Editor’s Note

What Progress Actually Looks Like

There is a particular quality to engineering progress that does not translate well into press cycles. It does not happen in announcements. It happens in test runs — 137 of them, in this week’s most important story — each one adding a data point to a distribution, each distribution narrowing the uncertainty envelope around a design that must work on another planet. NASA JPL (Jet Propulsion Laboratory) published the results of those 137 runs on May 7, two years after Ingenuity’s final flight and four years after its first, and the headline is both precise and extraordinary: the rotor blade tips of the next-generation Mars helicopter, a programme now named SkyFall, have exceeded the speed of sound without structural failure. That is the sentence that matters.

NVIDIA’s Space-1 Vera Rubin Module announcement from GTC in March has not appeared in these pages until now, and the delay has been useful: we can read it in the light of the thesis refinement we made last week, separating space-native orbital compute from the terrestrial-competing data centre vision. The Vera Rubin Module is unambiguously a space-native compute story — smaller, lighter, more power-efficient than anything that came before it, built for the orbital environment rather than adapted from it. Jensen Huang’s GTC framing — “intelligence must live wherever data is generated” — is the clearest public articulation of the space-native thesis this industry has produced.

And Europe, quietly, put thirteen satellites into orbit on a single Falcon 9 rideshare on the morning of May 3. Seven for Italy’s IRIDE constellation. Four for Greece’s national wildfire surveillance system — a world first. Two CubeSats testing optical data transmission in space, directly relevant to the photonics thesis that now anchors this publication. None of this required a press conference. It just happened, as accumulation does.

Section 01 · Human Spaceflight

SkyFall, Starship, and the Dragon at the Door

Ingenuity made its 72nd and final flight on January 18, 2024, having survived 1,000 Martian days in an environment it was never designed to endure. It carried no science instruments. Its entire purpose was to prove that powered, controlled flight on another world was physically possible. It did that, and it did it so thoroughly — so far beyond its planned five flights — that the question of what comes next shifted from “can it fly?” to “what can it carry?” The answer to that question is what JPL has been building toward in the low-density chamber at Pasadena, and what the May 7 results represent: a design capable of carrying instruments.

The programme now carrying that answer has a name: SkyFall. The rotor blades that will power it were developed and manufactured by AeroVironment in Simi Valley — the same company that built Ingenuity’s rotor system — and tested across 137 documented runs in a chamber that replicates the Martian atmosphere’s density of approximately 1% of Earth’s sea-level value. The blade tips were accelerated past Mach 1. They did not fail. The test results clear one of the most significant aerodynamic hurdles in the SkyFall development programme, and the engineering physics of exactly why this is hard is examined in this week’s Technical Deep-Dive in Section 07.

The explicit mission framing from JPL's video description is worth stating directly: SkyFall-class helicopters are designed to "carry small payloads — like instruments and sensors — to collect data in support of future human and robotic missions." This is the transition from technology demonstration to operational science platform. Ingenuity scouted terrain for Perseverance. SkyFall scouts terrain, takes measurements, deploys sensors, and returns data that mission planners cannot obtain from orbit or from a rover constrained to the surface. The scientific utility multiplier is substantial.

Elsewhere in this week’s section 01 edition: SpaceX CRS-34 launches today — the day this issue publishes — delivering over 6,400 pounds of science experiments and crew supplies to ISS on a Dragon spacecraft flying its sixth mission. Starship Flight 12 is imminent from Starbase, Texas: Ship 39 and Booster 19, the first version-3 integrated Starship system, targeting liftoff within this week’s window. Flight 12 is the first Starship to incorporate structural and propulsion changes from the lessons of Flights 7 through 11 — it is as much a new vehicle as a continuation of the test programme. And from China: the first uncrewed orbital flight of the Mengzhou spacecraft on the Long March 10A rocket remains on the 2026 schedule, carrying China’s crewed lunar programme a step closer to its first hardware-in-orbit validation.

Section 02 · Market Structure

Neutron Slips, and an Unexpected Industrial Grouping Emerges

Rocket Lab’s Neutron rocket — the medium-lift vehicle that represented the most credible commercial challenger to SpaceX’s Falcon 9 in its payload class — has slipped its debut to no earlier than the fourth quarter of 2026, following a first stage tank test failure earlier this year. The first Neutron flight will not be reusable; booster recovery is targeted for the second flight. This is a meaningful setback for a programme that had been tracking toward a mid-2026 inaugural. The tank test failure is not a catastrophic engineering problem — it is exactly the kind of anomaly that full-scale ground testing exists to find — but the schedule impact compounds Rocket Lab’s competitive position at a moment when Blue Origin’s New Glenn is itself grounded pending the NG-3 upper stage investigation. The medium-lift market between Falcon 9 and Falcon Heavy currently has no operational competitor. That gap persists for at least another two quarters.

Lockheed Martin announced this week that it is joining a collaboration with Firefly Aerospace and Seagate for offshore launch operations, building on Firefly’s Alpha rocket programme and Seagate’s data storage infrastructure interests. The grouping is unusual — a prime defence contractor, a commercial small launch provider, and a data storage company — and its strategic logic is not immediately obvious from the announcement. Offshore launch from maritime platforms removes regulatory constraints on launch azimuth, potentially enabling orbital inclinations not achievable from fixed land sites. Whether this is a genuine programme or an early-stage industrial exploration will become clearer when a launch site and customer manifest are named. Worth watching.

Section 03 · European Sovereign Space

Thirteen Satellites, One Morning, Three Countries

On the morning of Sunday May 3, a SpaceX Falcon 9 lifted off from Vandenberg carrying thirteen European satellites — seven for Italy, four for Greece, two for Greece again — and delivered all of them to orbit before most of Europe had finished breakfast. The compression of significance into a single rideshare launch is worth unpacking deliberately, because each of these three programmes represents something distinct about where European sovereign space capability is heading.

The seven IRIDE satellites are the latest addition to Italy’s national Earth observation constellation, now totalling 31 spacecraft on orbit. IRIDE is funded through Italy’s National Recovery and Resilience Plan — NextGenerationEU capital deployed into national space infrastructure — and coordinated by ESA with support from ASI. The HEO satellites carry multispectral high-resolution optical instruments for coastal monitoring, land use analysis, and emergency management. Italy has quietly built one of the most capable national EO constellations in Europe, and the May 3 launch advances it further.

The four Hellenic Fire System satellites are the more remarkable story in engineering terms. Greece has become the first country in the world to deploy a national satellite constellation specifically designed for wildfire detection and tracking. The Aegean and Mediterranean basin’s wildfire risk profile — exacerbated by every consecutive drought year — has driven a genuine national security imperative: detect ignitions from orbit before they become uncontrollable ground events. The four satellites provide near-continuous coverage of Greek territory at sufficient temporal resolution to serve as an early warning system rather than a post-event documentation tool. This is photonic and optical engineering applied to a climate adaptation problem, at national scale, funded by the same EU recovery mechanism that funded Neuraspace in Portugal. The policy architecture is beginning to produce operational space infrastructure.

The two Hellenic Space Dawn CubeSats are the most technically relevant to this publication’s thesis. They are testing satellite communication links and optical data transmission in orbit — laser and optical comms demonstrations on a Greek national programme, launched alongside the fire system satellites. ESA’s Laurent Jaffart, Director of Resilience, Navigation and Connectivity, noted explicitly that the mission “showcases innovative optical communications” as part of building Europe’s next-generation connectivity architecture. A national government deploying optical comms testbeds on CubeSats as part of a multi-satellite rideshare is no longer an experimental edge case. It is the emerging standard for how European space programmes are built.

Two other European developments this week deserve mention. ESA’s SMILE mission — the Solar wind Magnetosphere Ionosphere Link Explorer, a joint programme with the Chinese Academy of Sciences — is on the launch manifest schedule (May18/19) for a Vega-C from Kourou imminently, representing ESA’s most significant active collaboration with the Chinese space programme and a heliophysics mission that will study the dynamic interaction between solar wind and Earth’s magnetosphere in real time. And ESA launched a dedicated Extended Reality Competence Centre this week, releasing an XR Plugin built on Unreal Engine and OpenXR for space training and simulation applications — a development directly relevant to EWC Compute’s digital twins and simulation platform work.

Section 04 · Photonics for Space — Our Core Thesis

NVIDIA Declares the Frontier: The Space-1 Vera Rubin Module and What It Actually Means

This is where this publication dwells most deeply each week — the photonic and optical layer of space infrastructure, examined from inside the engineering domain. Everything else is context. This is the signal. And this week the signal comes from a GTC stage in March, covered here for the first time because it deserves more than a headline.

On March 16, Jensen Huang stood at the NVIDIA GTC conference and announced the Space-1 Vera Rubin Module — a space-qualified compute platform delivering up to 25 times the AI inferencing performance of the H100 GPU, engineered specifically for the size, weight, and power (SWaP) constraints of orbital platforms. The announcement named six partner companies already using NVIDIA accelerated computing for space missions: Aetherflux, Axiom Space, Kepler Communications, Planet Labs, Sophia Space, and Starcloud. Every one of those companies has appeared in this publication’s Section 04 analysis over the past four issues. That is not a coincidence. It is a map of the orbital compute ecosystem as it currently exists, centred on a single silicon supplier.

Huang’s framing is worth quoting precisely because it states the space-native compute thesis more clearly than anything previously published by a major technology company: “Space computing, the final frontier, has arrived. As we deploy satellite constellations and explore deeper into space, intelligence must live wherever data is generated. AI processing across space and ground systems enables real-time sensing, decision-making and autonomy, transforming orbital data centers into instruments of discovery and spacecraft into self-navigating systems.” The phrase “instruments of discovery” is not marketing language. It is the correct engineering description of what an inference-capable constellation does: it does not transmit raw data to be analysed on the ground. It generates knowledge at the point of observation.

The SWaP framing of the Vera Rubin Module is where the photonics dimension enters directly. Size, weight, and power constraints in orbital hardware are not independently optimisable — they are coupled through thermal management. A more powerful compute chip generates more heat. More heat requires more radiator area. More radiator area adds mass. More mass requires more launch Δv. The only architectural path that breaks this coupling is to reduce the heat generated per unit of compute — which means either more efficient silicon, or photonic interconnects that move data between chips at the speed of light with near-zero resistive heating. The Vera Rubin Module addresses the silicon efficiency dimension. The photonic interconnect generation is what follows it. This is the transition point that Precision with Light’s optical simulation capability is positioned to support: the design and validation of photonic chip interconnects for the thermal-budget-constrained orbital compute environment.

The partner ecosystem statements from the GTC announcement are worth reading as a sector map. Aetherflux CEO Baiju Bhatt: “NVIDIA Space-1 Vera Rubin Module delivers high-performance, energy-efficient AI at the edge in orbit, powered by solar energy — enabling autonomous operations and mission-critical services.” Kepler CEO Mina Mitry: “NVIDIA Jetson Orin brings advanced AI directly to our satellites, allowing us to intelligently manage and route data across our constellation.” Planet CEO Will Marshall: “By integrating NVIDIA’s accelerated platform from space to ground, we are supercharging our ability to index the physical world.” Three different companies, three different use cases — power beaming infrastructure, inter-satellite networking, Earth observation — all converging on the same silicon platform. The orbital compute ecosystem is forming around NVIDIA the same way the terrestrial AI infrastructure ecosystem did after the transformer architecture went mainstream. The platform lock-in dynamics are already visible.

One structural observation for this publication’s running thesis: the Vera Rubin Module announcement does not change the conclusion we reached last week about the SpaceX S-1 disclosure. The distinction between space-native orbital compute — serving the satellite’s own operations, the constellation’s own intelligence, the mission’s own needs — and terrestrial-competing orbital compute — selling processed outputs to Earth-based customers in competition with AWS — remains the correct analytical frame. Every partner in NVIDIA’s space ecosystem announcement is building the former. None of them is building a data centre that competes with Azure on price per FLOP. The thesis is intact and the Vera Rubin Module is its most significant hardware validation to date.

EWC intersections this week are the most direct of any issue to date. The Vera Rubin Module’s SWaP constraints and the photonic interconnect transition map directly onto the Precision with Light platform’s simulation domain. The Hellenic Space Dawn optical comms CubeSats are a European national programme demonstrating free-space optical transmission in orbit — the same class of link that the Cailabs MPLC ground station (Issue 004 Technical Deep-Dive) terminates on the ground side. The SkyFall rotor blade aerodynamics (Section 07) uses carbon fibre composite materials whose optical characterisation and structural monitoring is increasingly handled by photonic sensing — Brillouin scattering fibre sensors, fibre Bragg gratings — embedded in the blade structure. The photonics thread runs through every section of this issue.

Section 05 · One to Watch

Spaceflux: The Other Side of the Space Traffic Problem

EUROPE Active Signal

Last week we introduced Neuraspace — the Coimbra-based company using ML-driven fusion of optical and radar tracking data to predict conjunction events and recommend avoidance manoeuvres. This week, a complementary signal from the same sector: Spaceflux, a UK-based space domain awareness startup, extended its seed round with an additional £3.5 million, bringing total funding to £9 million. The company tracks objects in orbit using a combination of ground-based optical sensors and data analytics, focused on the real-time tracking layer that feeds into conjunction assessment systems like Neuraspace’s.

