The workforce above the Kármán line

Physical AI in Space and the Orbital Data Center Economy

Above the Kármán line there is no labor pool. Physical AI is not a substitute for the space workforce; it is the space workforce, at any price. This review prices that claim against the 2025–26 flight record, the launch-cost curve, and 83 sourced company and program records.

Grant Markhart, Industrial Research Fellow · The Charlot Lab & The Hiner Lab, Institute for Physical AI @ BMI

Review draft · data as of 5 August 2026 · compiled from ~90 primary and trade sources across English, Chinese, Japanese, French, German, Italian, Korean and Russian coverage · companion to the Physical AI in space topic

1. Headline numbers

2. Scope, definitions and method

Physical AI here means AI systems that act on the physical world through machines: robotic manipulation, autonomous mobility and rendezvous, onboard perception and decision-making, and fleet-scale autonomous operations. The review tracks both the narrow enabling stack that analysts size — chips, sensors, actuation, robot software, $1.5B in 2026 growing at 47%/yr77 — and the broader economic footprint that shows up inside other industries' P&Ls, from $5 trillion humanoid forecasts78 to the autonomy that operates 11,000 satellites.

Orbital data centers are space-based facilities for general-purpose or AI compute and data storage, from edge processors on relay satellites through free-flying GPU nodes to proposed gigawatt constellations, including lunar and cislunar data storage. Space logistics and servicing covers orbital transfer vehicles, satellite life extension and repair, refueling, debris removal, space situational awareness, reentry and downmass logistics, and cislunar delivery.

Method. Five parallel research tracks were compiled on 4–5 August 2026: the orbital-DC landscape; logistics and servicing companies; space robotics and planetary operations; market sizing and unit economics; and a non-English sweep across Chinese, Japanese, French, German, Italian, Korean and Russian sources, which surfaced programs largely absent from US coverage — Shanghai's Xingshu Plan31, CASC's Tiangong Kaiwu resource program, Japan's Space Strategy Fund allocations95, Orbital Lasers' Series A100. Roughly 90 primary sources are cited. Claims resting on single or machine-translated sources are flagged [UNVERIFIED] in the companion database. Figures for private companies are as reported by cited outlets and are indicative.

3. The space economy in 2026: baseline and capital

Novaspace puts the 2025 space economy at $626.4B, with commercial activity at roughly 78% and government spending of $137.4B — for the first time majority-defense at 54%1,5. Downstream, space-enabled services remain the largest slab, $236B growing to $323B by 2034. Projections diverge more on definitions than on optimism: Novaspace sees $1.01T by 2034 at 5.5% CAGR2; McKinsey and the WEF project $1.8T by 2035 in real 2023 dollars at 9% CAGR, with reach applications — logistics, mobility, agriculture riding on PNT and Earth observation — growing faster than the space backbone itself3.

Launch is the master variable. 2025 set records: 329 orbital launch attempts, up 25% year on year, and ~3,200 tonnes to orbit, two-thirds of it SpaceX9,10. Falcon 9 list pricing sits near $2,700–3,600/kg; Starship V3 is flying but at only 1–2 flights per quarter as of mid-2026, and targets $100–200/kg at maturity, with Google modeling a ~20% learning rate reaching under $200/kg by the mid-2030s12,13. Government budgets tell a defense-led story: US Space Force FY2026 tops $40B including Golden Dome funds6, NASA landed at $24.4B after Congress rejected proposed cuts7, and ESA's Bremen ministerial subscribed a record €22.1B for 2026–288.

Capital reset the sector's ceiling in 2026. Space Capital counted a record $55.3B of private investment in 2025 and $31.6B in Q2 2026 alone54. The watershed was the SpaceX IPO: priced 11–12 June 2026 at $135/share, a $1.78 trillion offering valuation; SPCX closed its first day up 19% at a ~$2.1 trillion market capitalization, instantly a top-six US company57. The listing validated space as an asset class and concentrated it at the same time: one firm now spans launch, broadband and, via its AI1 filing, orbital compute.

4. Physical AI: the workforce above the Kármán line

On Earth, Physical AI competes with human labor, so adoption is a cost-curve story. In orbit the comparison collapses. EVA hours are scarce, dangerous and confined to two stations; everything else — assembly, repair, refueling, inspection, capture, mining — is robotic or does not happen. That inversion makes space simultaneously the hardest proving ground for Physical AI and its purest market.

4.1 What has actually flown

Flight heritage, not renders, is the discriminator. The deepest lineage is MDA's: Canadarm2 and Dextre have operated on the ISS for over two decades, and the CA$1B Canadarm3 is in build91. GITAI operated dual 1.5 m autonomous arms outside the ISS in March 2024 and completed its S3 servicing-satellite flight model in June 2026, then delayed its own flagship demo to 2028+ after winning US Space Force interceptor-prototype work, a telling sign of where money is pulling the field90. The 21 July 2026 launch of DARPA and Northrop's MRV carried two 7-jointed dexterous arms to GEO, the first US robotic servicing mission in that orbit, arriving ~2027 for a planned decade of repairs, relocations and propulsion-pod installs58,59.

4.2 Autonomy at fleet scale is the quiet headline

The largest Physical-AI system in existence is not a humanoid. It is Starlink's collision-avoidance autonomy, which executed 207,152 maneuvers in the December 2025–May 2026 half-year and over 355,000 in twelve months, roughly 40 maneuvers per satellite per year, triple the 2024 rate, screening at a 3-in-10-million collision-probability threshold79. This matters economically because every orbital-capacity model assumes it: MIT's source-sink carrying-capacity estimate of ~12.6 million satellites in LEO is conditional on exactly this class of active management105, and CO₂-driven thermospheric contraction is projected to cut safe capacity 50–66% by 2100106. Autonomy is what keeps the n²σv collision arithmetic, and with it the entire megaconstellation economy, on the stable branch of the Kessler curve107.

