When the Launch Provider Becomes the Compute Provider: Google, SpaceX, and the Orbital Concentration Question

The week of 5 to 12 May 2026 will be studied in future enterprise architecture curricula. Not because something unprecedented was announced, but because several things that were already happening crossed the threshold from pattern to settled fact in a single calendar week.

On 12 May, the Wall Street Journal reported that Google is in advanced discussions with SpaceX to launch the Project Suncatcher orbital data centre satellites. Earlier that week, Anthropic confirmed it had agreed to use the full computing capacity of SpaceX's Colossus 1 facility in Memphis: more than 300 megawatts and over 220,000 NVIDIA GPUs, with the contract including potential collaboration on multi-gigawatt orbital data centres in future. And xAI, which merged with SpaceX in February 2026 at a combined valuation of reportedly US$1.25 trillion, was reorganised during the same week into a SpaceXAI division, per Elon Musk's announcement.

Three of the four major frontier AI labs operating at scale are now structurally bound to SpaceX: Anthropic as a compute customer with an orbital extension option, Google as a launch customer and existing minority shareholder, xAI as an internal business unit. OpenAI is the fourth major frontier lab, and it has not, to date, disclosed a comparable structural commitment to SpaceX.

This is the architectural starting condition for orbital compute in 2026. The question is not whether the concentration happened. It did, in one week, confirmed by multiple sources across five separate reporting outlets. The question is what it means for how organisations design, procure, and govern orbital workloads from this point forward.

The Three Clocks Reading

The Three Clocks framework has tracked the urgency dynamics of this series since Episode 1. It remains the sharpest lens for what happened last week.

The Commercial Clock is the primary driver. SpaceX is preparing for what is expected to be among the largest initial public offerings in history, at a reported valuation of approximately US$1.75 trillion. That figure is reported rather than audited and should be read as directional, not precise. But the directional signal is clear: the orbital data centre commitments being locked in this week are not incidental to the IPO preparation. They are the IPO story. A frontier AI lab signing a long-term compute agreement converts anticipated revenue into contracted revenue. Three such labs convert the revenue story into a category claim. That changes the risk profile for institutional investors categorically.

The Strategic Clock is running in parallel. Concentrating three frontier labs within a single company's infrastructure gives that company structural leverage that no competitor can quickly replicate. A new orbital data centre operator needs not only capital for manufacturing and launch, but years of customer relationship building. The frontier labs have now signalled where they are building. New entrants must compete for what remains.

The Regulatory Clock is the slowest of the three, and the gap between it and the other two widens each week. Regulators are still developing frameworks for terrestrial AI. They are not yet building equivalent frameworks for orbital compute concentration. SpaceX's filing with the FCC for authorisation to deploy up to one million satellites for orbital data centre purposes, made in February 2026, sits in regulatory territory with no orbital antitrust equivalent and no cross-jurisdictional data sovereignty framework. The governance vacuum documented in Space Mafia is not a future condition. It is the operating environment in which these deals are being struck.

The Conflict Clock, which entered the framework in Episode 3, is not the primary driver this week. The concentration described here is commercial, not adversarial. That does not make it less significant for sovereignty analysis. It makes it different.

The Vendor-Decides Pattern

There is a structural concept that the week's events bring into focus, one that I want to name explicitly because it will recur across this series and its adjacent ones.

I call it the Vendor-Decides Pattern.

When a vendor controls both the launch layer and the compute layer of a technology stack, decisions that users might otherwise make for themselves become vendor decisions by default. Which jurisdiction processes your data. Which legal framework governs a breach investigation. Whether meaningful redundancy is architecturally possible if that vendor experiences a disruption. These decisions are made at design time, by the entity that owns the infrastructure. Not through malice, but because the architecture does not permit another outcome.

Google's Project Suncatcher illustrates the pattern at orbital altitude. The target architecture is 81 satellites operating in a one-kilometre radius compute cluster, equipped with Google's custom Tensor Processing Units, manufactured by Planet Labs, with SpaceX as the reported launch provider. When those satellites are in orbit, the data processed on them will be subject to jurisdictional conditions established at design time: the registration of the launch vehicle, the legal domicile of the operating entity, and the contract terms negotiated before a single workload runs. Those conditions are being determined this week. They cannot be renegotiated once the infrastructure is operational.

Anthropic's Colossus 1 agreement extends the same pattern into terrestrial infrastructure with an orbital extension clause built in. The 300-plus megawatts of compute in Memphis is a SpaceX-operated facility. The contract's potential collaboration on multi-gigawatt orbital data centres means the architectural relationship is designed for continuity from terrestrial to orbital as cost curves shift. Same vendor, same operational terms, at altitude when the economics permit.

The xAI absorption into SpaceXAI is the most complete expression of the Vendor-Decides Pattern. xAI is no longer a customer making procurement decisions. It is an internal unit. The compute, launch, connectivity, and AI product decisions are made inside a single corporate structure. There is no procurement relationship to govern, because there is no separation to govern across.