The distinction between Spaceflux and Neuraspace is architectural rather than competitive. Neuraspace’s strength is in the ML prediction and manoeuvre recommendation layer — what happens after tracking data is collected. Spaceflux’s strength is in the sensor network and data quality layer — the observational infrastructure that tracking data is collected from. Both are necessary components of a functioning space traffic management system, and the European sector is building them in parallel, from different countries, with different funding sources. The convergence of these capabilities into interoperable infrastructure is the outcome the sector needs; the current landscape of independent national programmes and startup-level sensor networks is the reality it is working from. £9 million is early-stage funding for a serious infrastructure problem. The number to watch is when Spaceflux’s sensor network reaches the coverage density that makes its data commercially differentiated from what the US Space Surveillance Network already provides for free.

Section 06 · Numbers of the Week

Section 07 · Technical Deep-Dive — Issue 005

Why SkyFall Must Break the Sound Barrier: The Aerodynamics of Flying on Mars

Aerospace engineering · Rarefied aerodynamics · Rotorcraft design

Ingenuity’s 72 flights on Mars established that powered flight in a 1% density atmosphere is possible. What they also established, precisely through the data they generated, is the envelope within which it is possible — and the hard physical boundary that any heavier aircraft must push against. SkyFall must carry science instruments. Science instruments have mass. Mass requires more lift. More lift in a near-vacuum atmosphere requires faster blade tips. Faster blade tips eventually reach the speed of sound. What happens at and beyond that boundary — and how AeroVironment’s new rotor design survives it — is the engineering story of this week’s test results.

The Lift Equation in a Near-Vacuum

Lift generated by a rotor blade is proportional to the density of the fluid it moves through, the square of the relative velocity between blade and air, the blade’s planform area, and its lift coefficient. On Earth, atmospheric density at sea level is approximately 1.225 kg/m³. On the Martian surface, the CO₂-dominated atmosphere has a density of approximately 0.015 to 0.020 kg/m³ — roughly 1.2 to 1.6% of Earth’s sea level value. Mars gravity is 3.72 m/s², approximately 38% of Earth’s, which reduces the weight of the vehicle that must be lifted. These two effects partially offset each other, but the density deficit dominates: a Mars helicopter must generate lift in a fluid approximately 60 to 80 times less dense than the one that terrestrial rotorcraft operate in.

The engineering response to low density is to increase velocity — specifically blade tip velocity, which is the dominant term in the relative velocity experienced by the outer section of the rotor blade where most lift is generated. Ingenuity’s coaxial counter-rotating blades spun at approximately 2,400 rpm, producing blade tip speeds of roughly 200 m/s — about 65% of the speed of sound in the Martian atmosphere. For the 1.8-kilogram Ingenuity, this was sufficient. For SkyFall’s heavier payload requirement, it is not. The blade tips must spin faster. And faster means approaching, then crossing, Mach 1.

The speed of sound in the Martian atmosphere is approximately 240 m/s at surface temperatures, compared to 343 m/s at standard Earth sea level conditions. The lower value reflects both the cold surface temperatures (averaging around -60°C) and the higher molecular weight of CO₂ relative to Earth’s nitrogen-oxygen mixture. A blade tip operating at Mach 0.9 on Mars is moving at roughly 216 m/s — already within the transonic regime where compressibility effects become significant. The JPL tests pushed tips beyond Mach 1, into the supersonic regime, where shock waves form on the blade surface and drag increases dramatically. Surviving this regime structurally and aerodynamically across 137 test cycles is what the AeroVironment blade design has now demonstrated.

The Transonic Problem: Compressibility, Shockwaves, and Mach Tuck

In incompressible aerodynamics — the regime where Ingenuity’s blade tips operated for most of its mission — air behaves as a fluid that simply accelerates around the blade profile. As blade tip speed approaches the speed of sound, this assumption breaks down. The airflow over the upper surface of the blade accelerates locally beyond the free-stream velocity; at certain blade geometries and angles of attack, the local flow on the suction surface can go supersonic even when the free-stream tip speed is still subsonic. This produces a normal shock wave on the upper blade surface — a discontinuity in pressure and density that generates a sudden drag increase, strong enough to be felt as a torque spike by the drive system.

Above the critical Mach number, the shock wave moves progressively toward the trailing edge as tip speed increases. The pressure distribution over the blade surface shifts rearward, moving the centre of pressure toward the trailing edge and creating a nose-down pitching moment — the phenomenon known as Mach tuck in fixed-wing aviation, and its rotary equivalent in rotor blade dynamics. For a rotor blade cycling through varying angles of attack on every revolution, this pitching moment introduces torsional oscillation that must be absorbed by the blade’s structural stiffness and damping. The AeroVironment three-blade design — visible in the test facility images with its characteristic swept tip geometry — incorporates the blade planform shape and composite layup required to manage these torsional loads at supersonic tip speeds, across the thermal cycling of a Martian day, across 137 test runs without structural failure.

Material Science: Carbon Fibre at Supersonic Tip Speeds

The rotor blades are carbon fibre composite structures — unidirectional and woven carbon fibre plies in an epoxy matrix, layered and oriented to provide the specific combination of bending stiffness, torsional stiffness, and aeroelastic tailoring that the design requires. At supersonic tip speeds the structural demands are severe. The centrifugal load at the blade root — proportional to the square of rotational speed — increases substantially as rpm rises toward and beyond the Mach 1 threshold. The aerodynamic loads become highly non-linear in the transonic and supersonic regime. And the Martian thermal environment adds a further constraint: surface temperatures cycle between approximately -73°C at night and -13°C in midday sun, and the composite material properties — stiffness, strength, thermal expansion — vary across this range in ways that must be accounted for in the structural design.

The 137-run test dataset is significant precisely because it addresses the statistical question that a single successful run cannot. A single pass through Mach 1 without failure is a data point. One hundred and thirty-seven passes without failure is a Weibull distribution — an empirical reliability argument that tells JPL something meaningful about the blade’s survival probability across a mission profile involving many flights. The test methodology is designed to bracket the uncertainty that the Mars environment introduces: unknown particle contamination, thermal shock on the first morning spin-up, dust accumulation on blade surfaces over an extended mission. Demonstrating structural integrity across 137 runs under controlled conditions narrows the uncertainty envelope for the operational mission without eliminating it. This is how engineering decisions about planetary hardware are actually made: not by achieving perfection, but by characterising the failure modes well enough to design around them.

Ingenuity to SkyFall: The Architecture Shift

Ingenuity used a coaxial counter-rotating two-blade rotor configuration: two sets of two blades, spinning in opposite directions on the same shaft, cancelling rotor torque without requiring a tail rotor. The design is elegant for a technology demonstrator because it minimises complexity and maximises the blade area that can be packed into a given rotor diameter. SkyFall’s design — visible in the test images as a three-blade horizontal rotor alongside a two-blade vertical test configuration — indicates a different architectural approach for the science-capable successor. The addition of a third blade to one rotor increases blade area without increasing rotor diameter, which keeps the blade tip speed achievable at a given target lift. The swept tip geometry visible on the test blades is the standard approach to managing the transonic drag rise: sweeping the tip rearward increases the effective chord-wise Mach number at which the tip operates, delaying the onset of the shock wave and managing its location on the blade surface at higher tip speeds.

AeroVironment — the manufacturer — is not a name that surfaces often in space technology coverage, but it is the precise engineering partner this programme requires. The company’s background in small autonomous aircraft, high-altitude long-endurance vehicles, and unmanned aerial systems gives it the composite rotor blade manufacturing expertise and aerodynamic analysis capability that a Mars rotorcraft development demands. The continuity from Ingenuity to SkyFall within the same industrial relationship is an underappreciated advantage: the institutional knowledge of what works in that specific low-density test chamber, at those specific rpm ranges, does not need to be rebuilt from scratch.

SkyFall does not yet have a confirmed launch date. It is a technology development programme currently in the rotor blade qualification phase. What the May 7 test results confirm is that the most physically demanding constraint — aerodynamic survival at supersonic tip speeds — has been addressed in hardware, across a statistically meaningful test campaign. The next constraints are mass, power, avionics, and landing gear for terrain that Ingenuity’s scouts photographed but never touched. Each of those is also a solvable engineering problem. The sound barrier was the hardest one, and JPL’s team broke it 137 times in a row.

References & Sources

Issue 005 — Primary References

Section 01 — Human Spaceflight / SkyFall

• NASA pushes next-gen Mars helicopter rotor blades past Mach 1 — NASA JPL, May 7 2026

• Testing the Next Generation of Mars Helicopter Rotor Blades — NASA JPL YouTube, May 7 2026

• NASA ISS 2026 flight plan — CRS-34 May 12 — NASA

Section 02 — Market Structure

• Rocket Lab Neutron debut slips to Q4 2026 — Spaceflight Now

• Lockheed Martin / Firefly / Seagate offshore launch collaboration — Spaceflight Now, May 4 2026

Section 03 — European Sovereign Space

• Launch boosts European Earth monitoring and connectivity — ESA, May 4 2026

• Extended Reality at ESA opens new pathways — ESA / Copernical, May 7 2026

• SMILE Vega-C launch — Spaceflight Now launch schedule

Section 04 — Photonics for Space / NVIDIA Vera Rubin

• NVIDIA launches Space Computing, Rocketing AI Into Orbit — NVIDIA Newsroom, March 16 2026

• NVIDIA Space Computing platform — NVIDIA.com

Section 05 — One to Watch / Spaceflux

• Spaceflux extends seed round to £9M — European Spaceflight, April 29 2026

Section 07 — Technical Deep-Dive: SkyFall Rotor Aerodynamics

• NASA JPL rotor blade test results — primary source

• NASA JPL video — test chamber footage, blade geometry, SkyFall context

EWC Space Delta-V is a weekly technology intelligence publication by Engineering World Company, published every Tuesday. It covers the photonic and optical layer of space infrastructure, and the engineering physics that governs it. Nuno Edgar NunesFernandes is a physics engineer with a background in optoelectronics and photonics, building AI-assisted engineering platforms across photonics, quantum computing, and industrial simulation.

Editor’s Note

Two Registers, One Week

This issue opens with a structural change. From Issue 004 onwards, Europe gets its own dedicated section — Section 03 — not because European space news is more or less important than US news, but because it has been systematically underweighted by the English-language press that dominates the information environment we all swim in. The change is an editorial correction, not a political statement. The industry is global. This publication should reflect that.

The two dominant registers of the week sit in productive tension. In Europe, the commercial launch sector is grinding forward with exactly the kind of unglamorous, iterative persistence that serious engineering requires. Isar Aerospace is still on the pad at Andøya after five scrub attempts. Rocket Factory Augsburg has delivered both stages of RFA ONE to Scotland and is preparing for a summer inaugural. PLD Space in Spain has €180 million and a slot at Kourou. None of this has yet reached orbit. All of it is building the sovereign launch infrastructure that Europe strategically requires, and all of it is happening without the institutional resources or public attention that comparable US programmes attract.

Meanwhile in Washington, SpaceX filed its S-1 IPO document and, buried in the risk factors section, offered the most honest public assessment of orbital compute viability produced by any actor in the sector. We will engage with that assessment directly in Section 04, because it deserves neither dismissal nor panic — it deserves the kind of careful engineering and economic analysis this publication exists to provide. The Technical Deep-Dive this week examines the ground segment problem that all orbital compute ultimately depends on: how do you move data from orbit to Earth at terabit scales when the atmosphere is actively trying to destroy your laser beam? A French company called Cailabs has a photonic answer that has been in commercial operation since 2024. We go deep on the physics.

Section 01 · Human Spaceflight

Artemis III Slips, and the Dependencies Compound

Artemis III has slipped to late 2027. The immediate cause is dual: SpaceX’s Starship, which serves as the Human Landing System, has not yet demonstrated the in-orbit propellant transfer and long-duration cryogenic fluid management required for a crewed lunar landing mission. Blue Origin’s Blue Moon lander, which was added as an alternative HLS option, faces its own development timeline challenges following the NG-3 upper stage anomaly that grounded New Glenn indefinitely pending FAA investigation. Neither vehicle is ready for the mission profile Artemis III requires on the previously targeted 2026–2027 timeline. NASA has formally acknowledged the slip.

The cascade effect on the broader Artemis architecture is significant. The nuclear reactor on the lunar surface targeted for 2030 by NSTM-3 — which we covered in Issue 002 — was predicated on an Artemis III that placed crews on the South Pole by late 2026 or early 2027, providing the operational context for deploying Lunar Reactor-1. A 2027 Artemis III, dependent on Starship achieving propellant transfer at scale, pushes the 2030 nuclear reactor timeline into territory that requires simultaneous success across three distinct development programmes. Schedule dependencies in space programmes compound exponentially when they nest.

There is a quieter development on the European side of the human spaceflight story this week. ESA’s Celeste LEO-PNT demonstrator — the first two satellites of a planned 10-satellite Galileo augmentation constellation — launched on March 28 on a Rocket Lab Electron from New Zealand and delivered first navigation signals on April 8. The irony is structural: Europe’s sovereign navigation augmentation infrastructure launched on a non-European rocket. Electron is a New Zealand-origin vehicle operating from Māhia Peninsula. Isar, RFA, and PLD Space exist precisely to close this dependency over the next decade. Until they reach orbital cadence, European institutional payloads will continue flying on foreign vehicles.

Section 02 · Market Structure

Three Rockets in Five Days

The launch manifest this week produced an unusual density of distinct vehicle milestones. On April 27, ULA’s Atlas V launched 29 Amazon Kuiper LEO satellites — its heaviest-ever payload — in what was also one of the final Atlas V missions before the vehicle retires in favour of Vulcan Centaur. On April 29, SpaceX’s Falcon Heavy flew for the first time in 18 months, carrying the ViaSat-3 F3 communications satellite to geosynchronous transfer orbit — the 12th Falcon Heavy mission overall and the first to simultaneously use both Landing Zone 2 and Landing Zone 40 for booster recovery. And on April 30, Russia’s Soyuz-5 medium-lift rocket made its inaugural flight from Baikonur.