Between edge autonomy and gigawatt ambition sits the least glamorous, most consequential layer: making and assembling things in orbit. Today's working examples are small and real — Redwire's ISS pharmaceutical crystallization and bioprinting lines, Airbus's first metal parts 3D-printed in orbit (2024), and Varda's six recovered manufacturing capsules69,70. The next rung is structural: Arkisys's ground-demonstrated modular Port (first launch ~2027), Airbus's assembly roadmap, and the robotic in-orbit assembly that both ASCEND's 1 GW design and any multi-gigawatt compute cluster quietly assume38. The cautionary tale is governmental: NASA cancelled both OSAM-2 (2023) and OSAM-1 (2024) after overruns, while the DARPA-commercial hybrid RSGS survived to fly58. The lesson the market has drawn is that in-space industrialization advances when a paying operator, not a cost-plus program, owns the robot.

5. Orbital data centers: the global landscape

Twenty-two companies and programs now credibly claim the category. They stratify into four tiers: flying hardware (Starcloud, Kepler/Axiom, China's Three-Body/Star-Compute, HPE's ISS precursor); funded near-term builders (Cowboy Space, Sophia, Madari, Lonestar, NTT/Space Compass); hyperscale filings (SpaceX's 1M-satellite AI1 constellation, Blue Origin's 51,600-satellite Project Sunrise, Google's Suncatcher prototypes); and state-anchored programs (China's 2,800-satellite Star-Compute plan and Shanghai's 1,000-satellite Xingshu Plan31, the EU's ASCEND study38).

Eighteen months created the industry. China launched the first 12 Three-Body/Star-Compute satellites in May 202528,32; Starcloud-1 put the first NVIDIA H100 in orbit that November17; Google announced Project Suncatcher with an engineering paper days later13,14; ADA Space ran Alibaba's Qwen3 on an operational satellite, the first general LLM inference in orbit; Starcloud trained nanoGPT in orbit in December 2025, the first model trained in space17. In 2026: SpaceX filed for up to one million compute satellites in January20 and revealed the 150 kW AI1 design in June21; Blue Origin filed Project Sunrise's 51,600 satellites in March22; Kepler commissioned an NVIDIA-powered orbital cloud as NVIDIA launched its Space Computing line26,92; Starcloud raised $170M Series A at a $1.1B valuation16; and Shanghai unveiled the 1,000-satellite Xingshu Plan in July31,33.

Two structural observations. First, compute-adjacent infrastructure is arriving before the compute: optical relay (Kepler, Space Compass), edge processors, and standards. Second, the field's biggest names disagree in public. Musk claims space is the lowest-cost place to put AI within 2–3 years21; Bezos says gigawatts in 10–2023; Altman and AWS's Matt Garman dismiss near-term economics52; NVIDIA sells to all sides while Huang calls the economics poor right now53. When the principal beneficiaries of the hype are its loudest skeptics, discount rates matter.

6. The economics of orbital compute

The case reduces to one arbitrage: continuous, land-free solar power against the cost of getting mass to orbit and keeping it working. A dawn-dusk sun-synchronous orbit yields up to 8× the annual energy of the same panel at mid-latitudes, at over 95% capacity factor with no batteries, land, permits or water13,15. Terrestrial AI infrastructure meanwhile costs $30–45M per MW fully loaded, faces 18–36-month to 10-year grid interconnect queues, and will absorb $600–725B of hyperscaler capex in 2026 alone49,50.

Where the independent analyses converge. Strip away advocacy and the quantitative studies agree on the shape of the problem. Google's Suncatcher paper: at or below $200/kg, launch plus operations is roughly $810/kW-yr, inside the $570–3,000/kW-yr range of terrestrial data-center energy, with COTS TPUs proton-tested to ~2.7× a five-year mission dose and 1.6 Tbps optical links benched13. Forethought's independent model finds parity at ~$250/kg for space-solar-versus-off-grid and ~$100/kg for full competitiveness, flags a ~40% hardware overbuy from unrepairable failures, and notes the FAA's current launch-licensing ceiling would need a ~35× increase44. Aravolta lands at ~$200/kg with repairability as the binding constraint47. Against these, today's ~$1,500–3,600/kg says the trade is 10–30× out of the money, which is precisely why every serious player words its roadmap in Starship-cadence conditionals.

The skeptics' arithmetic. Thermal: in vacuum every watt must leave by radiation. A 700 W H100 needs ~1.4 m² of radiator at 60°C, growing ~40% over five years as surfaces degrade; a 40 kW rack needs ~80 m²; 100 MW implies ~2,500 such assemblies19. Radiators and arrays end up 65–70% of satellite mass. Flown radiator heritage runs ~14–17 kg/m²; the bulls' business cases assume advanced deployables at 3–10× better specific power44,45. Nanjangud's re-costing of Starcloud's 40 MW reference — ~900–1,100 tonnes of radiators, 17–22 launches rather than one — is the sharpest published rebuttal45. Reliability and radiation: no one fixes a GPU in orbit52. Estimated on-orbit failure rates near 9%/yr imply ~20% of accelerators dark within 26 months; five-year satellite lives clash with 1–2-year terrestrial refresh cycles; Google's own data shows HBM irregularities beginning at 2 krad, acceptable for inference, a real question for training13,18,42. Bandwidth: fleet laser links run 100–200 Gbps against terabit-class data-center fabrics; Suncatcher's 1.6 Tbps bench demo is the state of the art, and ground-segment egress remains a choke point, pushing near-term orbital compute toward inference and preprocessing rather than frontier training13,51. Aggregate: ABI Research estimates today's all-in orbital TCO at ~78× terrestrial while still forecasting 1.54 effective GW by 203543.

