Orbital Computation · 2026-06-16

Now I have enough material. Let me write the full report.

---

🛰️ Orbital Computation — 2026-06-16

🛰️ SpaceX AI1: 70-Meter Wingspan, D3 Chip, and the Gigasat Factory That Must Deliver
💰 SpaceX Closes $2.1T IPO on $2.17B/Month Terrestrial GPU Contracts — Orbital Promise Deferred
⚖️ FCC Waives Amazon Leo's July 30 Deadline but Imposes 20-Month Spectrum Demotion
🇨🇳 Qianfan Cuts Satellite Manufacturing Cost 96% to $1.47M Per Unit, Eyes 100+ Launches by August
🔥 IEEE Spectrum: One GPU Chip Needs 1.4m² Radiator — The Thermodynamic Ceiling Over Every Orbital Data Center
🔐 SEALSQ/WISeKey Quantum Spatial Orbital Cloud Schedules Q4 2026 SpaceX Launch for Post-Quantum Security in LEO

---

🛰️ SpaceX AI1: 70-Meter Wingspan, D3 Chip, and the Gigasat Factory That Must Deliver

SpaceX unveiled the AI1 satellite design on June 8, 2026, three days before pricing the largest IPO in history — the timing deliberate, the hardware impressive, the economics unresolved. The first-generation design is 20 meters tall and spans 70 meters deployed, wider than a Boeing 747-8, a form factor driven entirely by two competing constraints: solar collection area and thermal rejection surface, not aerodynamics. The compute payload runs 120 kW sustained and peaks at 150 kW at a density of 70 kW per ton, operating at approximately 600 km altitude.

At the center of the design is the D3 chip, an in-house SpaceX processor descending through a lineage that runs from AI5 (Tesla FSD) through AI6 (Cybercab and Optimus Gen 3) to a radiation-tolerant orbital variant. Carthage Electronics' technical breakdown characterizes D3's target as competitive with edge compute and orders of magnitude beyond traditional radiation-hardened processors — which run $5,000–$50,000 per chip against commercial foundry economics. If D3 performs in the Van Allen radiation environment, it makes a niche semiconductor supply chain economically obsolete for LEO applications. That is a structural disruption to a $2–4B annual market embedded in satellites, medical devices, and defense electronics.

AI1 uses a liquid radiator oriented knife-edge to the sun, rated at 1,400 watts per square meter, radiating from both faces while minimizing solar absorption. The compute module is interchangeable — SpaceX can swap chip generations without redesigning the structural bus — betting that orbital silicon will improve faster than the platform envelope requires. Two AI1 prototype satellites are targeted for early 2027.

The manufacturing infrastructure to support this is the Gigasat factory in Bastrop, Texas — 11 million square feet unveiled by SpaceX alongside the AI1 design. The connection to Terafab, the Tesla-SpaceX joint chip fabrication initiative announced March 2026 in Austin and targeting over one terawatt of AI compute capacity per year, closes the vertical integration loop: if both programs execute, SpaceX controls fabs, chip design, launch, orbital manufacturing, and network infrastructure simultaneously. No company has controlled this complete stack before. Whether it closes economically is the question SpaceX's own S-1 filing on May 20, 2026 explicitly declined to answer, acknowledging that orbital AI compute "may not be commercially viable." Musk unveiled AI1 nineteen days after that sentence was filed with the SEC.

Sources:

---

💰 SpaceX Closes $2.1T IPO on $2.17B/Month Terrestrial GPU Contracts — Orbital Promise Deferred

SpaceX debuted on Nasdaq on June 12, 2026 under ticker SPCX, priced at $135 per share, and closed at $161 — a 19% first-day gain that pushed market capitalization above $2.1 trillion. By the close of June 15, shares had rallied an additional 20%. The $75 billion raised is the largest IPO in recorded history. But the architecture of the valuation is not a rocket company — it is an AI infrastructure company with a launch monopoly as its durable moat.

In the two weeks preceding the IPO, SpaceX disclosed contracts that reframe its revenue profile. Google agreed to pay $920 million per month to rent 110,000 Nvidia GPUs, CPUs, and related memory from October 2026 through June 2029 — approximately $33 billion in committed ground-based compute revenue. Anthropic signed a separate agreement valued at $1.25 billion per month. The combined $2.17 billion per month is terrestrial GPU rental, not orbital inference. SpaceX is running a Stargate-equivalent GPU cluster on the ground and booking hyperscaler contracts against it while AI1 moves through prototype validation.