The Soyuz-5 deserves specific attention. It is the first genuinely new Russian orbital launch vehicle in years, designed to replace the aging Soyuz-2 in the medium-lift segment and eventually to serve as the first stage of the heavier Soyuz-6. Russia’s launch sector has been under sustained pressure since 2022 — international commercial customers have departed, Roscosmos has lost access to Western components, and the Baikonur facility suffered damage in the November 2025 Soyuz MS-28 launch incident. A successful Soyuz-5 inaugural is not merely a technical milestone; it is a signal that Russia’s launch programme retains domestic development capability despite those pressures. The geopolitical significance is not separable from the engineering one.

SpaceX passed its 50th launch of 2026 on April 26 and is tracking toward approximately 160 orbital missions this year — just below its 2025 record of 165. All 50 launches to this point have been Falcon 9. No Starship has flown in 2026; the programme is preparing for its 12th suborbital test at Boca Chica. The gap between SpaceX’s operational Falcon 9 cadence and every other launch provider remains the defining structural fact of the global launch market. The three distinct vehicle milestones this week are each significant in their own domain. None of them moves that gap.

Section 03 · European Sovereign Space — Inaugural Section

The Pad, the Stages, and the Policy: Europe’s Launch Week

This section makes its first appearance in Issue 004 as a standing feature. It will cover ESA programme developments, European commercial space, EU Space Policy, and the national agency contributions of Europe’s member states — with particular attention to those stories that the US-centric space press systematically misses or underweights.

The most visible European story this week is the one that has not yet happened: Isar Aerospace’s Spectrum rocket remains on the pad at Andøya Space in Norway, having accumulated five scrub attempts across three months. The most recent was a leak in a composite overwrapped pressure vessel — a COPV — discovered during final pre-launch checks. Isar’s CEO Daniel Metzler has been direct: “There is no question that we will reach orbit and demonstrate reliable access to space. Scrubs are part of the rocket industry; every successful rocket company has been here.” That statement is accurate. SpaceX’s early Falcon 9 programme accumulated extensive scrub history. What matters is the engineering response to each anomaly, not the anomaly count itself. What Isar has in its favour is that rockets for flights three through seven are already in production, and a new 40,000 square metre facility near Munich is opening in 2026. The programme is not scrubbing out of weakness. It is scrubbing out of caution, which is the correct disposition for a second test flight on a vehicle that failed 30 seconds into its first.

Rocket Factory Augsburg has delivered both stages of RFA ONE to SaxaVord Spaceport on the island of Unst in Shetland, Scotland — the northernmost point of the British Isles, chosen for its polar orbit access. The new 52-metre umbilical tower has been raised and the launch pad is in commissioning. RFA ONE is a 30-metre two-stage vehicle designed for 1,300 kg to LEO, with an optional Redshift kick stage for orbital flexibility. After an August 2024 pad fire destroyed the original first stage during a hot fire test, the company spent 18 months rebuilding and upgrading — adding improvements to the Helix engine, tank pressurisation systems, and ground procedures. Summer 2026 is the current target for the inaugural flight. Given the pace of pad commissioning activity, that timeline is credible.

In Spain, PLD Space has secured €180 million in fresh funding and confirmed a slot at the new multi-user commercial launch facility at the Guiana Space Centre for the inaugural orbital flight of its Miura 5 rocket by year-end. Miura 5 is a 19-metre vehicle targeting 450 kg to SSO — smaller than Spectrum and RFA ONE, occupying the microsatellite rideshare market rather than the small dedicated launch segment. PLD Space previously flew its Miura 1 suborbital vehicle in October 2023, giving it more flight heritage than either Isar or RFA at this stage.

The ESA (European Space Agency) and European Defence Agency signed an Implementing Arrangement on April 22 to jointly map strategic and technological gaps in Europe’s Earth observation capabilities for security and defence. This is the formal institutionalisation of a trend that has been visible for several years: the Copernicus programme and the Sentinel constellation, built for civilian Earth observation, are now explicitly being integrated into European defence planning. The dual-use nature of EO infrastructure — civilian in design, strategically critical in application — is no longer a subtext. It is the stated policy.

The US counterpart to this European policy convergence is the Golden Dome missile-defence architecture — a planned network of space-based sensors and interceptors requiring continuous tracking data shared in near real time. Its enabling technology is optical inter-satellite communications: laser crosslinks between LEO and MEO satellites, and from orbit to ground. The Space Force’s OPIR Space Modernization Initiative, with a $180 million FY2027 budget, is funding the MEO laser crosslink demonstrations that Golden Dome depends on. Europe’s ESA/EDA Earth observation roadmap and the US Golden Dome architecture are being built simultaneously, from different institutional starting points, toward the same structural conclusion: space-based optical sensing and communications are defence-critical infrastructure. The laser communications physics is the same on both sides of the Atlantic. The policy frameworks diverge. The engineering requirements converge.

A note on the broader European launch context. Orbex, the UK-based small launch provider that had been preparing for an inaugural flight from its Sutherland, Scotland facility, ceased operations in February 2026 after a failed acquisition attempt led to insolvency. The loss matters not just for Orbex’s own programme but for Sutherland Space Hub, which had been positioned as the UK’s first orbital launch site. The commercial small launch market is unforgiving: capital intensity is high, margins are thin before cadence is established, and the competitive pressure from SpaceX rideshare keeps prices under permanent downward pressure. RFA and Isar are both better capitalised than Orbex was, but the structural economics of the market are the same for all three.

Portugal in orbit — the Atlantic Constellation takes shape. Two events in the past five weeks mark a genuine inflection point for the Portuguese space programme, and both deserve coverage in a publication edited from Portugal. On March 30, six Portuguese satellites launched simultaneously on a SpaceX Falcon 9 Transporter-16 rideshare from Vandenberg — the largest single Portuguese launch in history, broadcast live from the Pavilhão do Conhecimento in Lisbon. Among the six were the Portuguese Air Force’s first SAR satellite and an optical satellite developed by CEiiA, the Centre for Engineering and Product Development based in Évora, now joining three Atlantic Constellation satellites already in orbit. Then on May 3 — this week — a second Air Force SAR (Synthetic Aperture Radar) satellite launched from Nebraska on another Falcon 9, further expanding Portugal’s all-weather, day-and-night Earth observation capability at 600 kilometres altitude.

The Atlantic Constellation is a Spain-Portugal joint programme targeting 26 satellites in total — 12 SAR and 14 optical — providing intraday revisit of the Iberian Peninsula, the Atlantic, and the Portuguese Exclusive Economic Zone. The SAR satellites are supplied by ICEYE, the Finnish company that has become the dominant provider of small SAR spacecraft to European national programmes. The optical satellite is a CEiiA product — a Portuguese-engineered spacecraft, not a procured foreign asset. Under the ICEYE partnership signed in December 2025, CTI Aeroespacial — the joint venture between the Air Force, Geosat Satélites, and CEiiA — will establish a satellite assembly facility at the Air Force General Material Depot in Alverca, near Lisbon. Portugal is not merely buying access to space. It is building the industrial base to operate in it.

The Portuguese Air Force’s framing of this programme — under the “Air Force 5.3” Flight Plan for 2024–2030, which identifies Space as the fifth operational domain alongside Land, Sea, Air, and Cyber — is not rhetorical. The SAR capability specifically strengthens maritime surveillance of Portugal’s EEZ (Exclusive Economic Zone), which at 1.7 million square kilometres is the largest in the European Union and among the largest in the world. Persistent, high-resolution, all-weather SAR imagery of that area changes the operational picture for the Air Force, Border and Coastguard, and NATO’s Atlantic flank in ways that no ground-based or aerial system can replicate at comparable cost. The fact that two of these satellites launched within five weeks of each other, on a programme funded through Portugal’s Recovery and Resilience Plan, is the kind of development that deserves more attention than it has received outside Portuguese media.

Section 04 · Photonics for Space — Our Core Thesis

This is where this publication dwells most deeply each week — the photonic and optical layer of space infrastructure, examined from inside the engineering domain. Laser communications, free-space optical links, photonic components for spacecraft, quantum channels in orbit — this is the section written at the level of an optoelectronics engineer, not a technology journalist. Everything else in the issue is context. This is the signal.

SpaceX submitted its pre-IPO S-1 registration document to the SEC this week. Buried in the risk factors section — as it is legally required to be, in plain language, for the benefit of prospective investors — is the following disclosure: “Our initiatives to develop orbital AI compute and in-orbit, lunar, and interplanetary industrialization are in early stages, involve technical complexity and unproven technologies, and may not achieve commercial viability.” SpaceX also disclosed plans for up to one million orbital data centres, intended initially for internal xAI and Tesla compute workloads rather than external customers.

This requires careful reading. An S-1 risk disclosure is not an editorial opinion, a research note, or a strategic assessment. It is a legal obligation: any material risk to the business must be stated plainly so that investors can make informed decisions. SpaceX is legally obligated to disclose uncertainty about orbital compute viability if that uncertainty exists and is material. The fact that it does so does not mean the company has abandoned the programme — it means the programme is real enough to be a material business line, and honest enough in its current state to warrant a risk disclosure. Both things are true simultaneously.

At a SpaceNews event on orbital data centres in Washington on April 30, Founders Fund partner Delian Asparouhov — whose firm is an early SpaceX investor and has backed both OpenAI and Anthropic — said he would be wary of competing directly with SpaceX in orbital compute. He pointed instead to the opportunity layer: companies built around services enabled by orbital data centres, rather than the data centres themselves. He also named the unresolved challenges with unusual frankness: power generation, thermal management, regulation, and whether AI demand persists at the scale required to justify space-based infrastructure if the terrestrial AI investment cycle contracts. These are not new concerns. They are the same constraints this publication has been naming since Issue 001. The difference is that they are now being named, in public, by a Founders Fund partner at a space industry event. The sector is beginning to do its own honest accounting.

None of this invalidates the thesis. What Starcloud demonstrated — hardware survives the environment — remains a genuine milestone. What Planet Labs demonstrated last week — production AI inference on an operational constellation — remains a category shift. What Kepler Communications has built — 40 NVIDIA Orin GPUs linked by laser inter-satellite communications across 10 satellites, running at 100% utilisation on inference workloads — remains the most commercially realistic near-term architecture in the sector. The SpaceX disclosure and the Asparouhov comments are a useful corrective to the hype cycle, not a refutation of the underlying engineering progress.

There is, however, a distinction the S-1 disclosure implicitly makes that deserves to be stated explicitly as a thesis refinement for this publication. SpaceX hedges on orbital compute competing with terrestrial data centres — the model where workloads run in orbit and outputs are sold to Earth-based customers via downlink. That model carries the ground-segment economics problem, the thermal constraints, and the launch cost burden that the S-1 correctly identifies. But orbital compute serving space-native operations is a structurally different proposition, and its logic is largely immune to those objections. Compute that serves a constellation’s own inference needs — Planet Labs’ aircraft detection, Kepler’s hosted payloads — never competes with AWS. Compute supporting lunar or deep space operations cannot be replaced by Earth-based infrastructure: a one-way signal delay to Mars of 3 to 22 minutes makes any round-trip to an Earth data centre physically unusable for real-time autonomous operations. And compute embedded in the military sensor-to-shooter network — as the K2 / Golden Dome programme makes explicit — cannot afford the latency of a ground trip. The orbital data centre serving Earth customers remains a secondary thesis under scrutiny. The orbital data centre serving space operations is the primary thesis, and it rests on physics that does not negotiate. This distinction will sharpen as the sector matures, and this publication will track both layers honestly.

The demand signal for laser communications in space is not coming only from commercial orbital compute. The US military is building it into the foundational architecture of its next-generation missile defence system. Space Force has selected K2 Space — a California startup whose first satellite, Gravitas, launched March 30 carrying a 20-kilowatt power system — for the Pentagon’s OPIR Space Modernization Initiative. The SMI carries a $180 million FY2027 budget, with $7.3 million specifically earmarked for optical inter-satellite crosslink demonstrations in medium Earth orbit. The work supports Golden Dome, the planned missile-defence architecture requiring a large constellation of space-based sensors and interceptors to track threats and share targeting data in near real time. K2’s head of strategy John Plumb, formerly in charge of space policy at the Defense Department, is direct about what makes MEO laser crosslinks technically distinct from the LEO links SpaceX already operates: “The distances are longer, the radiation environment is different. So it’s a different, harder problem, and we’re going to be the first ones there, really testing it out.” No one has yet solved space-to-space data links in MEO. K2 is being paid to try.

The GAO’s February 2025 report provides the necessary honest counterweight to the Golden Dome ambition. The Space Development Agency — the office responsible for building the Proliferated Warfighter Space Architecture that underpins Golden Dome — has taken steps to develop laser communications technology but has not yet fully demonstrated it in space. As of December 2024, only one of four prime contractors in the demonstration tranche had completed three of eight planned laser communications capabilities. The remaining two contractors had achieved none. The programme has lowered its expectations and still faces challenges. Nearly $35 billion in follow-on investment is contingent on demonstrating a minimum viable product that has not yet been demonstrated. The military demand signal for laser communications is real and enormous. The technology readiness is not yet at the level the programme’s financial commitments assume.