Demand-side check. The bull case does not require beating terrestrial compute everywhere, only capturing overflow demand that grids cannot serve. US AI capacity needs are projected to triple by 203143; interconnect queues in key metros run to a decade49; and Handmer's margin argument — inference tokens sell for ~100× marginal cost, so even 2×-terrestrial orbital costs preserve ~98% gross margins — reframes the question from cost parity to elasticity of supply48. The bear case answers that nuclear restarts, gas turbines and grid buildout are also racing, and a 3–5-year orbital deployment cycle competes against terrestrial capacity that can break ground tomorrow. Both arguments are about timing, not physics.

7. Space logistics, transport and maintenance: the enabling economy

If orbital compute is the skyscraper, logistics and servicing are the roads, cranes and maintenance crews, and in 2025–26 this layer produced harder milestones than the compute layer it will serve. The companion database profiles 39 organizations across six segments.

The strategic read: 2026 is to servicing what 2015 was to reusable launch, the year the reference missions flew. When MRV reaches GEO around 2027 and begins commercial repairs alongside China's demonstrated refueling capability, satellite design assumptions — single-fuel-load, disposable — begin to unwind, and with them the replacement-demand model underlying much of today's manufacturing revenue.

7.1 Rapid mobility and dynamic space operations

A distinct agility segment has crystallized around the US Space Force doctrine of dynamic space operations and maneuver without regret: repositioning across and between LEO, MEO and GEO at operational tempo rather than Hohmann patience. Orbital mechanics forces a three-way split in how firms buy that agility: fast chemical (Impulse's Helios, ~4 t from LEO to GEO in under a day, booked from 202765,66), high-efficiency thermal (Portal's solar-thermal Supernova, hours-class repositioning at far better propellant economy), and efficient-but-slow electric (Exotrail, D-Orbit, logistics-grade rather than tactical).

Two dependencies make this a Physical-AI story rather than a propulsion story. First, agility is effectively ΔV × autonomy ÷ decision latency: a ground-in-the-loop maneuver cycle takes hours an adversary's spacecraft does not grant, so terminal-phase RPO guidance and AI-accelerated low-thrust trajectory search are the binding technologies, with Starlink's 355,000 autonomous maneuvers per year as the fleet-scale existence proof79. Second, repeatable maneuver is a propellant-logistics statement: without regret only holds if tanks refill, which routes the doctrine back through the refueling economy (RAFTI, APS-R, PRM) and explains why China's SJ-25 to SJ-21 demonstration63 was read on both sides of the Pacific as a mobility milestone rather than a servicing one.

8. The orbital combat economy: a doctrinal net assessment

The rapid-mobility segment is at bottom a military-economic phenomenon, and its logic reads directly off the maneuver prices. This section treats those prices as a combat-economics model: an open-source, doctrinal net assessment in the tradition of published counterspace surveys111,112, built from two-body orbital mechanics and the documented 2025–26 programs, not from any operational source. The organizing insight is that orbital engagement is logistics-limited. Unlike terrestrial forces already in contact, an orbital asset must spend finite, largely non-renewable ΔV to bring an effect to bear, and that wallet does not refill without a depot. Three consequences follow — ΔV is ammunition, orbital position is terrain, and decision latency is tempo — and together they make the outcome of most engagements legible before they begin. Engagements end not in destruction but at Winchester: whoever can impose an effect for fewer m/s, more times, prevails.

Four laws organize the assessment. First, position is economy. The ΔV to put one engagement on a GEO asset spans 28× purely by starting orbit: 4,787 m/s climbing from a 51.6° LEO, 789 from an equatorial MEO perch, 170 from co-orbital GEO, while mid-fight plane changes stay unaffordable (a 10° tilt at LEO is ~1,340 m/s). The fight is decided by pre-positioning. Second, the high ground is coverage, not altitude. A co-orbital GEO picket can hold only ±35° of the belt at risk within a day for a 200 m/s reserve, so covering the full belt takes roughly six pickets; a single MEO vehicle laps the entire GEO belt by Keplerian drift about once a day and dives for 789 m/s when it chooses, while GEO defenders cannot cheaply climb to punish it. That is a ~6:1 coverage economy and the quantified case for the emerging MEO layer. Third, autonomy is priced in propellant. Evasion ΔV scales as 1/(warning − latency): against an autonomous defender, offense pays 30–450× the defender's ~1.5 m/s dodge, but the relation has a cliff. Once a decision loop is longer than the warning window, the defender cannot react and pays with the spacecraft. Fourth, refueling changes the exponent. A 1,500 m/s wallet goes Winchester after ~8 engagement cycles; an SJ-style +404 m/s refill every couple of cycles turns attrition into indefinite sustainment. RAFTI/PRM standardization and the SJ-25 to SJ-21 demonstration63 are therefore strategic acts, not maintenance.

The same model, with vehicles, propellants and full transfer solutions, runs in the maneuver console below.

Stability: the same equations, two readings. Whether this balance is dangerous is a question on which serious analysts divide, and the division is about weighting rather than facts. Those who read the economics as destabilizing note that cheap agility, autonomy and the inherent ambiguity of rendezvous — an inspector and an attacker fly identical approaches until the final seconds — compress decision time toward the reflexive, and that a fuelled maneuvering asset is a use-it-or-lose-it piece that degrades if it waits, rewarding the first mover. The historical rhyme is the pre-1914 mobilization timetable. Those who read the same facts as stabilizing counter that the dominant defensive move is simply to dodge — reversible, non-kinetic, harming no one and creating no debris — and that better space situational awareness widens the warning window, which by the figure above directly raises the price of surprise.