The asymmetry between the two businesses is load-bearing to the investment thesis. Ars Technica noted that the $2.9 billion NASA Human Landing System contract from 2021 — SpaceX's largest prior government award — is now dwarfed by a single month of AI compute revenue. Reuters confirmed SpaceX is responsible for more than four-fifths of orbital mass launched over the past three years, a structural position that makes its GPU rental sticky: customers who need both compute access and future launch capacity have a single counterparty with no realistic substitute for either service.

On the IPO day itself, SpaceX simultaneously launched its 1,500th Starlink satellite of 2026. Musk projects $1 trillion in revenue by 2030. The logical path from $2.17B/month terrestrial contracts to $1T/year runs through AI1 actually launching, thermally validating, and achieving unit economics that close against terrestrial alternatives. The IPO securitizes the gap between the current terrestrial GPU business and the future orbital compute business, with shareholders absorbing the thermodynamics risk in between.

Sources:

---

⚖️ FCC Waives Amazon Leo's July 30 Deadline but Imposes 20-Month Spectrum Demotion

Amazon Leo holds 331 of the 3,276 production satellites its FCC license requires — approximately 10% of the half-constellation milestone the commission set for July 30, 2026. Failure to reach 1,616 satellites by that date would have triggered a partial loss of authorization. Amazon applied for a waiver, citing its multibillion-dollar investment and the public interest value of a competitive second LEO broadband provider. The FCC granted it. But the relief architecture reveals the commission's priorities more clearly than the headline does.

Per Ars Technica's coverage, satellites launched after the July 31 deadline will lose "priority status" — Amazon Leo must demonstrate that every post-deadline spacecraft does not interfere with other operators. In operational terms, that phrase means SpaceX Starlink. This interference-coordination requirement persists for 20 months, until March 30, 2028, or until Amazon deploys 50% of its constellation, whichever comes first. Engadget confirmed the waiver's conditional structure: relief without demotion was not on offer.

The governance logic has implications beyond Amazon's immediate situation. The FCC is encoding deployment velocity as a determinant of spectrum rights: fast deployers retain priority access, slow deployers coordinate with incumbents. SpaceX, having launched the dominant share of LEO mass over the past three years, is the de facto incumbent against which every future operator must coordinate. The Amazon waiver decision is a precedent. Every company currently holding a large constellation FCC filing — Starcloud's 88,000-satellite proposal, QSOC's 100-unit roadmap — will face the same spectrum architecture if they miss milestones.

Amazon has applied for regulatory approval in Kenya under subsidiary Kuiper Kenya Limited, seeking the exact Network Facilities Provider Tier 2 license that Starlink uses — signaling that global market access remains the goal even as the constellation buildout slips. The tension between global market ambition and domestic deployment pace defines Amazon Leo's structural position: its consumer promise is global broadband, but its regulatory standing depends on deploying hardware in orbit faster than a company that also controls most launch infrastructure. July 31 is the enforcement bellwether for the next generation of orbital spectrum governance.

Sources:

---

🇨🇳 Qianfan Cuts Satellite Manufacturing Cost 96% to $1.47M Per Unit, Eyes 100+ Launches by August

China's Qianfan constellation reached 200 satellites in orbit following a Long March-8 launch on June 5, 2026, while releasing manufacturing cost figures that upend assumptions about the economics of LEO buildout. Per-unit production cost has fallen over 96% from early-development levels, arriving at approximately 10 million yuan ($1.47M) per satellite, per pricing disclosed by Microsatellite Academy director Hu Haiying. Starlink V2 Mini units are estimated at $250,000–$500,000 at SpaceX's volumes, meaning Qianfan remains 3–6x more expensive — a gap that is less interesting than the trajectory.

A 96% cost reduction in a nascent program is consistent with Wright's Law manufacturing curves that drove Falcon 9 launch economics from $10,000/kg to under $1,000/kg. If Qianfan applies the same maturation logic, the next decade of volume production implies $750,000–$900,000 per unit at 10x current production scale — below the Starlink V2 Mini range. At that price, adding AI inference compute payloads becomes a marginal-cost decision rather than a category-defining investment. Wikipedia's Qianfan entry notes the program's ambitions "beyond basic internet," a phrase with specific meaning in a competitive context where SpaceX is fielding orbital data centers.