The ground segment constraint is, however, where the orbital compute sector faces its most underappreciated physical bottleneck. An orbital data centre that cannot downlink its processed outputs at sufficient bandwidth and reliability is not a data centre — it is an isolated compute node. The satellite-to-ground laser communication link, operating through an atmosphere that introduces turbulence, scattering, and absorption at every wavelength of practical interest, is the critical path. Which brings us directly to a French company from Rennes, and to this week’s Technical Deep-Dive.

Before we go there: SES, the Luxembourg-based satellite operator, has contracted Cailabs’ TILBA-OGS optical ground stations for testing space-to-ground laser communication at 10+ Gbps. This is commercial validation of the technology from one of the world’s largest satellite fleet operators. The ground station is not a research prototype. It is in production, under contract, with more than ten stations currently committed. The physics underlying it is examined in Section 07.

EWC intersections: the SpaceX S-1 disclosure of orbital compute risk factors maps directly onto the physics constraints we have been naming — thermal management, radiation, launch qualification, power generation — all of which are engineering problems, not business problems. The Cailabs MPLC technology operates in the same free-space optical domain as the Precision with Light platform’s laser communication work. The ground station link budget analysis in Section 07 is directly relevant to EWC’s optical simulation capabilities.

Section 05 · One to Watch

Neuraspace: The Coimbra Company Keeping Orbital Traffic from Colliding - Portugal Active signal

This is the first Portuguese SpaceTech company to appear in EWC Space Delta-V, and it earns its place on technical merit rather than geography. Neuraspace, founded in 2020 and headquartered in Coimbra, provides automated space traffic management as a SaaS platform — real-time conjunction assessment, manoeuvre recommendation, and space situational awareness data. Its customers include ESA. It has raised €2.5 million from Armilar Venture Partners and a further €25 million through NextGenerationEU and Portugal’s Recovery and Resilience Plan funds for sensor infrastructure and growth.

The EISCAT partnership announced in August 2024 is technically the most interesting development in Neuraspace’s recent history. EISCAT — the European Incoherent Scatter Scientific Association, operating high-latitude radar facilities in the Arctic since the 1980s — provides incoherent scatter radar data that complements Neuraspace’s optical tracking capabilities. The technical distinction matters: optical sensors track objects by reflected sunlight in the visible spectrum, providing good angular resolution but limited range information in sunlit conditions. Incoherent scatter radar illuminates objects with radio frequency energy and analyses the backscattered return, providing independent positional measurements with different error characteristics and operating independently of solar illumination. Fusing optical and radar data produces conjunction predictions with tighter uncertainty ellipsoids than either source alone — which directly translates into fewer unnecessary avoidance manoeuvres and fewer missed conjunction events.

Neuraspace currently predicts the evolution of object uncertainty up to five days ahead of the time of closest approach. The EISCAT 3D next-generation system — currently in final development, with first test measurements expected shortly — will dramatically increase the radar aperture and sensitivity available to the partnership. EISCAT 3D is a phased array radar with distributed transmitters and receivers across northern Norway, Sweden, and Finland, covering the polar orbital inclinations where conjunction event frequency is highest for LEO constellations. Access to EISCAT 3D data will meaningfully advance Neuraspace’s machine-learning taxonomy models for resident space object characterisation.

In May 2025, Neuraspace expanded its service offering to five product areas: Space Traffic Management, Data and Tracking Services, Launch Screening, Mission Design, and Launch and Early Operations (LEOP) Tracking. The expansion reflects a deliberate strategic pivot: in CEO Chiara Manfletti’s framing, spacecraft operations must become more autonomous, less reliant on human intervention, and more responsive to the security dimension of space infrastructure as geopolitical tensions elevate the threat to orbital assets. Neuraspace is also a participant in EMISSARY, a €160 million European defence initiative for space situational awareness — the single largest investment in this domain in European defence history.

The structural relevance to one meaningful thesis of this publication - orbital scientific computing - is direct. Every orbital compute constellation we have covered — Starcloud, Kepler, EDGX, Lonestar — adds mass to an already congested LEO environment. The conjunction event rate scales with the number of objects. The software layer keeping those objects from colliding is not optional infrastructure. It is the operational prerequisite for the entire orbital compute sector. A company in Coimbra, Portugal, funded partly by NextGenerationEU, building that layer for European operators — and doing it with technically serious ML-driven fusion of optical and radar data — is worth knowing about.

Section 06 · Numbers of the Week

Section 07 · Technical Deep-Dive — Issue 004

Taming the Atmosphere: How Cailabs’ MPLC Technology Makes Satellite Laser Links Reliable

Free-space optical communications · Photonics · Wavefront engineering

The promise of orbital compute depends entirely on a link that most coverage ignores: the path from satellite to ground. A GPU in orbit is commercially useless if the data it processes cannot be retrieved at sufficient bandwidth and reliability. Radio frequency downlinks — the incumbent technology — are congested, spectrum-licensed, and physically bandwidth-limited by the carrier frequency. The maximum data rate achievable on an RF link scales with available bandwidth, which in turn scales with carrier frequency. Optical frequencies are five orders of magnitude higher than Ka-band RF, which means optical links can in principle carry five orders of magnitude more information through the same aperture. This is why free-space optical communication from satellite to ground is not an incremental improvement over RF. It is an architectural shift.

The obstacle is the atmosphere. The 10 to 20 kilometres of turbulent air between a low Earth orbit satellite and a ground receiver introduces wavefront distortions — variations in the refractive index of air caused by temperature fluctuations, wind shear, and humidity gradients — that spread and scramble a laser beam’s phase front. A perfectly coherent beam launched from a satellite arrives at the ground as a spatially incoherent speckle pattern, fluctuating on millisecond timescales. Coupling this distorted beam efficiently into a single-mode optical fibre — the standard interface for downstream optical processing — is the fundamental technical problem of satellite optical ground station design.

The Standard Approach: Adaptive Optics and Its Limits

The established solution is adaptive optics: a wavefront sensor measures the incoming distortion, a deformable mirror with hundreds of actuated segments applies the conjugate correction in real time, and the corrected beam couples into the fibre. Adaptive optics systems can achieve excellent performance under good seeing conditions. Their limitations are well known. They require fast, accurate wavefront sensing — typically using a Shack-Hartmann sensor or similar — a deformable mirror with sufficient stroke and actuator density, and a control loop running at kilohertz rates to track the atmospheric coherence time. The system is mechanically complex, thermally sensitive, requires calibration, and can saturate in conditions of severe turbulence. For an outdoor installation that must operate across years without specialist maintenance, adaptive optics represents a significant operational burden.

Multi-Plane Light Conversion: The Cailabs Approach

Cailabs’ approach is fundamentally different in architecture, and the difference is rooted in a principle from spatial mode theory rather than from classical adaptive optics. The core technology is Multi-Plane Light Conversion — MPLC — a passive optical system that performs unitary transformations on the spatial mode content of an optical beam through a sequence of phase masks and free-space propagation steps between them.

The principle is as follows. An atmospheric turbulence channel can be decomposed, in the mathematical framework of coherent mode decomposition, into a finite number of spatial modes — orthogonal field distributions that propagate independently through the channel. The dominant modes carry most of the power. A sufficiently sophisticated MPLC device can sort the incoming beam into these modes, phase-match them, and coherently recombine them into a single output mode suitable for fibre coupling — all without any electronic feedback, deformable optics, or moving parts. The physics is exact: MPLC implements a unitary transformation on the spatial mode basis, and unitary transformations are by definition lossless and reversible.

The practical implementation of MPLC involves a sequence of phase masks — typically implemented as computer-generated holograms on phase-only spatial light modulators, or as fixed etched optical elements once the design is optimised — separated by free-space propagation distances chosen so that each mask intercepts the beam at a specific plane determined by the desired mode transformation. The number of planes determines the complexity of the transformation achievable. Cailabs’ TILBA-ATMO device operates with sufficient planes to address the dominant atmospheric turbulence modes over a receive aperture relevant to a ground station telescope. Critically, once the masks are fabricated and the propagation geometry is fixed, the device operates entirely passively — no control electronics, no wavefront sensors, no feedback loops. It is a piece of glass performing a computation in photons.

The Three TILBA Building Blocks

Cailabs has structured the TILBA product family around three distinct functional building blocks, each addressing a specific aspect of the bidirectional optical ground station link:

TILBA-ATMO (receive-side turbulence mitigation). On the receive side, TILBA-ATMO addresses the coupling problem described above: an incoming beam distorted by atmospheric turbulence is decomposed into its dominant spatial modes, which are phase-corrected and coherently recombined into a single-mode fibre output. The device operates passively and is designed to handle the statistical distribution of turbulence conditions at a given site. Performance is characterised in terms of coupling efficiency improvement relative to direct fibre injection — the gain factor determines the link margin that TILBA-ATMO contributes to the system link budget.

TILBA-IBC (transmit-side incoherent beam combining). On the transmit side, the problem is different. A single high-power laser source propagating through turbulence will experience beam wander, scintillation, and spreading that reduce the intensity at the satellite receiver. TILBA-IBC uses incoherent beam combining: multiple independent laser sources are launched through different sub-apertures of the transmit telescope, producing multiple independently propagating beams that experience uncorrelated turbulence realisations. At the satellite receiver, the spatial diversity means that at least some fraction of the transmitted power arrives in a useful state at any given moment. The technique requires no phase coherence between transmitters — each source operates independently — which is architecturally simpler than coherent beam combining while still providing meaningful diversity gain against turbulence.

TILBA-CBC (coherent beam combining for future scalability). For the terabit-per-second feederlink capacities that large orbital compute constellations will eventually require, TILBA-IBC’s incoherent approach is insufficient — the transmitted power must scale to levels that single sources cannot achieve while maintaining beam quality. TILBA-CBC coherently combines multiple laser sources — phase-locking them so that they interfere constructively in the far field — to produce an effective transmit power approaching 100 watts from a compact aperture. Coherent beam combining at this level requires active phase stabilisation between sources, making TILBA-CBC the most technically complex of the three building blocks, and the one targeted at the longer-term market rather than the current commercial deployment.

Why This Matters for Orbital Compute

The orbital compute sector currently operates under an implicit assumption that the ground segment problem will be solved by the time large-scale orbital data centres exist. That assumption is not unreasonable — TILBA-OGS is already operational, and the technology trajectory is credible. What Cailabs' deployment with SES validates is not merely that the technology works in the laboratory. It validates that it works commercially, at scale, with a major institutional operator, on the timeline that orbital compute investors are using for their financial models.

The passive architecture of TILBA-ATMO is particularly significant for the multi-station network economics. An adaptive optics ground station requires specialist maintenance and calibration. A passive MPLC device, once installed and aligned, operates without ongoing expert intervention. For a network of dozens of ground stations distributed globally — the architecture required to provide adequate contact time for a large LEO compute constellation — the operational cost differential between adaptive optics and passive MPLC is substantial over a decade-long deployment lifetime.

From an EWC/Precision with Light perspective, the MPLC principle is a photonic computation in the spatial domain — the same class of operation that photonic integrated circuits perform in the spectral and temporal domains. The mode decomposition and recombination that TILBA-ATMO executes in free space is analogous to the demultiplexing and remultiplexing operations in a wavelength-division multiplexed optical communications system. The underlying mathematics — unitary transformations on a modal basis — is the same in both cases. This is not a coincidence. It reflects a deep structural similarity between spatial mode engineering and spectral/temporal mode engineering that is increasingly being exploited in both directions: free-space techniques informing integrated photonics, and integrated photonics approaches influencing free-space system design. Cailabs itself operates across both domains, with its AROONA and ONEMODE products addressing multimode-to-single-mode conversion in legacy fibre networks using the same MPLC principle. And one further connection worth naming: the OPIR infrared sensing satellites at the heart of Golden Dome — detecting missile launches by tracking heat signatures from space — are electrooptical and infrared systems operating across the same spectral domains that the Precision with Light platform models. The defence demand signal for MEO optical communications, the commercial demand signal for orbital compute downlinks, and the photonic engineering required to satisfy both — these are not three separate stories. They are one story, told from three different institutional vantage points.

References & Sources

Issue 004 — Primary References

Section 01 — Human Spaceflight

• Artemis 3 slips to late 2027 as Starship and Blue Moon lag — Space.com

• ESA Celeste LEO-PNT launch on Rocket Lab Electron — ESA Newsroom

Section 02 — Market Structure

• Falcon Heavy ViaSat-3 F3 / Atlas V Amazon Leo 6 / Soyuz-5 inaugural — Spaceflight Now launch schedule

• SpaceX 50th launch of 2026 — Space.com

• Soyuz-5 first flight — NASASpaceFlight.com

Section 03 — European Sovereign Space & Portugal

• Second Portuguese military satellite launches — Portugal Resident, May 3 2026

• Six Portuguese satellites launch on Transporter-16 — The Portugal News, March 30 2026

• ICEYE launches six SAR satellites for Poland and Portugal — The Defense Post

• ICEYE / CTI Aeroespacial agreement and Atlantic Constellation context — ICEYE newsroom

• Isar Aerospace Spectrum “Onward and Upward” scrub history — Isar Aerospace mission updates

• RFA ONE stages delivered to SaxaVord — Interesting Engineering

• PLD Space Miura 5 and European launch landscape 2026 — European Spaceflight

• ESA / EDA Implementing Arrangement, April 22 2026 — ESA Newsroom

• RFA ONE pad commissioning at SaxaVord — European Spaceflight

Section 04 — In-Orbit Compute

• SpaceX S-1 warns orbital AI data centres may not be viable — Dataconomy

• The opportunity beyond orbital data centres (Founders Fund / SpaceNews event) — SpaceNews

• Space Force taps K2 satellites to test laser communications for missile defence — SpaceNews

• Boeing opens EO/IR sensor production line for Space Force missile warning — Air & Space Forces Magazine

• Space Force taps K2 satellites for MEO laser crosslinks — Golden Dome / SMI — SpaceNews

• GAO questions SDA laser communications readiness — Breaking Defense

Section 05 — Neuraspace

• Neuraspace / EISCAT partnership — Neuraspace blog, August 2024

• Neuraspace launches five new service areas — Neuraspace blog, May 2025

• Neuraspace corporate — neuraspace.com

Section 07 — Technical Deep-Dive: Cailabs MPLC

• TILBA-OGS product page — Cailabs

• TILBA technology building blocks (ATMO, IBC, CBC) — Cailabs

• 5 reasons why Cailabs turns lasercom into reality — Cailabs blog

• Cailabs newsroom — including SES TILBA-OGS contract announcement

EWC Space Delta-V is a weekly technology intelligence publication by Engineering World Company, published every Tuesday. It covers the intersection of space infrastructure, AI compute, photonics, and the physics that governs all of them. Nuno Edgar Nunes Fernandes is a physics engineer with a background in optoelectronics and photonics, building AI-assisted engineering platforms across photonics, quantum computing, and industrial simulation.