9. Planetary operations

Roughly 40 robotic spacecraft currently operate beyond Earth orbit. Every one is a Physical-AI datapoint: light-lag makes autonomy mandatory, and the 2025–26 record shows both its progress and its price.

Mars. Perseverance and Curiosity remain active, with 30+ sealed samples including the Cheyava Falls potential-biosignature core. Mars Sample Return was effectively terminated when Senate appropriators declined to fund the existing program in January 2026, leaving those samples on Mars indefinitely85; China's Tianwen-3 is in spacecraft construction for a ~2028 launch and ~2031 return, now positioned to return Mars samples first86. ESA's Rosalind Franklin has NASA contributions formally in implementation for a 2028 launch, and ESCAPADE arrives in 2027.

The Moon. The 2025 commercial scorecard was one full success (Firefly's Blue Ghost 1, 14 days of surface ops80), one tip-over (IM-2 at the south pole81), and one crash (ispace's Resilience, laser-rangefinder failure82), which says landing autonomy remains genuinely hard. Artemis II flew crewed around the Moon 1–10 April 2026, splashing down 2.9 miles from target with dramatically improved heat-shield behavior, keeping Artemis III on the books for 202784. The pipeline into 2027 is the densest in history: Chang'e-7 targeting H2 2026 with a hopping probe for shadowed craters83, Blue Ghost 2 to the far side, IM-3 carrying JPL's CADRE autonomous rover trio, Astrobotic's Griffin-1, VIPER revived on Blue Origin's second Blue Moon MK188, and NASA's $219M/$220M lunar-terrain-vehicle awards89.

Small bodies and deep space. China's Tianwen-2 reached quasi-moon Kamoʻoalewa in July 2026 ahead of a touch-and-go sampling and ~2027 return87. Psyche executed its Mars gravity assist in May 2026 after resolving a propulsion anomaly; ESA's Hera arrives at the DART impact site in December 2026. Commercial asteroid ambitions were humbled: AstroForge's Odin was lost en route in 2025, while CASC's new Tiangong Kaiwu program laid out a state roadmap from Chang'e-8 ISRU (2029) toward asteroid extraction at 2040-scale.

10. Regional landscape and policy

United States pairs the deepest capital markets (SpaceX at ~$2T, record VC54,57) with a defense pull now strong enough to distort commercial roadmaps: GITAI's postponed servicing demo90 and Golden Dome's $25B are the same story told twice. The FCC absorbed two unprecedented compute-constellation filings while the FAA moved to streamline launch environmental reviews, even as its ~145-launch/yr licensing ceiling sits ~35× below fast-scenario needs44.

China is running an industrial policy for space compute with no Western equivalent: MIIT's systematic planning posture, a CAICT-coordinated committee spanning 100+ organizations, a Beijing innovation center for satellite AI chips and thermal systems, and a projected >¥250B domestic market by 203029,30. Beneath policy sit three operating constellation programs (Three-Body, Star-Compute, Xingshu31), the SJ-25/SJ-21 refueling first63, commercial servicing entrant Sustain Space64, and a lunar and resource roadmap synchronized with its Five-Year Plan. The gap between announced POPS and audited capability is real; the coordination advantage is realer.

Japan built the servicing sector's first genuine business result: Astroscale's first positive gross profit and ¥30.6B raise with JSAT and Hulic as strategic investors62, alongside a ¥1T Space Strategy Fund95, NTT's methodical space-DC roadmap36,37, and new entrants such as Orbital Lasers (¥3B Series A)100. Europe leads on rules and lags on hardware: the ASCEND study validated a 1 GW climate-positive orbital DC by 205038 but no demonstrator is funded, ClearSpace-1 slipped to ~202899, both 2025 reentry-capsule attempts failed71,72, and the draft EU Space Act remained stuck in Council94. Standouts: D-Orbit's €120M ESA GEO-servicing contract67 and The Exploration Company's reported >$2B valuation talks. Elsewhere: India converted its SPADEX docking success into a 2028 Chandrayaan-4 path with SSA (Digantara97) and refueling-interface startups; Korea's KASA opened its first OTV budget line96; the UAE's Madari is the Gulf's first orbital-DC entrant39; Russia's Sfera program consolidated amid funding complaints, with no visible compute or servicing initiative98.

11. Risk register

Launch-cost schedule risk (dominant). Every orbital-compute business case discounts a Starship price curve that has not yet materialized: mid-2026 cadence is 1–2 flights/quarter against models assuming thousands/yr12,44. A three-year slip in $/kg moves every breakeven right by roughly the same amount. Thermal-mass physics. Radiator specific power must improve 3–10× over flown heritage for bull-case masses; otherwise radiators, not chips, are the product19,44,45. Reliability without repair. ~9%/yr accelerator attrition and 5-yr asset lives against 1–2-yr terrestrial refresh force overbuild and freeze architectures, unless in-orbit servicing matures on schedule, which is itself the Physical-AI bet18,44. Debris and carrying capacity. Filings totaling >1M satellites enter an environment already generating 355k avoidance maneuvers/yr79; capacity is management-conditional105 and shrinking with thermospheric contraction106. A single bad conjunction cascade in a compute shell would reprice the sector's insurance and its politics simultaneously107. Bandwidth ceiling. Until multi-Tbps optical mesh and ground egress scale, orbital compute is inference-shaped; training-class interconnect remains unproven13,51. Demand rerating. If terrestrial power constraints ease or AI capex cools, orbit's overflow demand evaporates ahead of its cost curve50. Concentration and governance. One ~$2T firm spans launch, broadband and compute filings; Brookings flags the feasibility gap itself as a governance risk, with spectrum and orbital slots claimed by paper constellations46. Geopolitical bifurcation. Parallel US/China compute-and-servicing stacks with dual-use ambiguity raise both escalation risk and compliance cost for every commercial operator63.