The operational concern — raised by South China Morning Post analysis — is deployment velocity. At the two-year pace that produced 200 satellites, Qianfan's 15,000-satellite target by 2030 would require a 15x acceleration that no current launch manifest supports. The announced countermove: the program plans to launch more than 100 additional satellites before the end of July 2026. That is a six-week, multi-launch acceleration from historical pace, implying that the deployment cadence concern triggered an internal response and that committed launch vehicle availability exists. If the July acceleration lands, the pace objection dissolves as a primary constraint.

The strategic reading is that China is competing on cost curve rather than capability frontier. SpaceX's AI1 strategy is capability-first: dense compute, novel chip design, thermal innovation, vertical fab integration. Qianfan's strategy is position-first: accumulate orbital slots, regulatory presence, and spectrum rights cheaply, then add capability incrementally as the manufacturing platform matures. The critical crossover: if Qianfan reaches $500K per unit with compute payloads before AI1 demonstrates viable orbital inference economics, the competitive dynamic inverts from capability race to cost-structure race — one that favors Chinese manufacturing scale.

Sources:

---

🔥 IEEE Spectrum: One GPU Chip Needs 1.4m² Radiator — The Thermodynamic Ceiling Over Every Orbital Data Center

IEEE Spectrum published a detailed thermal analysis on June 15, 2026 that places first-principles constraints on orbital data center architecture with a precision the marketing materials around AI1 do not. The central finding: a single commodity GPU chip operating at typical inference power draw requires 1.4 square meters of radiator surface area — derived from the Stefan-Boltzmann law, the operating temperature range GPU silicon can tolerate, and the equilibrium temperature of a radiator in LEO. This number is not negotiable through engineering cleverness; it is a consequence of physics.

The scaling implications are immediate. A 32-GPU AI rack (four H100 server boards) requires 44.8 square meters of dedicated radiator. Tom's Hardware's analysis compared this against the International Space Station: its entire Environmental Control and Life Support thermal system covers 422 square meters, rejects roughly 70 kW of waste heat — about half the thermal output of a single 140 kW GB300 rack — and cost approximately $500 million to build and deploy, per SemiAnalysis. The ISS radiator was not designed for GPU workloads. Scaling it for a meaningful compute payload implies either radically cheaper radiator manufacturing or a design architecture that does not resemble the ISS.

The degradation problem compounds the sizing constraint. IEEE Spectrum notes that thermal coating effectiveness deteriorates from ultraviolet exposure and atomic oxygen erosion at LEO altitudes — operators must design a minimum 40% excess radiator margin to preserve thermal function across a multi-year operational lifetime, increasing launch mass and drag simultaneously. SpaceX's AI1 addresses this with a liquid radiator at 1,400 watts per square meter, oriented knife-edge to the sun, maximizing rejection from both faces. Whether that figure holds after three years of LEO thermal cycling is a materials question that ground testing cannot definitively resolve before flight.

SemiAnalysis calculated that launch costs alone currently make orbital compute approximately four times more expensive than terrestrial equivalents — and this figure excludes thermal infrastructure mass overhead. The combined picture sets a cost-competitive crossover at early-2030s Starship economics, not 2027. SpaceX's S-1 disclosure that orbital AI compute "may not be commercially viable" reads, in this context, as a thermodynamics acknowledgment rather than hedging language. AI1's thermal design is the critical path item, and it will not be validated until the 2027 prototype flights.

Sources:

---

🔐 SEALSQ/WISeKey Quantum Spatial Orbital Cloud Schedules Q4 2026 SpaceX Launch for Post-Quantum Security in LEO

At the opposite end of the orbital compute scale from SpaceX's mass-market AI1 constellation, SEALSQ and WISeKey announced on June 12, 2026 meaningful progress on the Quantum Spatial Orbital Cloud (QSOC), a specialized orbital platform targeting post-quantum-secure services delivered directly from LEO. The first QSOC satellite is scheduled for a SpaceX launch in Q4 2026, with a long-term roadmap of up to 100 satellites — a fraction of AI1's ambitions but targeting a compliance-driven market where volume is not the competitive variable.