Next issue: Tuesday May 12, 2026. Links, papers, and signals — especially from European and Portuguese space — are always welcome.

Editor’s Note

A Week of Sharp Edges

Some weeks in this industry move in a single direction. This was not one of them. Issue 3 covers a week of contradictions — engineering milestones arriving alongside engineering failures, institutional ambition running into budget reality, and a genuinely significant threshold being crossed in orbit that most of the space press missed entirely.

Start with the threshold. Planet Labs ran production AI inference directly on its Pelican-4 satellite this week — not a demonstration, not a testbed, but an operational pipeline on a commercial constellation, detecting aircraft over Alice Springs and returning geo-rectified structured data rather than raw imagery. Latency from collection to actionable intelligence dropped from hours to minutes. This is the moment the orbital compute thesis moves from “compelling argument” to “deployed product.” We examine it in detail in Section 04, alongside EDGX’s STERNA platform and Lonestar’s orbital sovereign data storage announcement — three distinct points on the maturity curve of the same underlying infrastructure shift.

Then there is Blue Origin’s NG-3. The first reused New Glenn booster landed cleanly. The upper stage failed, left the payload in the wrong orbit, and the FAA grounded the vehicle. Both facts matter, and both need to be held simultaneously rather than allowing one to eclipse the other. Section 02 does that analysis without sentiment in either direction.

GPS III-8 — named Hedy Lamarr, for reasons that turn out to be historically precise rather than merely ceremonial — closed the GPS III production block this week while carrying the first laser communications demonstration on a GPS satellite. That intersection of navigation infrastructure and free-space optical engineering sits directly in this publication’s domain, and Section 03 examines it accordingly.

And this issue introduces something new: the first Technical Deep-Dive, in Section 07, examining the Nancy Grace Roman Space Telescope as a piece of optical engineering rather than as a news story. The 2.4-metre NRO heritage mirror, the 300-megapixel HgCdTe focal plane, the Coronagraph Instrument’s 10⁻⁹ contrast target and dual deformable mirror wavefront control system — these deserve treatment at the level of engineering specifics, not press release summary. That is what Section 07 attempts. It will appear in issues where the technical material justifies the depth. This week, it clearly does.

Section 01 · Human Spaceflight & Policy

The 46% Cut and the Telescope That Defied It

The Trump administration's FY2027 budget proposal, released this week, requests a 23% overall cut to NASA's budget and a 46% reduction to its science programmes — cancelling 53 science missions in a single document. The proposal is a continuation and deepening of a pattern visible in last year's budget, which Congress rejected. On April 22, NASA Administrator Jared Isaacman appeared before Congress for a hearing that revealed unusually broad bipartisan opposition. Republican Representative Brian Babin, chair of the House Science Committee's space subcommittee, stated plainly that he does not support the proposal and is confident Congress will reject it again.

That said, the annual ritual of an extreme executive budget proposal being moderated by Congress should not produce complacency. The proposals do damage even when rejected: they create planning uncertainty that causes key personnel to leave, that delays procurement decisions, and that signals to the international partner community that US science commitments are conditionally reliable. ESA, JAXA, and the Canadian Space Agency all have programmatic investments tied to missions now under proposed termination. The diplomatic cost of that uncertainty is not recoverable by a subsequent budget line.

Against this backdrop, NASA unveiled the fully assembled Nancy Grace Roman Space Telescope at Goddard on April 21 — completing integration eight months ahead of its original schedule and under budget. Isaacman set a September 2026 launch window on a SpaceX Falcon Heavy to the Sun-Earth L2 point, 1.5 million kilometres from Earth. What would take Hubble 2,000 years to survey, Roman can complete in a single year. It is the best news in NASA science this month, and it arrived on the same week as the proposal to gut the programme that built it.

The Habitable Worlds Observatory — NASA’s next planned flagship, intended to directly image Earth-like exoplanets around nearby stars in the 2040s — depends critically on Roman’s coronagraph as its technology demonstration. Roman’s Coronagraph Instrument is the proof of concept for the starlight suppression capability that HWO requires. Cancelling the science infrastructure that feeds HWO’s development does not merely delay it; it removes the engineering confidence that HWO’s architecture currently rests on. The Roman coronagraph is not a luxury. It is a prerequisite. We examine it in technical detail in this issue’s first ever Technical Deep-Dive section below.

Section 02 · Market Structure

Blue Origin’s Partial Victory: What NG-3 Actually Proved

New Glenn launched on April 19 from Cape Canaveral. The first stage — “Never Tell Me the Odds,” previously flown on NG-2 — executed a clean separation, re-entered, and landed successfully on the droneship Jacklyn in the Atlantic. It was the first reuse of a New Glenn booster, and the first stage performed exactly as Blue Origin needed it to. Then the upper stage failed.

One of the BE-3U vacuum engines did not produce sufficient thrust on a critical second burn. The payload — BlueBird-7, AST SpaceMobile’s broadband satellite — was left in an orbit significantly lower than planned, too low to sustain operations. The FAA classified the event as a mishap and has grounded New Glenn pending investigation. Blue Origin had planned up to twelve additional launches this year; all are now contingent on the FAA’s findings and the upper stage anomaly resolution timeline.

The engineering lesson here is important and often blurred in launch coverage. A launch vehicle is not a single system. It is at minimum two independent vehicles in sequence — a first stage whose job ends at separation and an upper stage whose job begins there. Blue Origin successfully demonstrated first-stage reuse for the first time. It also revealed an unresolved reliability problem in its upper stage. Both facts are true simultaneously and should be reported simultaneously. The booster landing was not diminished by the upper stage failure. The upper stage failure was not mitigated by the booster landing.

The AST SpaceMobile dimension adds commercial stakes to the anomaly. BlueBird-7 was intended to expand AST's direct-to-device broadband constellation, which is the most credible commercial challenger to Starlink's dominance in satellite-to-smartphone connectivity. A lost satellite is a setback for a programme that is itself under investor scrutiny. Blue Origin must resolve the BE-3U anomaly, complete the FAA mishap investigation, and demonstrate corrective action before NG-4 can proceed — all while its commercial customers watch and assess whether New Glenn's reliability profile justifies its place in their launch plans.

For context, SpaceX passed its 600th Falcon 9 booster landing this week and crossed 1,000 Starlink satellites launched in 2026, bringing the active constellation above 10,200. These are not records being chased — they are the operational tempo of a mature industrial programme. That is the gap Blue Origin is working to close, and NG-3’s upper stage anomaly has widened rather than narrowed it for now.

Section 03 · Defence & Policy

GPS III Closes, Laser Comms Opens

The GPS III-8 satellite — named “Hedy Lamarr” by Space Force, after the actress and inventor whose World War II-era frequency-hopping research underlies the spread-spectrum techniques at the heart of GPS, WiFi, and Bluetooth — launched this week on a Falcon 9, completing the GPS III production block. Space Force framed the launch explicitly as the foundation for the GPS IIIF generation that will follow: more capable, more survivable, and architecturally updated for the contested environment the Symposium spent three days discussing.

The embedded story is more technically interesting than the block completion. GPS III-8 carries an optical inter-satellite link demonstration — a laser communications system being validated on orbit before integration as a standard capability on GPS IIIF satellites. This matters for two reasons. First, laser inter-satellite links offer dramatically higher bandwidth and lower probability of interception than radio frequency crosslinks — both relevant to a constellation operating in a contested environment. Second, this is the first laser comms demonstration on a GPS satellite, and its success or failure will directly influence the architecture of the next generation of the most critical space-based infrastructure the US operates.

The naming of GPS III-8 after Hedy Lamarr is not merely ceremonial. Lamarr and composer George Antheil filed US Patent 2,292,387 in 1942 for a frequency-hopping spread-spectrum communication system intended to make torpedo guidance signals unjammable. The patent expired before it could be commercialised, but the underlying technique became the foundation of CDMA, GPS signal structure, WiFi, and Bluetooth. Naming the satellite that closes the GPS III block after the woman whose wartime engineering insight underlies the entire system is one of the more historically coherent decisions the Space Force has made.

Section 04 · In-Orbit Compute — Our Core Thesis

Demo to Production: The Week AI Actually Moved to Orbit

This is where this publication dwells most deeply each week — the meat of our editorial ethos. Everything else is context. This is the signal. And this week, the signal crossed a threshold that previous issues could only anticipate: orbital AI inference moved from engineering demonstration to production operation, on a commercial constellation, generating commercially relevant outputs. That is not a press release. It is a category shift.

Planet Labs published results this week from its Pelican-4 satellite, which ran an object-detection AI model directly on its NVIDIA Jetson Orin compute unit, detecting aircraft over Alice Springs, Australia. The satellite produced geo-rectified GeoJSON outputs — structured, analysis-ready geospatial data — directly in orbit, without downlinking raw imagery. The latency from collection to actionable intelligence dropped from hours to minutes. The downlink bandwidth requirement dropped from gigabytes of raw image data to kilobytes of derived detections. The initial inference accuracy was approximately 80%, which Planet considers sufficient for operational use at this stage and expects to improve as models are refined and updated on-orbit.

The significance is architectural, not merely technical. Planet’s Pelican constellation is one of the largest daily-imaging Earth observation constellations in operation. Deploying production AI inference at the constellation level — not on a testbed satellite, but on an operational platform — transforms the economic model of Earth observation. The value chain that previously ran from satellite collection through ground processing to customer delivery can now short-circuit the ground processing step for a growing class of detection tasks. Insurance, agriculture, defence, logistics, infrastructure monitoring: every sector that consumes derived geospatial intelligence is affected by a latency reduction from hours to minutes.

The Jetson Orin is not an H100. It is an edge inference chip — drawing 10 to 60 watts depending on configuration, designed for embedded AI applications in robotics and autonomous vehicles. Its relevance for orbital compute is precisely its power efficiency: it can run continuous inference within the thermal and power budgets that small satellites can sustain, without the radiator area and power generation infrastructure that high-end data-centre GPUs require. Planet’s deployment validates the edge-inference architecture that Kepler Communications described last week as its core thesis. The orbital compute sector is bifurcating: high-performance training and heavy inference on large platforms such as Starcloud-2, and distributed edge inference on operational constellations such as Planet and Kepler. These are complementary, not competing.

Two further developments this week advance the picture. Belgian spacetech company EDGX launched its STERNA compute platform on SpaceX’s Transporter-16 rideshare mission, with two hosted payloads now operational. STERNA is an NVIDIA-powered edge computer that scales dynamically between 10 and 45 watts — the dynamic power range designed to match available solar generation across the sunlit and eclipse portions of a LEO orbit. EDGX is offering compute-as-a-service commercially in 2026, positioning itself as a European-jurisdiction orbital compute provider at a moment when data sovereignty concerns are giving European customers structural reasons to prefer non-US orbital infrastructure.

And Lonestar Data Holdings announced StarVault this week — described as the world’s first commercially operational space-based sovereign data storage platform, targeting a launch in October 2026. Lonestar reports that demand from governments and financial institutions already exceeds expectations, with sovereign data requirements — data that must be stored under a specific national jurisdiction — driving early interest. The storage use case is distinct from the compute use case but shares the same underlying infrastructure logic: data that is generated, processed, or stored in orbit is subject to different jurisdictional rules than data handled on terrestrial servers. The legal architecture of orbital compute is as underdeveloped as the technical architecture is nascent. Both are being built simultaneously, in real time, with commercial customers already making purchasing decisions.

EWC intersections this week: Planet’s production AI inference pipeline — collect, infer, output GeoJSON — is structurally identical to the EWC Compute platform’s sensor-to-model-to-output architecture. The EDGX dynamic power scaling between 10 and 45 watts is a thermal management problem with a direct Precision with Light parallel in photonic integrated circuit thermal stability. The Lonestar sovereign data jurisdiction question will become relevant to any EWC platform that handles regulated engineering data. The GPS III-8 laser link demonstration is a free-space optical engineering programme sitting squarely in the PwL domain.