12. Scenarios and outlook to 2035

Signposts over the next 18 months. Starcloud-2 (Oct 2026): does a Blackwell-class node run at full power, fixing Starcloud-1's radiator shortfall, and hold customers' SLAs16. Starship: cadence × published pricing; the first credible sub-$1,000/kg external quote resets every model12. MRV at GEO (~mid-2027): first commercial robotic repair, the servicing market's Falcon 9 landing moment58. Google's Planet-built prototypes (early 2027) and any published radiation/thermal telemetry, the first peer-reviewable orbital-DC data13. China: Star-Compute batch cadence toward 1,000 sats by 2030, and whether POPS claims convert to sold capacity28,30. Starship ship-to-ship propellant transfer: unlocks the heavy-cargo cislunar economy, still unflown as of Aug 2026102. FAA licensing ceiling and the EU Space Act: the regulatory throughput fast scenarios silently assume44,94.

13. Conclusions

1. The theses are one thesis. Orbital data centers, space logistics and planetary operations all reduce to the same input: machines that can work in places humans cannot. Physical AI is the space economy's labor supply, and its maturity curve, not solar flux and not chip supply, sets the sector's speed limit.

2. Hardware has started telling the truth. In eighteen months the orbital-compute question moved from could it work to at what $/kg, with what radiators, at what reliability — questions with numeric answers arriving on knowable dates. The 10⁶:1 ambition-to-hardware gap will compress from both directions.

3. The servicing economy is the safer branch of the same bet. Whether or not gigawatt compute arrives on schedule, the 2025–26 record — MRV launched, GEO refueling demonstrated, six Varda recoveries, ADRAS-J's inspection, 355k autonomous maneuvers — says orbital operations are industrializing now. Its revenue is nearer, its physics kinder, and every compute scenario needs it.

4. Watch China's coordination, not just its claims. Three constellation programs, a national committee, a refueling first, and a Five-Year-Plan-synchronized resource roadmap constitute the most complete state strategy for the orbital economy anywhere. The West's answer is capital depth and SpaceX's cost curve. The 2030 landscape depends on which compounds faster.

5. For builders and investors, the asymmetric positions are in the tax collectors, not the skyscrapers: autonomy software with flight heritage, manipulation and refueling standards (RAFTI/PRM), optical interconnect, SSA, high-cadence reentry, and thermal hardware. These are the capabilities every scenario must rent.

Companions

Both consoles are schematic: two-body impulsive kinematics, for intuition rather than mission design. Peer-reviewed treatments are cited above (Kessler 2010107; MIT MOCAT105).