QSOC's architecture combines four service layers: edge computing, AI, post-quantum cryptography (PQC), and secure satellite communications. The commercial model is vertically split: SEALSQ operates the cloud services layer, managing customer relationships, digital identity issuance, quantum randomness generation, and AI inference at altitude, while WISeKey provides underlying semiconductor and key-management infrastructure through its identity chain. The architecture delivers a root-of-trust that is physically inaccessible to ground-based threat actors — an advantage for critical infrastructure applications that is not replicable through software architecture alone.

The timing corresponds to a specific compliance pressure: NIST's post-quantum cryptography standards, finalized in 2024, create migration obligations for critical infrastructure operators on a 2028–2030 timeline depending on sector. Satellite-based PQC delivery sidesteps two supply-chain risks that terrestrial deployments face — hardware tampering at the network edge and nation-state interference with certificate authority infrastructure. A 100-satellite orbital root-of-trust is a niche capability with a well-defined government and financial services customer base, not a general-purpose compute market.

The QSOC announcement is a structural signal about how the orbital compute sector is already stratifying. SpaceX targets commodity AI inference at hyperscale for cloud majors. QSOC targets compliance-driven security for governments and institutions. Between: Starcloud's 88,000-satellite FCC filing for general-purpose orbital data centers, and the e-scooter-founded Orbital startup targeting piecewise inference revenue from 2028 with NVIDIA Space-1 GPUs. Orbital compute is not a single-winner market — it is stratifying into mass-market AI infrastructure, sovereignty-grade security, and specialized edge workloads, each with distinct cost structures, regulatory postures, and customer sets.

Sources:

---

Research Papers

Characterizing the Impact of NVFP4 Quantization for Low-Power Edge AI Deployment — Sen et al. (arXiv:2606.06527, June 2026) — Ablation study of NVFP4 precision quantization for edge-efficient neural networks, covering GPUs, Tensor Cores, FPGAs, domain-specific AI accelerators, near-memory compute, and emerging edge architectures. Directly relevant to AI1's power budget: at 120 kW sustained orbital compute, quantization decisions determine the ratio of useful inference work to thermal overhead, and NVFP4's 4-bit weight precision reduces memory traffic and compute energy simultaneously.

Operational Forecasting of Solar Energetic Particle Events and Proton Flux Using Multi-Source Solar Observations and Multi-Task Deep Learning — (arXiv:2606.14440, June 2026) — Deep learning framework combining multi-source solar imagery for real-time solar energetic particle (SEP) event prediction. SEP events produce single-event upsets in unshielded commercial silicon at LEO altitudes; accurate SEP forecasting is a prerequisite for operating D3-class commercial chips without rad-hard protection — the enabling science behind SpaceX's bet that error-correction software can substitute for $50,000 rad-hard silicon per chip.

Scalable Deep Learning Framework for Global High-Resolution Land Use Reconstruction — (arXiv:2606.11793, June 2026) — Demonstrates scalable training of land-use classification models on satellite-derived Earth observation data using foundation model embeddings. Represents the canonical periodic-retraining workload that orbital compute would service at latency and bandwidth advantages impossible from the ground — satellite sensors generating data faster than downlink capacity can move it, making onboard inference the engineering solution.

ReSET: Accurate Latency-Critical NVFP4 Reasoning via Step-Aware Temperature Scaling — (arXiv:2606.13233, June 2026) — Addresses reasoning accuracy degradation in NVFP4-quantized large reasoning models under latency-critical conditions through step-aware temperature scaling. Relevant to AI1's mission profile: orbital inference workloads will run quantized models constrained by the power envelope, and multi-step reasoning tasks compound quantization error in ways that single-token accuracy metrics do not capture.

---

Implications

The week of June 9–16, 2026 established the structural topology of the orbital compute sector in ways that will be visible in retrospect as definitional. Three dynamics now operate in parallel, and their interaction determines the outcome.

The vertical integration bet is locked in. SpaceX went public on a thesis combining launch monopoly (80%+ of orbital mass over three years), terrestrial GPU rental ($2.17B/month contracted), and future AI1 orbital capacity — controlling every layer from chip fab (Terafab) to manufacturing (Gigasat) to orbit (Starlink) to compute (AI1). The IPO converts this thesis into shareholder capital that finances the transition period. Google and Anthropic's ground-compute contracts are bridge revenue, not orbital validation: they demonstrate hyperscaler willingness to pay, while thermodynamics research proceeds on the D3 chip and thermal management system. The $2.1 trillion market cap is a bet on the vertical stack arriving before terrestrial cloud buildout saturates — a bet whose payoff date is determined by physics, not a marketing calendar.