Section 05 · One to Watch

The Security Layer Nobody Is Writing About

As compute migrates to orbit, the cyberwarfare attack surface migrates with it. Deloitte — not a name that typically appears in space infrastructure coverage — announced this week that its Deloitte-2 and Deloitte-3 satellites are now operational alongside Deloitte-1, all carrying an on-orbit intrusion detection system called Silent Shield. Project Constellation is planned to eventually include nine satellites, with a software-only version of Silent Shield that can be pushed to third-party satellites already in orbit — an over-the-air security update for hardware that was never designed to receive one.

The problem Silent Shield addresses is real and largely ignored in the orbital compute coverage. A satellite running an NVIDIA GPU and an AI inference model is, architecturally, a compute node on a network. It has a command uplink, a data downlink, software that can be patched, and models that can be poisoned. The security assumptions of early satellite operations — that the hardware was purpose-built, the software was fixed at launch, and the communication protocols were proprietary and obscure — are invalidated the moment you put a general-purpose GPU and a commercial operating system in orbit. The attack surface of an orbital data centre is not qualitatively different from that of a terrestrial one. It is, however, subject to different response times: a compromised terrestrial server can be isolated and patched in minutes; a compromised satellite may require a ground station contact window, a command uplink, and a software update cycle measured in orbits.

We will track this space carefully. The orbital cybersecurity layer is the most underreported dimension of the in-orbit compute sector, and it will become the most commercially consequential one as soon as the first significant breach occurs.

Section 06 · Numbers of the Week

April 29, 2026 — The Week in Figures

Section 07 · Technical Deep-Dive — Issue 003 Inaugural

The Roman Space Telescope: What Modern Optical Engineering for Space Actually Looks Like

The Nancy Grace Roman Space Telescope is, by any measure, a remarkable optical engineering achievement. It is not the largest space telescope ever built, nor the one operating at the shortest wavelengths. What makes it remarkable is the combination of aperture, field of view, detector technology, and coronagraph sophistication assembled into a single observatory — and the fact that its primary mirror arrived from the National Reconnaissance Office as a donation, having been built for an entirely different purpose and then rendered surplus to intelligence community requirements. Understanding what Roman actually is, optically, requires working through each of those elements in turn.

The Primary Mirror and Optical Heritage

Roman’s 2.4-metre primary mirror is the same aperture as the Hubble Space Telescope — but it is not the same telescope. The NRO mirror was built with a shorter focal length and a correspondingly wider field of view. Where Hubble operates at focal ratio f/24 — a long, narrow beam that produces high magnification over a small field — Roman operates at f/7.9, a much faster optical system that trades magnification for sky coverage. The consequences of this choice propagate through the entire instrument design. A fast focal ratio means the focal plane subtends a much larger angle on the sky; the same physical detector area maps to far more square degrees of coverage. This is the fundamental optical trade that defines Roman: speed and area over magnification and depth.

The f-number of a telescope is the ratio of focal length to aperture diameter. Hubble’s f/24 means its focal length is 24 times its mirror diameter — 57.6 metres of effective focal length folded into a compact tube via its Cassegrain design. Roman’s f/7.9 means 19 metres of effective focal length. Lower f-number means a wider cone of light — faster in photographic terms, larger in angular coverage. For a survey telescope whose purpose is to map billions of galaxies, an f/7.9 system is far better suited than f/24. The NRO mirror’s original purpose — high-resolution imaging of small ground targets from orbit — required precisely the opposite characteristic. The donation was physically appropriate for Roman’s mission because the aperture was right, not because the original design intent was aligned.

The Wide Field Instrument: 300 Megapixels of HgCdTe

Roman's primary science instrument is the Wide Field Instrument — a 300.8-megapixel camera covering 0.48 to 2.30 micrometres, from the blue visible through the near-infrared. The detector material is mercury cadmium telluride (HgCdTe), a ternary semiconductor compound whose bandgap can be tuned by adjusting the mercury-to-cadmium ratio, allowing sensitivity to be optimised for specific wavelength ranges. This is not a conventional silicon CCD. Silicon is transparent to photons with wavelengths above approximately 1.1 micrometres — the near-infrared regime where Roman does much of its cosmological science. HgCdTe detects efficiently out to 2.3 micrometres, making it the detector technology of choice for infrared space astronomy.

The focal plane consists of 18 individual H4RG-10 detectors from Teledyne Technologies, arranged in a 6×3 array. Each H4RG-10 is a 4096×4096 pixel array with 10-micrometre pixel pitch, designed specifically for space astronomy applications. The “H4RG” designation refers to Hawaii-4RG — a lineage of HgCdTe detector arrays developed initially for ground-based telescopes, progressively space-qualified, and now flying on both JWST and Roman. The combined focal plane provides a 0.28 square degree field of view at a resolution of 0.11 arcseconds per pixel — comparable to Hubble’s resolution over a patch of sky approximately 100 times larger than Hubble’s cameras can cover in a single exposure.

The hexapod actuator system supporting the focal plane array deserves specific mention. The 18-detector array must maintain diffraction-limited focus across its entire extent throughout the mission lifetime, despite thermal changes as the telescope moves in and out of shadow. Rather than relying solely on a fixed mechanical design, the focal plane is mounted on a hexapod — a six-degree-of-freedom positioning mechanism — that can reposition it based on ground-commanded focus analysis. This is active optical alignment by ground control, not passive thermal stability. It is an engineering admission that no passive design can hold the focus of a 300-megapixel infrared focal plane to the required tolerance across years of thermal cycling.

The Coronagraph Instrument: Starlight Suppression at 10⁻⁹ Contrast

The Coronagraph Instrument (CGI) is where Roman’s most sophisticated optical engineering resides, and where its legacy for future missions is most consequential. CGI operates at shorter wavelengths than WFI — 575 to 825 nanometres, the visible regime — and its purpose is not wide-field survey but extreme precision: suppressing the light from a star to a contrast ratio of 10⁻⁹, revealing planets and circumstellar discs that are up to a billion times fainter than their host stars.

The optical train of CGI is built around eight off-axis paraboloidal (OAP) mirrors arranged as four relay pairs, creating four distinct pupil planes and four focal planes along the optical path. This multi-relay architecture is not complexity for its own sake. Each relay pair serves a specific purpose: the pupil planes are where spatial filtering, wavefront correction, and coronagraphic masks are applied; the focal planes are where the stellar image is formed and where occulting masks are inserted to block the star’s light while transmitting the surrounding region where planets orbit.

The Wavefront Control Problem and Dual Deformable Mirrors

Achieving a contrast ratio of 10⁻⁹ in a space observatory is not an alignment problem. It is a wavefront control problem, and it is one of the hardest problems in applied optics. To understand why, consider what 10⁻⁹ contrast requires: the light from the star must be suppressed by a factor of one billion relative to the planet. Any residual optical aberration — any deviation from a perfect wavefront across the telescope aperture — scatters starlight into the region of the focal plane where the planet signal resides, raising the noise floor and reducing contrast. The aberrations that matter at this level are not large manufacturing errors. They are nanometre-scale surface irregularities, sub-nanometre thermal deformations as the telescope passes between sunlight and shadow, and picometer-level vibrations from reaction wheels and other moving components.

CGI addresses this with two Boston Micromachines deformable mirrors, each carrying 48×48 independently actuated segments — 2,304 degrees of freedom per mirror, 4,608 total. These mirrors can correct the wavefront in real time as the telescope’s thermal state changes. The correction loop runs continuously: the Low Order Wavefront Sensor (LOWFS) measures wavefront errors at low spatial frequencies, the deformable mirrors apply corrections, and the resulting residual error is small enough that the coronagraphic masks can suppress the remaining scattered starlight to the required contrast floor.

CGI addresses this with two Boston Micromachines deformable mirrors, each carrying 48×48 independently actuated segments — 2,304 degrees of freedom per mirror, 4,608 total. These mirrors can correct the wavefront in real time as the telescope’s thermal state changes. The correction loop runs continuously: the Low Order Wavefront Sensor (LOWFS) measures wavefront errors at low spatial frequencies, the deformable mirrors apply corrections, and the resulting residual error is small enough that the coronagraphic masks can suppress the remaining scattered starlight to the required contrast floor.

Why This Matters Beyond Roman

The Roman Coronagraph Instrument is officially designated a technology demonstration. Its science programme — three months of observations spread across the first year and a half of operations — is intentionally limited, preserving most of Roman's time for the wide-field survey programme. But the technology demonstration is the point. Every technique validated by CGI in flight — the deformable mirror control algorithms, the Shaped Pupil mask designs, the post-processing pipelines, the calibration procedures — feeds directly into the Habitable Worlds Observatory, which will need to image Earth-analogue planets in reflected visible light at contrast ratios approaching 10⁻¹⁰. Roman proves the approach works in the space environment. HWO scales it up.

This is how space optical engineering progresses: not in single leaps but in validated increments, each mission de-risking the next. The Roman Telescope exists because Hubble proved space-based optical diffraction-limited imaging. JWST existed because Hubble proved space infrared astronomy and space optical fabrication. HWO will exist — if it is funded and built — because Roman proved space coronagraphy. The 46% science budget cut proposed this week does not merely threaten Roman’s successor. It threatens the chain of validated learning that Roman represents the current link in.

References & Sources

Issue 003 — Primary References

Section 01 — Human Spaceflight & Policy

• NASA’s Nancy Grace Roman Space Telescope will launch in September — Scientific American

• The Nancy Grace Roman Space Telescope is ready to fly — Sky & Telescope

• Roman: NASA’s next great observatory is finally complete — Space.com

• House Science Committee chair opposes NASA budget cuts — SpaceNews

Section 02 — Market Structure & Blue Origin NG-3

• New Glenn reused booster; upper stage anomaly leaves satellite in wrong orbit — Spaceflight Now

• Blue Origin reuses New Glenn booster — but loses payload to orbit anomaly — Ars Technica

• FAA investigating New Glenn NG-3 anomaly — FAA newsroom

Section 03 — Defence & Policy

• GPS III-8 “Hedy Lamarr” launch completes the GPS III block — SpaceNews

• GPS III-8 optical inter-satellite link demonstration — US Space Force

Section 04 — In-Orbit Compute

• Planet Labs runs production AI inference in orbit — TechCrunch

• EDGX STERNA orbital edge compute launches on Transporter-16 — TechCrunch

• Lonestar announces StarVault sovereign orbital data storage — SpaceNews

Section 05 — One to Watch

• Deloitte Project Constellation and Silent Shield orbital intrusion detection — SpaceNews

Section 07 — Technical Deep-Dive: Roman Telescope

• Nancy Grace Roman Space Telescope — Wikipedia (technical specifications)

• Roman Space Telescope — MAST Archive / STScI

• Roman Coronagraph Instrument Overview and Status — arXiv:2309.08672

• Roman Coronagraph Calibration Plan — arXiv:2202.05923

• NASA completes Roman Space Telescope construction — NASA JPL

• Nancy Grace Roman Space Telescope — NASA Goddard official site

EWC Space Delta-V is a weekly technology intelligence publication by Engineering World Company, published every Tuesday. It covers the intersection of space infrastructure, AI compute, and the physics that governs both. Nuno Edgar Nunes Fernandes is a physics engineer with a background in optoelectronics and photonics, building AI-assisted engineering platforms across photonics, quantum computing, and industrial simulation.

Editor’s Note

The Week After the Moon

Issue 1 of EWC Space Delta-V covered a historic week — Artemis II returned humans from the vicinity of the Moon, SpaceX filed for the largest IPO in financial history, and a startup proved that a data-centre GPU could survive in orbit. That was the week the world was watching.

This week’s edition - reflecting SpaceTech developments taking place in the lasy week and during the weekend - was the week the industry watched itself. The 41st Space Symposium ran April 13–16 at The Broadmoor in Colorado Springs — the gathering colloquially known as the “Davos of Space,” 12,000 attendees from 60 countries, convening immediately in the afterglow of Artemis II. The announcements made on that stage were, in some respects, more consequential than the mission that preceded them. The White House released a nuclear power policy. NASA unveiled a strategic pivot. The orbital compute ecosystem made its most concrete engineering advance yet. And Blue Origin quietly prepared to cross the threshold that separates a reusability demonstration from an operational reusability programme. None of this made front pages. All of it matters.

Section 01 · Human Spaceflight

From Celebration to Strategy: NASA’s “Ignition” Pivot

Artemis II splashed down on April 10. By April 14, NASA Administrator Jared Isaacman was on stage at the Space Symposium outlining what comes next — and the shape of it is significantly different from what the Artemis programme looked like a year ago.

Under a newly unveiled initiative called “Ignition,” NASA is formalising a pivot from the expeditionary model of early Artemis — fly to the Moon, take samples, return — toward something more ambitious and more demanding: a permanent human presence at the lunar South Pole by 2030. The orbital Gateway station, which was the planned stepping-stone between Earth and the surface, has been formally deprioritised and its resources redirected toward heavier surface infrastructure. Cargo Human Landing Systems will ferry habitation modules to the surface in a “near monthly” cadence of drone missions beginning early next year. Between 2033 and 2036, NASA plans to deliver approximately 150,000 kilograms of payload to the Moon, including pressurised rovers capable of sustaining astronauts outside their suits for extended durations.

This is the engineering consequence of cancelling the Gateway: the cislunar waypoint disappears and all capability must be delivered directly to the surface. That raises the Δv budget for every subsequent mission, concentrates risk on the lunar landing system, and eliminates the orbital infrastructure that was intended to support long-duration stays. It is a bolder strategy and a harder one.

