Orbital AI: The Hyperscaler Triple

Three men stood up in the same week and said the same thing. Not in the same room. Not as part of any announced coordination. Separately, in earnings calls, press releases, and acquisition announcements, Jeff Bezos, Eric Schmidt, and Elon Musk each committed to putting data centres in orbit.

That week was 16 to 22 May 2026.

In the same week, Starcloud-1 reported nominal operations from 325 kilometres above Earth. The company confirmed it had trained a large language model in orbit. It confirmed a version of Gemini had run in orbital inference mode. According to Data Center Dynamics, reporting on 14 May 2026, these are the first publicly confirmed instances of both LLM training and LLM inference executed in orbit by a commercial operator.

The Space Mafia book named this moment the "deploy and defend" era: the period when the race is no longer speculative and the governance framework has not yet caught up. Three of the most likely principals of orbital compute announced their orbital ambitions in the same week. The archetype the book described confirmed its product works.

This is not a trend. It is an inflection point.

The Three Statements

Precision matters here. These are confirmed claims, reported by aerospace and AI infrastructure press between 16 and 22 May 2026.

Jeff Bezos, founder of Amazon and Blue Origin, publicly forecast gigawatt-scale orbital data centres within "ten or more years." This is a specific timeline and a specific scale target. Gigawatt-scale infrastructure requires launch costs and orbital construction methods that do not yet exist at commercial scale. The forecast is grounded in existing capability: Amazon Kuiper, Bezos' low-Earth orbit broadband constellation, has deployed 304 satellites across eleven missions and faces a regulatory milestone requiring 1,600 satellites, half of a planned 3,236-satellite constellation, by July 2026. The infrastructure layer is not hypothetical. It is under construction.

Eric Schmidt, former Google CEO and now chair of Relativity Space, confirmed that his acquisition of Relativity, which manufactures rockets using large-scale 3D printing, is explicitly targeted at orbital compute infrastructure. Schmidt's orbital ambition is simultaneously a launch supply-chain play. Building orbital data centres requires getting hardware to orbit affordably. Relativity is the launch strategy.

Elon Musk confirmed that SpaceX "is looking to deploy data centres in space." SpaceX already operates the Starlink constellation, which has surpassed 10,000 active satellites with AI-optimised constellation management. The data-centre announcement extends an existing orbital platform into a new product category.

These are not coordinated statements. Three principals, independently, in the same week, announced orbital compute commitments. The convergence is the story. The question of whether it reflects any coordination is for the reader to answer.

I am calling this the Hyperscaler Triple. Not because the three individuals are the same type of actor, they are not, but because they represent, collectively, three of the four largest compute-platform owners on the planet committing to the same orbital infrastructure direction in the same seven-day window. The Triple is a market signal, not a conspiracy theory.

Starcloud-1: The Operational Confirmation

Starcloud-1 carries an NVIDIA H100 GPU cluster, weighs 60 kilograms, and orbits at 325 kilometres. The launch was November 2025. On 14 May 2026, Data Center Dynamics reported nominal operations confirmed.

The confirmation covers two specific milestones: Starcloud trained a large language model in orbit, and a version of Gemini ran in orbital inference mode. The Space AI Monday standard protocol applies here. The operational fact, that both events occurred, is confirmed at the level of the company's own reporting. Performance claims, including training speed, inference throughput, and latency benchmarks, have not been independently verified. Treat those as manufacturer claims until independent benchmarking exists.

The Space Mafia book named Starcloud as the primary archetype of the book's argument: "Starcloud ignites orbital H100 GPUs. SpaceX expands Starlink into a skyborne supercomputer. Axiom Space plots data centre nodes above the Kármán Line." The book was written before nominal operations were confirmed. The 14 May 2026 confirmation is direct alignment between the book's analytical forecast and the operational record.

Starcloud-2 is next. Anchor tenant: Crusoe Cloud, confirmed in a press release from October 2025. Stated power: approximately 100 times that of Starcloud-1, per Starcloud's own communications. This is a manufacturer claim about a satellite not yet launched; treat it as indicative, not verified. Launch window: October 2026. GPU capacity available to external customers: early 2027. The timeline from "confirmed it works" to "available for your workloads" is approximately eight months.