References

1. 1 Novaspace, “Global Space Economy Reaches $626 Billion” (Jan 2026) — https://nova.space/press-release/global-space-economy-reaches-626-billion-mark…
2. 2 Via Satellite, “Novaspace Report Predicts Global Space Economy Could Top $1 Trillion by 2034” (Jan 30, 2026) — https://www.satellitetoday.com/finance/2026/01/30/novaspace-report-predicts-gl…
3. 3 World Economic Forum / McKinsey, “Space: The $1.8 Trillion Opportunity” (Apr 2024) — https://www3.weforum.org/docs/WEF_Space_2024.pdf
4. 4 Space Foundation, The Space Report 2025 Q2 — https://www.spacefoundation.org/2025/07/22/the-space-report-2025-q2/
5. 5 Novaspace, “Global Space Spending Reaches $137B, Marking a Defense-Led Era” — https://nova.space/press-release/global-space-spending-reaches-137b-marking-a-…
6. 6 SpacePolicyOnline, “Adding Reconciliation, Space Force Budget Tops $40 Billion in FY2026” — https://spacepolicyonline.com/news/adding-reconciliation-space-force-budget-to…
7. 7 Space.com, “Congress passes $24.4 billion budget for NASA” (Jan 2026) — https://www.space.com/space-exploration/in-rejection-of-trumps-deep-cuts-congr…
8. 8 Space.com, “Europe passes record-breaking space budget” (ESA Bremen Ministerial, Nov 2025) — https://www.space.com/space-exploration/europe-passes-record-breaking-space-bu…
9. 9 J. McDowell, “Space Activities in 2025” (Feb 4, 2026) — https://planet4589.org/space/papers/space25.pdf
10. 10 Aviation Week, “Global Orbital Launch Rate Jumped 25% in 2025” — https://aviationweek.com/space/launch-vehicles-propulsion/spaceops-global-orbi…
11. 11 Orbital Radar, “Launch Cost Trends” (2026) — https://orbitalradar.com/space-economy/launch-cost-trends
12. 12 SpaceOrbitals, “Starship V3 Cost per kg” (2026) — https://spaceorbitals.com/articles/starship-v3-cost-per-kg-2026/
13. 13 Google Research, “Towards a future space-based, highly scalable AI infrastructure system design” (Project Suncatcher paper, Nov 2025) — https://services.google.com/fh/files/misc/suncatcher_paper.pdf
14. 14 Google, “Project Suncatcher” announcement (Nov 4, 2025) — https://blog.google/innovation-and-ai/technology/research/google-project-sunca…
15. 15 Starcloud, “Why We Should Train AI in Space” (whitepaper) — https://starcloudinc.github.io/wp.pdf
16. 16 GeekWire, “Starcloud hits $1.1B valuation” (Mar 2026) — https://www.geekwire.com/2026/orbital-ai-seattle-area-startup-starcloud-hits-1…
17. 17 CNBC, “Nvidia-backed Starcloud trains first AI model in space” (Dec 10, 2025) — https://www.cnbc.com/2025/12/10/nvidia-backed-starcloud-trains-first-ai-model-…
18. 18 IEEE Spectrum, “Orbital Data Center Hype” (Jul 2026) — https://spectrum.ieee.org/orbital-data-center-hype
19. 19 IEEE Spectrum, “The Thermal Problem with Orbital Data Centers” (Jun 11, 2026) — https://spectrum.ieee.org/orbital-data-centers-heat
20. 20 Data Center Dynamics, “SpaceX files for million-satellite orbital AI data center megaconstellation” (Jan 2026) — https://www.datacenterdynamics.com/en/news/spacex-files-for-million-satellite-…
21. 21 Yahoo Finance, “SpaceX reveals its first orbital data center — AI1” (Jun 9, 2026) — https://finance.yahoo.com/sectors/technology/article/spacex-reveals-its-first-…
22. 22 Broadband Breakfast, “Blue Origin submits satellite application for space-based data centers” (Mar 2026) — https://broadbandbreakfast.com/blue-origin-submits-satellite-application-for-s…
23. 23 CNBC, “Bezos on Blue Origin space data centers” (May 20, 2026) — https://www.cnbc.com/2026/05/20/bezos-blue-origin-space-data-centers.html
24. 24 Axiom Space, Orbital Data Center program — https://www.axiomspace.com/orbital-data-center
25. 25 Introl, “First Orbital Data Center Nodes Reach Space” (Jan 2026) — https://introl.com/blog/orbital-data-center-nodes-launch-space-computing-infra…
26. 26 Kepler Communications, “Kepler deploys first space-based scalable cloud infrastructure powered by NVIDIA” (Mar 16, 2026) — https://kepler.space/kepler-deploys-first-space-based-scalable-cloud-infrastru…
27. 27 Via Satellite, “Aetherflux rebrands as Cowboy Space, expanding to in-orbit compute and launch” (May 11, 2026) — https://www.satellitetoday.com/finance/2026/05/11/aetherflux-rebrands-as-cowbo…
28. 28 SpaceNews, “China launches first of 2,800 satellites for AI space computing constellation” (May 2025) — https://spacenews.com/china-launches-first-of-2800-satellites-for-ai-space-com…
29. 29 Xinhua, space computing industry coverage incl. MIIT posture and CAICT committee (Apr 2026) — https://www.news.cn/tech/20260408/b0b6a0ccd4f14902b7c31383215ffdf4/c.html
30. 30 Xinhua, China space computing market >250B yuan by 2030 (Apr 25, 2026) — https://english.news.cn/20260425/b4368923827e4988af1bd030dd01d9cc/c.html
31. 31 ITHome, “Shanghai Xingshu Plan unveiled at WAIC” (Jul 18, 2026) — https://www.ithome.com/0/978/530.htm
32. 32 Zhejiang News, Three-Body Computing Constellation first launch (May 14, 2025) — https://zjnews.zjol.com.cn/yc/qmt/202505/t20250514_30994148.shtml
33. 33 Quanzhou Evening News, Three-Body Constellation expansion launch (Jul 24, 2026) — https://www.qzwb.com/gb/content/2026-07/24/content_9243456.htm
34. 34 Factories in Space, Lonestar Data Holdings profile — https://www.factoriesinspace.com/lonestar
35. 35 StorageNewsletter, “Sidus Space expands agreement with Lonestar for StarVault payload” (Apr 22, 2026) — https://www.storagenewsletter.com/2026/04/22/sidus-space-expands-existing-agre…
36. 36 NTT Group, aerospace business information (Space Integrated Computing Network) — https://group.ntt/en/aerospace/information/
37. 37 Space Compass, news releases 2025–2026 — https://space-compass.com/news/
38. 38 Thales Alenia Space, ASCEND feasibility study results — https://www.thalesaleniaspace.com/en/press-releases/thales-alenia-space-reveal…