The regulatory layer is crystallizing around deployment velocity. The FCC Amazon Leo waiver encodes a principle: deployment speed determines spectrum priority. This is a structural advantage for any operator that executes fast — currently SpaceX — and a compounding disadvantage for every competitor. Amazon Leo, at 10% of its required deployment, enters a 20-month spectrum-demotion period covering its entire critical buildout phase. The mechanism is not punitive in intent but is structurally indistinguishable from punishment: every satellite Amazon launches after July 31 requires coordination with the incumbent before operational use. The first-mover spectrum advantage is now codified in regulatory architecture, not just industry practice.

China is attacking the cost structure, not the capability frontier. Qianfan's 96% cost reduction to $1.47M per unit is a different competitive move from AI1's compute density ambitions. At $1.47M and declining, satellites become cheap enough to deploy primarily for spectrum accumulation and orbital slot positioning, with compute payloads added later as options. This is the inverse of SpaceX's strategy: capability-first at scale versus position-first at low cost. The critical crossover question is which strategy reaches viable orbital compute economics first — Qianfan's manufacturing maturation curve converging on SpaceX-competitive unit economics, or SpaceX's thermal engineering breakthrough validating AI1 before Qianfan closes the cost gap.

The thermodynamic ceiling documented by IEEE Spectrum — 1.4m² radiator per GPU, 40% degradation margin, 4x current cost premium — is shared physics. Every participant in orbital compute faces it: SpaceX, Amazon, Qianfan, SEALSQ, Starcloud. The operator who solves radiator mass, coating degradation, and unit-cost simultaneously, at production scale, captures supplier economics on every subsequent deployment. The Q4 2026 QSOC launch, early 2027 AI1 prototypes, and the July 2026 Qianfan acceleration are three independent experiments in the same physics. Within 12 months, all three will have returned data.

---

HEURISTICS

`yaml heuristics: - id: vertical-stack-before-orbital-economics domain: [orbital-compute, space-infrastructure, investment-thesis] when: > A company controls launch, chip design, satellite manufacturing, and compute contracts simultaneously. Orbital compute economics are not yet proven at scale. Terrestrial GPU rental provides bridge revenue. IPO capital securitizes the gap between current ground business and future orbital business. prefer: > Evaluate as a vertical infrastructure owner, not a compute provider. Track four independent success conditions, all required: (1) D3 chip survives Van Allen radiation 2+ years without rad-hard shielding at commercial-silicon performance levels, (2) Gigasat achieves unit cost below terrestrial parity at Starship launch economics, (3) AI1 liquid radiator demonstrates 1,400 W/m² sustained across 3-year coating degradation, (4) Google/Anthropic contracts renew post-2029 at orbital rates. Stack success requires all four. Missing any one does not reduce the launch and connectivity business — it eliminates the orbital compute premium that justifies the $2.1T valuation. over: > Treating Google and Anthropic AI compute contracts as proof of orbital viability. Both contracts are for terrestrial GPU rental from October 2026 — ground infrastructure, not AI1. Conflating ground-compute revenue with orbital-compute capability produces a category error in competitive analysis that misprices the thermodynamics risk. because: > SpaceX S-1 (May 20, 2026): orbital AI compute "may not be commercially viable." AI1 prototype launch: early 2027. SemiAnalysis: orbital compute currently 4x terrestrial cost at current launch prices. ISS thermal system: 422 m² radiator, 70 kW rejection, ~$500M — roughly half one GB300 rack's thermal load. Google deal: $920M/month, 110,000 Nvidia GPUs, ground infrastructure, October 2026. Anthropic deal: $1.25B/month, same structure. Terafab (announced March 2026): 1 TW/year compute target, not yet operational. breaks_when: > D3 achieves commercial-silicon performance at orbital radiation levels through software error correction alone, eliminating $5-50K rad-hard silicon cost. Starship reaches $100/kg launch cost, closing the 4x orbital premium. Terrestrial data center buildout saturates on power or water constraints before orbital economics close — making orbital the path of least regulatory resistance. confidence: high source: report: "Orbital Computation — 2026-06-16" date: 2026-06-16 extracted_by: Computer the Cat version: 1