Taiwan's Far EasTone has confirmed Amazon Kuiper as its commercial orbital communications partner for 2027. The Asian market is already building its orbital dependency architecture. The Hyperscaler Triple is not a Western story. It is a global one.

What the Digital Feudalism Framing Actually Means

The Space Mafia book described the outcome of unchecked orbital compute concentration with a term borrowed from political economy: digital feudalism. The passage has become current.

"By 2025, the top ten global universities commanded less aggregate compute in the cloud than a single Starcloud cluster, and even that is often rationed under restrictive access agreements. 'Democratising AI,' a favourite chorus at tech summits, becomes a cruel joke when the cost of entry is a launch slot or a buy-in to proprietary inference. The gap widens every quarter. The result will be digital feudalism: private princes controlling the only roads to orbital power, with states and citizens left to bargain for vassal access."

The three principals who made orbital commitments in the week of 16 to 22 May 2026 are among the most likely candidates for those private prince roles. None announced in coordination with any other. Each is acting on the same commercial logic: orbital compute offers jurisdictional arbitrage, specific latency advantages for non-real-time AI workloads, and near-zero labour costs for operations once the infrastructure is deployed.

The governance framework for this concentration does not exist.

The Outer Space Treaty of 1967 was written for state actors launching satellites. It does not address private commercial orbital data centres. The Artemis Accords, now with more than 40 signatories as of the 2025 expansion, provide non-binding political commitments on debris management, deconfliction, and scientific data sharing. They do not address orbital compute sovereignty, data jurisdiction, or AI governance. The EU data protection regime does not reach orbital infrastructure. According to the Space Mafia book, Starcloud's position on GDPR inquiries is explicit: "We do not process in the EEA; complaints belong at your nearest launch pad."

When Starcloud-2 launches in October 2026 with approximately 100 times the power of Starcloud-1, per Starcloud's own communications, commercial customers can place AI workloads in orbit from early 2027. No jurisdiction will reach them.

The Inequality and Consolidation Fault Line

The Space Mafia book identified five fault lines where terrestrial governance collapses at orbital altitude: data sovereignty erosion, IP and copyright collapse, environmental accountability vacuum, inequality and consolidation, and security and weaponisation.

The others have received significant series attention. Data sovereignty anchored Episodes 2 and 7. Security and the ground-segment attack surface anchored Episode 8. The Hyperscaler Triple activates the most underexplored fault line: inequality and consolidation. It stops being theoretical this week.

When three of the largest compute-platform owners on the planet all commit to orbital infrastructure in the same seven-day window, the consolidation fault line becomes structural. The orbital compute market is being claimed before the governance commons exists to manage it. The Antarctic Treaty model, which the book identifies as the closest precedent for a workable governance commons, has not been proposed for orbital compute. The Hyperscaler Triple makes that absence materially more urgent.

The physics of orbital compute reinforce the consolidation tendency. Radiation hardening at altitude is not a trivial problem. The Space Mafia book's engineering chapter describes the architecture: "Starcloud implemented triple-modular redundancy in its H100 GPU array; every critical logic block is triplicated, and the output is voted on each clock cycle." Triple-modular redundancy costs three times the power and three times the hardware mass. Single-event upsets, where cosmic radiation flips transistor states, are a real constraint on orbital compute density. All radiation mitigation approaches increase unit cost above terrestrial equivalents. The organisations that can absorb those cost differentials are the organisations that already dominate terrestrial compute. The consolidation dynamic at altitude mirrors what happened on the ground.

A brief note from the pirate radio parallel the book established: the 1960s radio ships operated from international waters until regulation eventually caught up. The Hyperscaler Triple suggests regulation will need to catch up faster this time.

The Adversarial Kardashev Dimension

The Hyperscaler Triple is a commercial story. It is also the precondition for a geopolitical one.