The more striking announcement was the nuclear power timeline. On April 14, the White House released a six-page policy — designated NSTM-3 — directing NASA, the Department of Defense, and the Department of Energy to develop space nuclear power systems that could launch as soon as 2028. The policy is concrete: NASA is directed to begin work within 30 days on a mid-power space reactor generating at least 20 kilowatts, with a variant designed for the lunar surface. It calls for parallel design competitions, specifies extensibility to at least 100 kilowatts for later designs, and assigns binding timelines to federal agencies — not aspirational language, but a formal interagency directive.

The first deliverable is the SR-1 Freedom — a nuclear electric propulsion spacecraft that Isaacman described as “the first-of-its-kind nuclear-powered interplanetary spacecraft,” targeting a 2028 launch. By 2030, a scaled fission surface system called Lunar Reactor-1 is to be deployed on the Moon, providing the 20-kilowatt continuous power supply required to sustain operations through the 354-hour lunar night — the period during which solar power is non-viable and during which any permanent base must remain operational.

Isaacman noted that NASA has spent over $20 billion on nuclear power and propulsion projects over the past several decades, none of which flew. The explicit intent of NSTM-3 is to reverse that pattern. "We're taking it out of the lab," he said — drawing directly on the Navy's nuclear reactor development model as the operational template. The last time the United States operated a nuclear reactor in space was approximately 60 years ago. The procurement and regulatory precedent established by SR-1 Freedom will determine whether the 2030s nuclear space infrastructure programme is real or aspirational.

The defence dimension is not peripheral. Todd Harrison of the American Enterprise Institute noted at the Symposium that a reliable nuclear power source in orbit unlocks capabilities that currently lack a credible power supply: space-based data centres running at scale, persistent missile warning, directed energy systems, and strategic communications nodes that cannot be power-starved by adversarial action. The nuclear policy and the orbital compute thesis are not separate stories.

Section 02 · Market Structure

Blue Origin Steps Across the Line

The reusability story of the space industry has been, for a decade, almost entirely a SpaceX story. Falcon 9’s booster programme redefined launch economics; the Falcon 9 fleet has now completed over 600 booster landings, and individual stages have flown more than 25 times. No other orbital vehicle had crossed from landing a booster to reflying one — until this weekend.

Blue Origin’s New Glenn executed a static fire of its first previously flown booster on April 16, clearing the final pre-launch hurdle for the NG-3 mission targeting April 19. The booster — named “Never Tell Me the Odds” — previously flew and landed successfully on the NG-2 mission in late 2025, which delivered NASA’s EscaPADE twin probes toward Mars. The decision made for this reflight is technically revealing: Blue Origin replaced all seven BE-4 engines rather than inspecting and recertifying the originals, added a thermal protection system upgrade on one engine nozzle, and reserved the flown engines for future missions. CEO Dave Limp was direct about the reasoning. The company elected to swap hardware where uncertainty was highest and treat the NG-3 flight as a learning data point rather than a cost-optimised reuse.

This is the correct engineering posture for a first reflight — the same logic that led SpaceX to over-inspect early Falcon 9 refurbishments before the cadence data justified faster turnarounds. But it also reveals exactly where Blue Origin stands on the long road from recovering hardware to reusing it economically. Each New Glenn booster is designed for up to 25 flights. SpaceX arrived at 25-flight cadence through years of progressive data accumulation, not through declared ambition. Blue Origin has declared its ambitions; NG-3 is the beginning of accumulating the data that will either validate or revise them.

The payload for NG-3 is BlueBird-7, AST SpaceMobile's broadband satellite. This matters beyond the reuse milestone: AST SpaceMobile's constellation is the most credible commercial challenger to Starlink's direct-to-device satellite internet dominance. A successful NG-3 gives Blue Origin a high-value commercial customer relationship at the moment it is trying to establish itself as a viable alternative launch provider to SpaceX. The mission is consequential on two axes simultaneously.

Meanwhile SpaceX passed two milestones this week with the quiet efficiency of industrial routine: the 600th Falcon 9 booster landing, and the 1,000th Starlink satellite launched in 2026 — on April 14, the same day Artemis II astronauts were attending post-splashdown ceremonies. The Starlink constellation now exceeds 10,200 active satellites. These numbers are significant not as records but as indicators of an infrastructure operating at a scale and cadence that no competitor currently matches. The New Glenn programme is not yet competing on cadence. It is competing on the existence of a credible alternative — which has its own market value, particularly for national security customers who have structural reasons to avoid sole-source dependency on a single provider.

Section 03 · Defence & Policy

Colorado Springs: The Symposium’s Underlying Signal

The themes on the Space Symposium stage this year were, in aggregate, a policy statement: contested satellite manoeuvring as the new battlefield; AI’s role in space operations; the convergence of national security and private space operations; lunar economics and permanent presence. These are not conference panel abstracts. They are the organisational logic of the US national security space programme in 2026.

The Andromeda programme we covered in Issue 001 — $1.8 billion to replace GSSAP geosynchronous surveillance satellites with commercial equivalents — is one expression of this logic. So is the nuclear power policy. So is the Proliferated Warfighter Space Architecture: hundreds of small, interconnected, expendable satellites rather than a small number of large, expensive, vulnerable ones. The through-line in all of these is the same: the US military is redesigning its space architecture around the same economic and technical principles that govern commercial cloud infrastructure. Resilience through redundancy. Capability through software. Acquisition through competition rather than sole-source programme management.

At the Symposium, OSTP Director Michael Kratsios placed the nuclear policy explicitly in this frame: nuclear power enables the persistent, high-availability infrastructure that the military requires in space — data centres, missile warning, directed energy — at power levels that solar arrays in current configurations cannot sustain. The policy is not separable from the compute and communication requirements driving the rest of the space architecture discussion. It is the power supply for the distributed system that the other programmes are building.

One signal worth tracking from the Symposium periphery: New Glenn has been awarded a West Coast launch site at Vandenberg Space Force Base, designated SLC-14, announced on April 14. This gives New Glenn polar orbit capability — access to sun-synchronous orbits critical for Earth observation and the reconnaissance satellite market. Blue Origin is not simply building a heavy-lift rocket. It is building the launch infrastructure necessary to compete for the national security payloads that currently flow exclusively to SpaceX and ULA.

Section 04 · In-Orbit Compute — Our Core Thesis

The Mesh Is Live: Distributed Edge Compute Arrives in Orbit

This is the part of each issue where we dwell most of our writing, attention and detailed analysis — it represents the meat of this publication’s ethos. Everything else in space news is context. This is the signal.

Last issue we covered Starcloud: one GPU satellite, a single H100, a proof of concept that data-centre-class silicon could survive in orbit. This week the orbital compute story advances architecturally, and in a direction that is more immediately realisable than the large orbital data centre vision. On April 13, TechCrunch reported that Canada’s Kepler Communications has brought online the largest compute cluster currently in orbit — 40 NVIDIA Orin edge processors distributed across 10 operational satellites, all interconnected by laser communications links. The cluster is not a monolith. It is a network.

Kepler does not describe itself as a data centre company. CEO Mina Mitry frames its mission as infrastructure for applications in space — a networking and processing layer that other satellites can use, rather than a stand-alone compute facility. The distinction is architecturally significant. The Starcloud model requires a single large platform to carry substantial compute and power generation; the Kepler model distributes modest compute across many nodes and connects them with high-bandwidth laser links. These are two different approaches to the same problem, drawing on two different engineering philosophies: concentrated compute versus distributed edge processing.

Kepler’s thesis is that the near-term orbital compute workload is inference, not training. “Because we have the belief it’s more inference than training, we want more distributed GPUs that do inference, rather than one superpower GPU that has the training workload capacity,” Mitry told TechCrunch. “In our case, our GPUs are running 100% of the time.” The utilisation argument is compelling: a 20-kilowatt GPU satellite running at 10% utilisation because it is waiting for a large training job is an expensive satellite. Forty edge processors running continuous inference on sensor data they collected moments earlier is a different economic structure entirely.

The partnership announced this week between Kepler and Sophia Space sharpens the picture further. Sophia is developing passively cooled space computers — and this is where the engineering gets interesting. Passive cooling in the context of a GPU satellite means designing the thermal system to reject heat by radiation alone, without active refrigeration loops, pumps, or heavy deployable radiators. For a small edge processor operating at modest power, this is tractable. For an H100-class GPU operating at several hundred watts, it is a significant constraint. Sophia’s approach is to build hardware that fits within the passive thermal envelope rather than expanding the envelope to accommodate commodity silicon.

In the partnership, Sophia will upload its operating system to one of Kepler’s satellites and attempt to configure and run it across six GPUs on two spacecraft. This is the first time multi-GPU orchestration will be attempted in orbit — the kind of operation that is trivially routine in a terrestrial data centre, and that has never been demonstrated in the radiation environment, thermal cycling, and communications latency of low Earth orbit. If it works, it de-risks a critical capability for Sophia’s first planned satellite launch in late 2027 and validates Kepler’s network as a testbed for orbital software development at a cost far below the price of launching dedicated hardware.

The thermal physics of orbital compute deserves more attention than it typically receives in coverage of this sector. Space is cold — approximately 3 kelvin in deep space — but that is an incomplete picture. A satellite in LEO spends roughly 60% of its orbit in sunlight, absorbing solar radiation, and 40% in eclipse. The thermal environment cycles between extremes on every 90-minute orbit. More critically, space is a vacuum: there is no air to carry heat away by convection. All heat generated by electronics must be rejected by radiation — by emitting infrared from the surface of the satellite. The Stefan-Boltzmann law governs this: radiated power scales as the fourth power of temperature, and the effective radiating area is physically constrained by the satellite's geometry and mass budget. An H100 GPU dissipates approximately 700 watts at full load. Radiating 700 watts passively requires a radiator area that conflicts directly with the mass and volume constraints of a small satellite. This is not a solvable problem by cleverness alone. It is a physics constraint, and it is the principal reason that Sophia's passively cooled chip architecture — rather than commodity data-centre silicon — may prove to be the more scalable approach to orbital compute.

There is a further data point from the Starcloud story that deserves to be surfaced explicitly. In the TechCrunch Series A coverage, Starcloud CEO Philip Johnston disclosed that an NVIDIA A6000 board — a second GPU onboard Starcloud-1 — failed during launch. This is not a failure of the space environment; it is a failure of launch loads: vibration, acoustic pressure, and mechanical shock at levels that commodity data-centre hardware is not qualified to survive. The H100 that has operated successfully in orbit did so; the A6000 did not make it past the launch. This is precisely the kind of detail that press releases omit and engineering analyses require. The radiation environment in LEO is harsh but manageable with appropriate shielding. The launch environment is violent, and qualifying commercial silicon to survive it is a non-trivial engineering programme in its own right.

Finally, a detail from the TechCrunch article that deserves its own paragraph, because it signals something about the terrestrial policy environment that the orbital compute sector will increasingly use as a market argument: the state of Wisconsin last week adopted a ban on new data centre construction. Several members of Congress are reportedly pursuing similar restrictions at the federal level, citing energy grid impacts, water consumption, and local opposition. Whether or not these bans spread, the fact that they exist at all changes the conversation about orbital compute from “interesting long-term thesis” to “potentially near-term regulatory arbitrage.” Sophia CEO Rob DeMillo’s comment — “There’s no more data centers in this country. It’s gonna get weird from here” — is not a joke. It is a market signal.

EWC Compute intersections this week are direct and multiple. The Kepler laser inter-satellite links are a free-space optical communications engineering problem — link budget, pointing acquisition and tracking, coherent vs. intensity modulation — that sits squarely within the Precision with Light platform domain. Sophia's passively cooled chip architecture is a thermal engineering and photonic integration problem: photonic interconnects generate far less heat per bit than electronic equivalents, which is why photonic compute architectures are being studied for the same power-constrained environments. The distributed GPU orchestration problem across Kepler's constellation is architecturally analogous to the EWC Compute platform's multi-node simulation bridge. These are not analogies. They are the same class of problem in different physical environments.

Section 05 · One to Watch

SR-1 Freedom: Nuclear Propulsion Leaves the Laboratory

LONG SIGNAL

The SR-1 Freedom programme deserves more than a footnote in the nuclear reactor announcement. It is the first US nuclear electric propulsion spacecraft — a vehicle that uses a fission reactor to power ion or Hall-effect thrusters, achieving specific impulse values that chemical propulsion cannot approach. The physics case for nuclear electric propulsion on deep space missions is compelling: Isp values in the range of 3,000–10,000 seconds compared to approximately 450 seconds for the best chemical engines. For a Mars cargo mission, the Δv advantage translates directly into payload fraction — far more mass delivered per launch mass. The rocket equation favours nuclear electric propulsion for any mission beyond the Moon with a significant payload requirement.

The reason this has not been done before — NASA has spent over $20 billion on nuclear space power and propulsion across several decades without flying a single system — is not physics. It is procurement culture, regulatory complexity, launch approval processes for nuclear materials, and the institutional risk aversion of a programme that cannot afford a visible failure. NSTM-3’s explicit reference to the Navy nuclear reactor programme as a model is instructive: the Navy succeeded by accepting a doctrine of progressive deployment — launch something that works at modest power, accumulate flight heritage, then scale — rather than pursuing the 100% solution before committing. Isaacman’s phrase “we’re not trying to nail the 100% solution” applied to SR-1 Freedom suggests a similar doctrine. A 20-kilowatt reactor flying in 2028 is not the endgame. It is the regulatory and engineering precedent for everything that follows.

We will track SR-1 Freedom closely. The 30-day window for NASA to begin design competition work starts now. The first contractor selections will reveal whether this programme has the procurement velocity its timeline requires, or whether it will join the long list of nuclear space initiatives that generated excellent reports and no hardware.