39. 39 Data Center Dynamics, “UAE’s Madari Space to pilot sending data centers into space in 2026” — https://www.datacenterdynamics.com/en/news/uaes-madari-space-to-pilot-sending-…
40. 40 SpaceNews, “Relativity names Eric Schmidt as CEO” (Mar 2025) — https://spacenews.com/relativity-names-eric-schmidt-as-ceo-as-it-updates-terra…
41. 41 GeekWire, “Sophia Space raises $10M for space computing network” (Feb 2026) — https://www.geekwire.com/2026/sophia-space-10m-space-computing-network/
42. 42 JLL, “Data Centers in Space” research report (2026) — https://www.jll.com/content/dam/jllcom/en/global/documents/reports/research-re…
43. 43 ABI Research, “Data Centers in Space” (2026) — https://www.abiresearch.com/blog/data-centers-in-space
44. 44 Forethought, “Will We Really Put Data Centers in Space?” (May 22, 2026) — https://www.forethought.org/research/will-we-really-put-data-centers-in-space
45. 45 A. Nanjangud, “Space Data Centers, Part 1” (2025, rev. Jul 2026) — https://angadh.com/space-data-centers-1
46. 46 Brookings, “Orbital data centers’ feasibility gap is a governance risk” (Jun 25, 2026) — https://www.brookings.edu/articles/orbital-data-centers-feasibility-gap-is-a-g…
47. 47 Aravolta, “Datacenters in Space” (Jan 23, 2026) — https://www.aravolta.com/blog/datacenters-in-space
48. 48 C. Handmer, “I Guess We’re Doing Moon Factories Now” (Feb 10, 2026) — https://caseyhandmer.wordpress.com/2026/02/10/i-guess-were-doing-moon-factorie…
49. 49 Axis Intelligence, “AI Data Center Cost per MW” (2026) — https://axis-intelligence.com/ai-data-center-cost-per-mw/
50. 50 Futurum Group, “AI Capex 2026: The $690B Infrastructure Sprint” — https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint…
51. 51 Advanced Television, “SpaceX reveals Starlink laser capacity” (Feb 2024) — https://www.advanced-television.com/2024/02/02/spacex-reveals-starlink-laser-c…
52. 52 TechRadar Pro, “Sam Altman says plans for data centers in space are ridiculous” (Feb 2026) — https://www.techradar.com/pro/sam-altman-says-plans-for-data-centers-in-space-…
53. 53 Benzinga, “Jensen Huang: orbital datacenters have poor economics right now; cooling is the bottleneck” (Feb 2026) — https://www.benzinga.com/news/space/26/02/50876881/jensen-huang-thinks-orbital…
54. 54 Space Capital, Space IQ quarterly investment reports (2025–2026) — https://www.spacecapital.com/space-iq
55. 55 Reuters (via KFGO), “Space investment in first quarter hits record as SpaceX IPO buzz lifts sector” (Apr 21, 2026) — https://kfgo.com/2026/04/21/space-investment-in-first-quarter-hits-record-as-s…
56. 56 Seraphim Space, SpaceTech sector newsletter Q2 2026 — https://www.investegate.co.uk/announcement/rns/seraphim-space-investment-trust…
57. 57 Yahoo Finance, “SpaceX IPO by the numbers” (Jun 2026) — https://finance.yahoo.com/markets/article/spacex-ipo-reveals-one-stunning-fact…
58. 58 DARPA, “Robotic Servicing of Geosynchronous Satellites lifts off” (Jul 21, 2026) — https://www.darpa.mil/news/2026/robotic-servicing-of-geosynchronous-satellites…
59. 59 Northrop Grumman, MRV launch release (Jul 2026) — https://news.northropgrumman.com/launch/northrop-grummans-mission-robotic-vehi…
60. 60 Space.com, “Two private satellites undock after pioneering life-extension mission” (Apr 2025) — https://www.space.com/space-exploration/satellites/2-private-satellites-undock…
61. 61 Astroscale, “ADRAS-J mission completes operations” (Mar 2026) — https://www.astroscale.com/en/news/astroscales-adras-j-mission-completes-opera…
62. 62 Space Connect (JP), Astroscale FY2026 results analysis — https://space-connect.jp/astroscale-2026/
63. 63 SCMP, “China’s Shijian satellite pair appears to dock in orbit in historic refuelling mission” (2025) — https://www.scmp.com/news/china/science/article/3318050/chinas-shijian-satelli…
64. 64 NASASpaceflight, China roundup April 2026 (Sustain Space Xiyuan-0 demo) — https://www.nasaspaceflight.com/2026/04/china-roundup-202604/
65. 65 Business Wire, “Impulse Space secures $300M Series C” (Jun 2025) — https://www.businesswire.com/news/home/20250603503683/en/Impulse-Space-Secures…
66. 66 Space.com, “Impulse Space raises $500M” (Jun 2026) — https://www.space.com/space-exploration/launches-spacecraft/impulse-spacex-500…
67. 67 European Spaceflight, “D-Orbit wins €119M ESA contract for satellite life extension (RISE)” (Oct 2024) — https://europeanspaceflight.com/d-orbit-wins-119m-euro-esa-contract-for-satell…
68. 68 GeekWire, “Starfish Space to attempt docking with Australian satellite” (2026) — https://www.geekwire.com/2026/starfish-space-australian-satellite-docking/
69. 69 PR Newswire, “Varda announces $187M Series C” (Jul 2025) — https://www.prnewswire.com/news-releases/varda-announces-187-million-in-series…
70. 70 PR Newswire, “Varda W-6 reentry validates autonomous navigation and advanced TPS” (May 2026) — https://www.prnewswire.com/news-releases/varda-space-industries-successfully-r…
71. 71 European Spaceflight, “The Exploration Company declares Mission Possible a partial success” (Jun 2025) — https://europeanspaceflight.com/the-exploration-company-declares-mission-possi…
72. 72 Atmos Space Cargo, “Phoenix 2: first of seven missions to launch 2026” — https://atmos-space-cargo.com/milestones/phoenix-2-first-of-seven-missions-to-…
73. 73 Spaceflight Now, “Sierra Space’s Dream Chaser debut delayed again; no longer docking to station” (Sep 2025) — https://spaceflightnow.com/2025/09/26/sierra-spaces-dream-chaser-debut-mission…
74. 74 Grand View Research, Satellite Servicing Market report (2025) — https://www.grandviewresearch.com/industry-analysis/satellite-servicing-market…
75. 75 Allied Market Research, In-Space Manufacturing, Servicing and Transportation Market ($135.3B by 2040) — https://www.alliedmarketresearch.com/press-release/in-space-manufacturing-serv…