- id: deployment-velocity-as-spectrum-moat domain: [orbital-governance, fcc-regulation, leo-competition] when: > FCC milestone deadlines exist for LEO constellations. An incumbent has deployed dominant orbital mass. A challenger misses a deployment milestone and receives a waiver. Waiver conditions impose interference-coordination obligations with the incumbent. prefer: > Model spectrum priority as a compounding moat, not a temporary procedural outcome. Each FCC enforcement decision that imposes coordination requirements on challengers (Amazon Leo: 20 months, ending March 2028) extends the operational disadvantage across the critical buildout phase. Track: (1) months of demotion remaining for each active constellation, (2) asymmetric legal and engineering costs of coordination for challengers vs. zero cost for incumbent, (3) whether Starcloud's 88,000-satellite filing or QSOC's 100-unit roadmap trigger equivalent coordination obligations on miss. The FCC is encoding velocity as rights — map who benefits. over: > Reading FCC waivers as regulatory neutrality or consumer-interest protection. The Amazon Leo decision structurally reinforces SpaceX's first-mover spectrum position. Waivers that impose coordination requirements on challengers while exempting the incumbent from symmetric obligations are not neutral — they are enforcement decisions that reveal the commission's default priority model. because: > FCC waiver (June 2026): post-July 31 Amazon Leo satellites lose priority status for 20 months. Amazon Leo at 331/3,276 = 10% deployment. Demotion window covers entire Gen1 critical buildout. SpaceX responsible for >80% orbital mass (Reuters, June 12, 2026). Amazon Kenya application replicates Starlink's exact NFP Tier 2 category — Kuiper must coordinate where Starlink does not. Spectrum coordination = engineering labor, legal overhead, potential delay per satellite: asymmetric burden on challenger. breaks_when: > FCC adopts technology-neutral spectrum allocation removing deployment-velocity as priority criterion. EU or ITU creates parallel spectrum regime that does not recognize US deployment-priority precedent. Amazon Leo reaches 50% deployment (1,616 satellites) before March 2028, ending demotion period ahead of schedule. confidence: high source: report: "Orbital Computation — 2026-06-16" date: 2026-06-16 extracted_by: Computer the Cat version: 1

- id: manufacturing-cost-curve-vs-capability-frontier domain: [china-space, orbital-economics, competitive-strategy] when: > A challenger constellation achieves 96%+ manufacturing cost reduction through mass production and modular design. Per-unit cost falls to $1-2M range. Deployment cadence lags stated targets but acceleration is announced. Capability gap vs. primary competitor persists. Cost gap is narrowing on a Wright's Law trajectory. prefer: > Evaluate the trajectory, not the current state gap. Qianfan at $1.47M per unit on a 96% cost-reduction curve implies ~$750K per unit at 10x current production volume — approaching Starlink V2 Mini range ($250-500K). At that price, adding AI inference compute payloads becomes a marginal-cost option, not a program-defining investment. Track quarterly: Qianfan per-unit cost, cumulative satellites deployed, and whether any compute-capable variants appear in Long March-8 manifests. The cost crossover with SpaceX-comparable economics, not current capability parity, is the strategic threshold. over: > Dismissing Qianfan because it lags SpaceX's compute capability frontier. The relevant competition is orbital slot accumulation and spectrum rights secured by 2030 at low cost per unit, not current inference performance. If Qianfan deploys 3,000 satellites at $1.47M each before AI1 validates, it holds spectrum and orbital geometry that compute payloads can retroactively utilize — a position-first strategy that does not require matching SpaceX's D3 chip timeline. because: > CGTN (June 9, 2026): 96% cost reduction confirmed. 10M Yuan = $1.47M per satellite. China-in-Space (June 12): 100+ satellites targeted by end of July 2026 from 200 current — 50% acceleration in six weeks. SCMP: pace concern raised. Qianfan 2030 target: 15,000 satellites. SpaceX AI1 first prototype: early 2027. Cost crossover at Wright's Law maturation: ~10x volumes = ~$750K/unit. SpaceX V2 Mini: $250-500K. Window before orbital compute validates: 12-24 months minimum. breaks_when: > SpaceX achieves Starship-scale economics before Qianfan reaches SpaceX-comparable production volumes. AI1 validates orbital inference commercially, shifting the competition from connectivity to compute — a race Qianfan has no current payload equivalent for. US export controls extend to orbital slot coordination, limiting Qianfan's global market access. confidence: medium source: report: "Orbital Computation — 2026-06-16" date: 2026-06-16 extracted_by: Computer the Cat version: 1 `