The Week in Figures

References & Sources

Editor’s Note & Symposium Context

• 41st Space Symposium — Space Foundation

Section 01 — Human Spaceflight & NASA Ignition

• NASA shifts focus to permanent lunar base and nuclear propulsion — SatNews

• White House releases space nuclear policy (NSTM-3) — SpaceNews

• NASA chief: US needs nuclear-powered interplanetary spacecraft — Washington Times

• NASA & DOE to develop lunar surface reactor by 2030 — NASA official release

Section 02 — Market Structure & Blue Origin

• Blue Origin hot fires first previously flown booster — Spaceflight Now

• Blue Origin reusing New Glenn stage for first time — Space.com

• Blue Origin’s second life: New Glenn booster reuse attempt — Space Daily

• SpaceX launches two Starlink groups, 1,000th satellite of 2026 — Space.com

Section 03 — Defence & Policy

• Global space and defence industry at Space Symposium 2026 — Colorado Springs Chamber

• Nuclear reactors in orbit by 2028, Moon by 2030 — Eurasian Times

Section 04 — In-Orbit Compute

• The largest orbital compute cluster is open for business — TechCrunch

• Sophia Space raises $10M seed to demo novel space computers — TechCrunch

• Why the economics of orbital AI are so brutal — TechCrunch

• Four things we’d need to put data centres in space — MIT Technology Review

• Starcloud raises $170M Series A — TechCrunch

• Will data centres in space work? — NPR

Section 05 — One to Watch

• Colorado’s key role in probing outer space — Colorado Springs Gazette

EWC Space Delta-V is a weekly technology intelligence publication by Engineering World Company. It covers the intersection of space infrastructure, AI compute, and the physics that governs both. Nuno Edgar Nunes Fernandes is a Physics Engineer with a background in optoelectronics and photonics, building AI-assisted engineering platforms across photonics, quantum computing, and industrial simulation.

What EWC Space Delta-V Is, its Leitmotiff

EWC Space Delta-V exists because the space industry these days generates more than regular news, it has real signals for our lives, economies and Societies. There are publications that track space technologies, launches, spacetech related funding rounds, and publications that translate press releases into readable prose. What is rarer — and what we intend to be — is a publication written from inside the engineering stack: one that reads the physics, the engineering specifics before it reads the headline, that asks what a claim actually requires to be true before amplifying it, and that treats the rocket equation as a real constraint rather than a metaphor.

The leitmotiff of this publication is the convergence of space infrastructure and compute. Orbit is becoming a distributed computing environment. That thesis carries implications for every domain we work across — photonics, physical AI, digital twins, simulation — and it is the lens through which we will read each month’s news. We will not be completely certain, as is normal, sometimes. We will name those mistakes when they happen. What we will not do is mistake motion for progress, or a press release for an engineering result. Welcome to the first issue.

Section 01 · Human Spaceflight

The Mission That Actually Mattered

On April 1, NASA launched Artemis II from Kennedy Space Centre. On April 10, the Orion spacecraft — named Integrity by the crew — splashed down in the Pacific off San Diego at 24,661 miles per hour, decelerating through 35 times the speed of sound, with four astronauts inside. Commander Reid Wiseman, pilot Victor Glover, mission specialist Christina Koch, and Canadian astronaut Jeremy Hansen had spent ten days travelling 252,756 miles from Earth — a new record, surpassing Apollo 13’s 248,655 miles, a distance Apollo 13 reached not by design but by catastrophe. Artemis II reached it by plan.

The spectacle has been covered extensively. What deserves more attention are the anomalies, because this was a test flight and the anomalies are the point. The mission accumulated a non-trivial list of items requiring engineering action before Artemis III can proceed. The heat shield — made of an ablative material called Avcoat, the same chemistry used in Apollo — returned with divots and cracking. This was anticipated, and NASA is planning a material redesign for Artemis III, but the post-splashdown inspections are now underway and the data will be closely watched. The service module requires a valve redesign. There was a small helium leak in the European Service Module propulsion system; non-critical to mission safety, but requiring characterisation. The toilet malfunctioned. And the planned skip-reentry profile — which briefly dips into the upper atmosphere to dissipate energy and improve landing precision — was replaced with a steeper direct entry after heat shield erosion was observed in Artemis I. That decision closed a capability that may be needed for future long-duration deep space returns.

The Lunar Gateway was cancelled in March 2026. This removes a key piece of the original Artemis architecture and shifts the programme toward direct surface access. Artemis III is now aimed at the lunar South Pole — arguably the most technically demanding landing site ever attempted, chosen for its ice resources rather than its accessibility. The engineering challenges compound.

None of this diminishes what was achieved. Humans flew beyond low Earth orbit for the first time in over fifty years, and came home safely. The Orion capsule performed. The SLS flew its second mission. Glover became the first person of colour to travel around the Moon; Koch the first woman; Hansen the first non-American. These are genuine firsts. But the honest engineering read of April 2026 is that Artemis II proved the hardware can get there and back, while simultaneously generating a precise list of what needs to change before we go further. That list is the useful output. The celebration is warranted, but the work it reveals is what matters for Artemis III.

Section 02 · Market Structure

The Trillion-Dollar IPO and What It Means for the Stack

On April 1 — the same day Artemis II lifted off — SpaceX filed confidentially with the Securities and Exchange Commission, targeting a Nasdaq listing in June at a reported valuation of $1.75 trillion. If that valuation holds at listing, it would be the largest IPO in history by a substantial margin. For context, the current global space economy is estimated at $626 billion. SpaceX alone would be valued at nearly three times the annual economic output of the entire industry it dominates.

The market structure implications of a public SpaceX are significant and underappreciated. A public company has quarterly reporting obligations, shareholder accountability, and capital access that changes the tempo and nature of its investment decisions. The aggressive internal cross-subsidy that currently flows from Starlink’s revenues into Starship development becomes a line item that analysts will scrutinise. The pressure to demonstrate returns on the Starship programme — which currently has no commercial revenue — will intensify. This may accelerate commercialisation of Starship in ways that benefit the broader industry, or it may constrain the programme’s ambition in ways that a private company could absorb. We do not know which. We will watch the S-1 closely when it is filed publicly.

What is already clear is that SpaceX’s vertical integration — launch, connectivity, ground segment, increasingly compute — makes it structurally different from every other public aerospace company. It is not Boeing or Lockheed. It is closer to Amazon circa 2010: a logistics infrastructure company with embedded software services and a founder who treats market rules as constraints to route around. The space economy’s trajectory to $1 trillion by 2034 runs through whatever SpaceX becomes as a public company. That is either a very good or a very complicated thing, depending on whether you are building on that infrastructure or competing with it.

Section 03 · Defence & Policy

The Military-Commercial Blur Is Now Structural

The US Space Force is doubling its budget and the distinction between military and commercial space is collapsing not as a trend but as deliberate policy. Two developments in April crystallise this. First, Space Force awarded the Andromeda programme — formerly designated RG-XX — to fourteen companies, with a $1.8 billion ceiling through April 2036. Andromeda’s purpose is to replace GSSAP, the Geosynchronous Space Situational Awareness Programme: the satellites that perform close-proximity inspection and characterisation of objects in geosynchronous orbit. This was previously an exclusively government-operated capability. It is now being transitioned to commercial providers under rolling competitive task orders. The capability is the same; the ownership model is not.

Second, the broader budget push is explicitly designed to move away from a small number of large, expensive, and inherently vulnerable satellites toward a Proliferated Warfighter Space Architecture — hundreds of smaller, interconnected satellites built to be expendable and replaced rapidly. The model here is not aerospace procurement. It is closer to cloud infrastructure: resilience through redundancy, capability through software, unit economics that favour scale over per-unit capability. This is the logic of hyperscale data centres applied to national security space assets.

The engineering consequence is significant: when military space architecture converges on the same design philosophy as commercial satellite constellations, the technology base merges. Components, software stacks, ground systems, and operational practices developed for commercial LEO constellations become directly relevant to defence applications. The barrier between the two sectors, already low, approaches zero.

Section 04 · In-Orbit Compute — Our Core Thesis

Orbit as Data Centre: The Race Has Started

This is the part of this news channel’s first article to be of most focused attention by the potential interested readers/subscribers. It is here that we will dwell most of our writting, attention and detailed analysis, it represents the meat of the publication’s ethos. Everything else in space news is context. This is the signal.

In November 2025, a Washington-based startup called Starcloud launched a 60-kilogram satellite — roughly the size of a domestic refrigerator — carrying an NVIDIA H100 GPU into low Earth orbit at 350 kilometres altitude. The H100 represents approximately 100 times the GPU compute of anything previously operated in space. In December, the satellite ran Google DeepMind’s Gemma large language model in orbit and trained nanoGPT — Andrej Karpathy’s compact language model — on the complete works of Shakespeare. These were not demonstrations designed for press releases. They were engineering validation milestones: proof that data-centre-class silicon could operate reliably through vacuum, radiation exposure, and repeated thermal cycling between the sunlit and shadow portions of each 90-minute orbit.

In March 2026, Starcloud raised $170 million, reaching a $1.1 billion valuation — the fastest Y Combinator company to reach unicorn status, at 17 months post-demo-day. In February, it filed with the FCC for a constellation of up to 88,000 satellites. Starcloud-2 launches in October 2026, carrying NVIDIA’s Blackwell B200 chip alongside multiple H100s, with 100 times the power generation capacity of the first satellite. The Crusoe Cloud platform will deploy on that satellite, offering commercial GPU capacity from orbit beginning in early 2027.

Starcloud is not alone. At NVIDIA’s GTC conference in March, Jensen Huang announced the Space-1 Vera Rubin Module — a space-qualified compute platform delivering up to 25 times the AI performance of the H100 for orbital inferencing. Aetherflux, Axiom Space, Kepler Communications, Planet Labs, and Sophia Space are all working with NVIDIA accelerated platforms for space missions. The orbital compute ecosystem is forming around NVIDIA the same way the terrestrial AI infrastructure ecosystem did.

The physics case for orbital compute deserves honest examination, because it is simultaneously compelling and subject to hype. The genuine advantages are: near-continuous solar power without grid dependency, radiative cooling to a 3-kelvin heat sink with no water consumption, and latency-free processing of data generated in orbit — particularly relevant for synthetic aperture radar and hyperspectral imagery, where downlinking raw data to ground is increasingly the bottleneck. The genuine constraints are: the radiation environment causes bit errors that propagate through neural network weights in ways that are not yet fully characterised at H100-class compute densities; launch costs per kilogram, while falling, still impose strict hardware mass budgets; orbital debris risk makes large constellations a systemic fragility as well as a capability constraint; and the economics of competing with terrestrial hyperscale data centres — which are also improving — will require the launch cost curve to continue its current trajectory for a decade or more.

What Starcloud has demonstrated in Starcloud-1 is not that orbital data centres are economically viable at scale today. What it has demonstrated is that the hardware can survive and operate in the environment. That is the necessary first step, and it is a genuine engineering milestone, not a press release. The valuation reflects the option value on the trajectory, not the current economics. Both things can be true simultaneously.

From an EWC perspective: the intersection with our Precision with Light platform (Tidy3D adapter, optical domain simulation) and the laser inter-satellite link programmes (ESA Transporter-16 CubeSats, SpaceX Starlink ISLs) is direct. Free-space optical communication between orbital compute nodes is a photonics engineering problem. We will be reporting on this intersection in depth in coming issues.

But there’s even more to read:

Section 05 · One to Watch

NASA’s Commercial Station Programme: Hurry Up and Wait

SLOW SIGNAL

Less visible than Artemis but more structurally important for the long-term commercial space economy: NASA’s Commercial Low Earth Orbit Destinations (CLD) programme — the mechanism for replacing the International Space Station with privately operated successors — has slowed significantly after a burst of apparent urgency last year. NASA signalled acceleration in mid-2025, drafted a solicitation in September, and targeted awards by April 2026. Those awards have not materialised on schedule. The ISS is planned for deorbit in 2030. That leaves a narrowing window.

The tension is architectural. Commercial station developers like Starlab argue that NASA’s requirements culture — thousands of granular standards developed over decades of government spaceflight — is incompatible with the four-year timeline from contract to operational station. NASA’s safety requirements should drive design at the system level, the argument goes, not specify implementation details. NASA, for its part, cannot disclaim responsibility for crew safety on a vehicle it is paying to operate. This is not a bureaucratic dispute. It is a genuine engineering governance problem about where the boundary between customer specification and contractor design freedom should sit. We will track this closely, because it will determine whether commercial LEO stations exist before 2030 or after it.

Section 06 · Numbers of the Month

April 2026 at a Glance:

252,756

miles — new record for human deep space distance (Artemis II)

$1.75T

reported SpaceX valuation targeting June Nasdaq listing

$1.8B

Space Force Andromeda programme ceiling across 14 companies

$1.1B

Starcloud valuation, fastest YC unicorn ever (17 months)

$626B

global space economy in 2025, up 7% year-on-year

88,000

satellites in Starcloud’s FCC filing for orbital data centre constellation

EWC Space Delta-V is a monthly technology intelligence publication by Engineering World Company. It covers the intersection of space infrastructure, AI compute, and the physics that governs both. Nuno Edgar Nunes Fernandes is a physics engineer with a background in optoelectronics and photonics, building AI-assisted engineering platforms across photonics, quantum computing, and industrial simulation.

Next issue: May 2026. If you have links, papers, or programme developments worth tracking, send them in.