76. 76 US GAO, “In-Space Servicing, Assembly and Manufacturing” GAO-25-107555 (Jul 2025) — https://www.gao.gov/products/gao-25-107555
77. 77 MarketsandMarkets, Physical AI Market ($15.24B by 2032) — https://www.marketsandmarkets.com/PressReleases/physical-ai.asp
78. 78 Morgan Stanley, “Humanoid robot market: $5 trillion by 2050” — https://www.morganstanley.com/insights/articles/humanoid-robot-market-5-trilli…
79. 79 Starpath, “Starlink satellites now dodging collisions almost weekly as maneuvers triple in a year” (Jul 15, 2026) — https://starpath.global/news/starlink-satellites-now-dodging-collisions-almost…
80. 80 Spaceflight Now, “Firefly’s Blue Ghost Mission 1 concludes successful 14-day mission on the Moon” (Mar 2025) — https://spaceflightnow.com/2025/03/18/firefly-aerospaces-blue-ghost-mission-1-…
81. 81 Spaceflight Now, “Intuitive Machines IM-2 mission ends with lander on its side” (Mar 2025) — https://spaceflightnow.com/2025/03/07/intuitive-machines-im-2-mission-ends-wit…
82. 82 SpaceNews, “Laser rangefinder problems blamed for second ispace lunar lander crash” (2025) — https://spacenews.com/laser-rangefinder-problems-blamed-for-second-ispace-luna…
83. 83 SpaceNews, “China’s Chang’e-7 arrives at spaceport for lunar south pole mission” (Apr 2026) — https://spacenews.com/chinas-change-7-arrives-at-spaceport-for-lunar-south-pol…
84. 84 NASA, “NASA on track for future missions with initial Artemis II assessments” (Apr 2026) — https://www.nasa.gov/missions/nasa-on-track-for-future-missions-with-initial-a…
85. 85 Live Science, “NASA’s Mars Sample Return is dead” (Jan 2026) — https://www.livescience.com/space/mars/nasas-mars-sample-return-is-dead-leavin…
86. 86 SpaceNews, “China’s Tianwen-3 Mars sample return moves into spacecraft construction phase” (Apr 2026) — https://spacenews.com/chinas-tianwen-3-mars-sample-return-mission-moves-into-s…
87. 87 Forbes, “A Chinese spacecraft just arrived at Earth’s quasi-moon” (Tianwen-2, Jul 4, 2026) — https://www.forbes.com/sites/jamiecartereurope/2026/07/04/a-chinese-spacecraft…
88. 88 SatNews, “NASA solidifies lunar strategy with Blue Origin VIPER delivery task order” (Dec 2025) — https://satnews.com/2025/12/17/nasa-solidifies-lunar-strategy-with-blue-origin…
89. 89 Payload, “NASA awards LTV lunar transport contracts” (May 2026) — https://payloadspace.com/nasa-awards-ltv-lunar-transport-contracts/
90. 90 Via Satellite, “GITAI delays robotic servicing mission to prioritize SBI work” (Jun 17, 2026) — https://www.satellitetoday.com/technology/2026/06/17/gitai-delays-robotic-serv…
91. 91 MDA Space, “$1B contract for Canadarm3 flight system” (Jun 2024) — https://mda.space/article/mda-space-awarded-1b-contract-to-design-and-deliver-…
92. 92 NVIDIA, “Space Computing” platform announcement (Mar 16, 2026) — https://nvidianews.nvidia.com/news/space-computing
93. 93 Ubotica, “One year of CogniSAT-6: pioneering live Earth intelligence” (2025) — https://ubotica.com/one-year-of-cognisat-6-pioneering-live-earth-intelligence/
94. 94 US Office of Space Commerce, “EU Space Act update” (May 2026) — https://space.commerce.gov/eu-space-act-update-may-2026/
95. 95 JAXA, Space Strategy Fund portal — https://fund.jaxa.jp/
96. 96 NewsSpace (KR), KASA 2026 R&D plan — https://www.newsspace.kr/news/article.html?no=11638
97. 97 Payload, “Digantara raises $50M Series B” (Dec 2025) — https://payloadspace.com/digantara-raises-50m-series-b/
98. 98 ComNews (RU), “Sfera absorbed into new national space project” (Jan 2025) — https://www.comnews.ru/content/237436/2025-01-29/2025-w05/1007/sfera-priobrela…
99. 99 ESA, ClearSpace-1 mission page — https://www.esa.int/Space_Safety/ClearSpace-1
100. 100 Orbital Lasers, Series A announcement (Mar 19, 2026) — https://www.orbitallasers.com/
101. 101 MarketsandMarkets, On-Orbit Satellite Servicing Market ($5.1B by 2030) — https://www.marketsandmarkets.com/PressReleases/on-orbit-satellite-servicing.a…
102. 102 Wikipedia, Starship Propellant Transfer Demonstration — https://en.wikipedia.org/wiki/Starship_Propellant_Transfer_Demonstration
103. 103 LeoLabs, “Record bookings in 2025” — https://leolabs.space/press/leolabs-achieves-record-bookings-in-2025/
104. 104 Astroscale, ADRAS-J2 mission page (JAXA CRD2 Phase II) — https://www.astroscale.com/ja/missions/adras-j2
105. 105 MIT ARC Lab / JSR, “Carrying Capacity of Low Earth Orbit Computed Using Source-Sink Models” — https://arxiv.org/abs/2206.05345
106. 106 MIT News, “Climate change will reduce the number of satellites that can safely orbit in space” (Mar 2025) — https://news.mit.edu/2025/study-climate-change-will-reduce-number-satellites-s…
107. 107 D.J. Kessler, “The Kessler Syndrome: Implications to Future Space Operations” (2010) — https://aquarid.physics.uwo.ca/kessler/Kessler%20Syndrome-AAS%20Paper.pdf
108. 108 Data Center Dynamics, “Aetherflux orbital data center to be operational by Q1 2027” (Dec 2025) — https://www.datacenterdynamics.com/en/news/aetherflux-orbital-data-center-to-b…
109. 109 True Anomaly — Jackal, Mosaic, VICTUS HAZE flight operations, $650M raised (accessed Aug 2026) — https://www.trueanomaly.space/
110. 110 Blue Origin — Blue Ring multi-orbit mobility platform (accessed Aug 2026) — https://www.blueorigin.com/blue-ring
111. 111 Secure World Foundation, “Global Counterspace Capabilities: An Open Source Assessment” (2025 ed.) — https://swfound.org/counterspace/
112. 112 CSIS Aerospace Security Project, “Space Threat Assessment 2025” — https://www.csis.org/analysis/space-threat-assessment-2025

This review synthesizes public reporting and company statements as of 5 August 2026. It is not investment advice; private-company figures are as reported by the cited outlets and valuations are indicative. Forecast figures are the cited analysts', not the author's.