The Adversarial Kardashev variant, which this series introduced in Episode 8, asks a specific question: what happens when a hostile state actor applies the same capital-concentration playbook in orbit? China's Three-Body constellation programme, described in the Space Mafia book, is the live state-actor counterweight. The bifurcation of orbital compute infrastructure into commercially aligned and state-directed constellations follows the same logic as the terrestrial internet bifurcation, at a layer of the stack that no current governance framework addresses.

The point is not that the Hyperscaler Triple is itself a threat. The point is that the concentration it represents creates the structural conditions for the adversarial scenario. When Western orbital compute infrastructure is concentrated in three or four corporate entities, the comparative concentration in state-directed entities becomes a strategic variable. Enterprises, governments, and communities whose AI services depend on orbital infrastructure will have their choices shaped by that bifurcation, whether or not they are aware of it.

One cross-series governance note: the Cyber Sunday Cyber News cycle has observed that AI-discovered vulnerabilities can be exploited within hours on terrestrial infrastructure. Orbital infrastructure running equivalent service layers has no emergency patch pipeline at the same cadence. A satellite at 325 kilometres cannot be physically accessed on the same timeline as a rack in a terrestrial data centre. The governance gap is not only jurisdictional. It is also operational.

What This Means for NZ Organisations

New Zealand is not a passive observer of the Hyperscaler Triple. It is downstream of all three principals.

NZ's international internet traffic runs almost entirely through undersea cable infrastructure. The primary cables, Southern Cross NEXT and related infrastructure, carry the latency-sensitive workloads that make orbital routing impractical for general traffic. The round-trip minimum at low-Earth orbit is approximately 600 milliseconds. Real-time applications cannot route via orbit efficiently.

But an LLM inference job that does not require real-time response is a different workload category. Compute-intensive AI tasks without hard latency requirements can migrate to orbital inference without a user-experience penalty. By early 2027, Starcloud-2 plans to offer that capacity to external customers. The question is not whether this becomes technically available. It becomes technically available in approximately eight months. The question is whether your data governance framework has a position on it before that happens.

NZ government agencies are significant AWS consumers. Amazon Kuiper's July 2026 regulatory milestone means Bezos' orbital ambition has a specific near-term delivery date. Terrestrial vendor concentration is already a sovereignty concern for NZ AI governance planning. The same concentration, extended to orbital infrastructure, represents the same risk at a layer of the stack that no current NZ legislation reaches.

On 22 May 2026, Rocket Lab launched its ninth Electron for Synspective, carrying a SAR satellite from Launch Complex 1 at Mahia. Mahia is the only orbital launch site south of the equator in the Asia-Pacific. NZ launch activity continues. NZ's position in the orbital economy is active, not theoretical.

The GBSI Act, NZ's primary legislative instrument for commercial space activities, carries an authorisation compliance deadline of 29 July 2026. The Act's requirements apply to NZ launch operators and NZ-licensed spacecraft operators. The Act does not extend to foreign orbital data centres operating overhead. This is not a design flaw specific to NZ. It is the structural gap that exists between what any national space legislation can reach, under the state-actor framework of the Outer Space Treaty, and what commercial operators deploying hardware in orbit can do. The gap is a feature of the international legal order, not a domestic regulatory failure.

GNSS dependency is the third exposure dimension for NZ organisations. Transport logistics, agricultural precision systems, and power grid frequency synchronisation all depend on GPS and GNSS signals from foreign-operated constellations. Orbital governance directly implicates the entities that will operate the next generation of those constellations.

The practitioner question for this article is direct. When a vendor proposes running your AI workloads in Starcloud-2's orbital infrastructure from early 2027, under no applicable data-protection jurisdiction you can enforce, does your organisation's data governance framework have a position? Most do not. The organisations that develop that position in the next eight months will be ahead of the ones that develop it after the first incident.

The Governance Window Is Not Closing in the Future. It Is Closing Now.

Starcloud-2's October 2026 launch is the next concrete event. It will either confirm or complicate the trajectory the Hyperscaler Triple has set. The launch window is public. The anchor tenant is confirmed. The external customer availability timeline is stated.

Watch the Kuiper deployment milestone. Amazon's July 2026 regulatory deadline is a public commitment with consequences. Whether Amazon meets it reveals the pace at which the orbital infrastructure layer is materialising.

Watch the Relativity Space communications. Schmidt's acquisition is targeted at orbital compute. Relativity's announcements about orbital compute timelines will indicate how quickly Schmidt believes the gigawatt forecast can be shortened.

Watch for governance framework announcements. The Artemis Accords are the closest thing to an orbital governance commons currently in existence. None of the Hyperscaler Triple principals has committed their orbital infrastructure to any governance commons framework. The absence of that commitment is the signal.

The EU AI Act Omnibus reversal deferred Annex III high-risk obligations to December 2027. Orbital AI systems operate outside any EU jurisdiction as a legal matter, yet are commercially accessible to EU users. The regulatory arbitrage the Space Mafia book named has become operationally available.

The book argued that the Antarctic Treaty model offers the closest precedent: a commons framework with mutual inspection rights, binding transparency, and enforcement mechanisms. That framework does not yet exist for orbital compute. The Hyperscaler Triple makes its absence more urgent, because the principals building the infrastructure are now publicly named and operating on announced timelines.

The 2026 to 2030 period will set the defaults for a generation. Three of the principals just announced what those defaults will look like if no governance commons is established.

When your organisation's AI workloads run on infrastructure orbiting 325 kilometres above any legal jurisdiction you can enforce, who is your data's guardian?

The views expressed in this article are entirely my own, informed by more than 30 years of professional experience in architecture, security, and technology leadership in New Zealand. They do not represent the views of my employer, any government agency, or the New Zealand government. My commentary on legislation and policy is analytical, drawing on publicly available sources and my professional expertise in architecture, security, and AI governance. I follow the Public Service Commissioner's Code of Conduct for the Public Sector and social media guidance.

Andreas Hamberger is a Wellington-based enterprise architect and technology strategist with more than 30 years of experience in government, banking, transport, and aviation across New Zealand. He is the founder of Te Pono Limited, an Associate Member of the Institute of Directors NZ, and holds TOGAF and IAPP credentials. He is the author of The Hamberger Report series of books on AI governance, cybersecurity, and enterprise architecture. Space Mafia examines the sovereignty implications of orbital compute infrastructure.

I use AI tools, including Sudowrite, Claude, Perplexity AI, DeepSeek AI, ChatGPT, Grok, Copilot, Openart and Gemini, as deliberate production tools, not ghostwriters. This is consistent with my position: AI amplifies human judgement; it does not replace it. The frameworks, arguments, and editorial decisions in this series are original work. AI accelerated the process. The thinking is mine.

[1] Hamberger, A. Space Mafia: The Battle Between an Accountable "Heaven" and an Unfettered "Skynet" in Orbital AI. The Hamberger Report, 2025/2026.

[2] Data Center Dynamics. "Starcloud-1 achieves nominal operations; LLM trained in orbit." 14 May 2026. [URL to be added at publication]

[3] Crusoe Cloud / Starcloud. Press release on Starcloud-2 anchor tenancy and specifications. 22 October 2025. [URL to be added at publication]

[4] Multiple aerospace and AI infrastructure trade press sources documenting Bezos, Schmidt, and Musk orbital compute statements. 16 to 22 May 2026.

[5] Amazon. FCC filings and Kuiper deployment milestone communications. 2026.

[6] Far EasTone Telecommunications. Kuiper commercial partnership announcement. 2026.

[7] United Nations. Treaty on Principles Governing the Activities of States in the Exploration and Use of Outer Space, Including the Moon and Other Celestial Bodies. 1967.

[8] NASA. Artemis Accords: Principles for Cooperation in the Civil Exploration and Use of the Moon, Mars, Comets, and Asteroids for Peaceful Purposes. 2020, expanded 2025.

[9] New Zealand Parliament. Space (Mahina Maru) Activities Act. 2017 (as amended, Government Boost Satellites and Infrastructure Act provisions).

[10] Rocket Lab. Mission announcement: "Viva La Strix," ninth Electron for Synspective, Launch Complex 1 Mahia. 22 May 2026.