Three lab relationships with SpaceX. Three different structural forms: customer, customer-shareholder, internal unit. One pattern.

The Startup Layer Confirms It

If you are inclined to read the frontier lab concentration as temporary, driven by IPO timing and specific relationships, the Cowboy Space Series B is the counter-argument.

Cowboy Space, formerly Aetherflux, reportedly raised US$275 million in a Series B at a valuation of approximately US$2 billion, according to Data Centre Dynamics on 13 May 2026. The company's model is structural, not opportunistic: take a rocket upper stage, convert it into a one-megawatt orbital data centre at deployment, target a demonstration satellite for Q4 2026. It was founded by a co-founder of Robinhood.

At the startup layer, the launch-plus-compute integration is the designed architecture. The venture capital thesis for orbital compute is explicitly built on controlling both layers. The Vendor-Decides Pattern is not a consequence of one company's particular strategy. It is the structure that investors are funding across the category.

The Economics Have Not Converged. The Decisions Have.

There is an important distinction to draw clearly, because conflating these two things would misread the risk.

Orbital compute has not reached cost parity with terrestrial alternatives. Today's satellite construction and launch costs make orbital data centres significantly more expensive per workload than terrestrial facilities. The Anthropic, Google, and Cowboy Space commitments are bets on future cost reduction, particularly on reusable heavy-lift launch, not on current price parity.

The architecture decisions, however, are being made now, and at a scale that forecloses options. A contracted compute relationship does not dissolve because a competitor improves its pricing. The organisations making these commitments are locking in operational dependencies for years. Orbital infrastructure does not decommission lightly.

There is a workload suitability constraint worth understanding for your own architectural assessment. Orbital compute is currently suited to asynchronous, batch-processing workloads. The Earth-to-orbit latency profile makes real-time inference, the kind that drives customer-facing AI services, better placed terrestrially for now. Thermal management across sun and eclipse temperature ranges and the radiation hardening requirements for long-duration operation also constrain hardware choices in ways terrestrial facilities do not face.

This matters for how you read the concentration risk. The workloads for which orbital concentration becomes relevant are batch training runs, bulk inference tasks, and large-scale data processing operations: the workloads that move in response to cost and capacity signals, not latency signals. If your highest-value data processing sits in that category, the orbital supply chain consolidation is directly relevant to your vendor risk register. If it does not, the relevance is indirect but still present, because the infrastructure decisions being made this week will shape the cost structures and contract terms your organisation will encounter when it does engage with orbital compute.

According to IEA projections, global data centre electricity consumption is expected to exceed 1,000 terawatt-hours by the end of 2026, roughly equivalent to Japan's annual electricity consumption. Connection waiting times for new gigawatt-scale AI facilities in established markets reportedly reach up to seven years. The frontier labs are not choosing orbital compute because it is currently cheaper. They are choosing it because terrestrial capacity at the scale they need is increasingly unavailable. The orbital layer is being built under pressure from a terrestrial constraint that is binding now, not at some future planning horizon.

In Episode 9, the GAO's finding that AI data centres could account for approximately 12 percent of US electricity demand by 2028 set the context for why the pivot to orbit has accelerated. Anthropic's 300-megawatt commitment to a single facility, signed in the same week as Google's reported launch discussions, is the commercial confirmation of that finding landing in contract form.

What This Means for NZ Procurement

New Zealand organisations are not yet running orbital workloads in any meaningful volume. That does not mean the decisions being made this week are irrelevant to NZ enterprise architecture.

The procurement decisions being made by Anthropic, Google, and the SpaceXAI division will shape the cost structures, contract terms, and dependency chains that NZ organisations encounter when they do engage with orbital compute. And the structural concentration being established now is easier to describe than to exit.

Three practical observations for enterprise and architecture teams.

Jurisdiction follows architecture, not contract intent. NZISM v3.9 controls on identity and cryptographic protection apply to NZ government data wherever it is processed, including in orbit. An architecture that routes sensitive NZ workloads through orbital infrastructure without explicit jurisdiction analysis is not NZISM v3.9 compliant regardless of the vendor's contractual assurances. The upcoming 29 July 2026 authorisation deadline under the GBSI Act requires NZ entities engaging in space activities to hold appropriate authorisation. The definition of what constitutes a space activity has implications for organisations that process data on satellite infrastructure. A legal assessment specific to your organisation's orbital footprint, if any, is the appropriate starting point, not a vendor's standard terms.

Orbital concentration mirrors terrestrial concentration at a different layer. NZ government and commercial workloads already run heavily on a small number of global cloud providers. Three frontier AI labs structurally tied to one launch and compute provider is, architecturally, the same question NZ procurement already manages on the ground, with one material difference: orbital concentration operates without the geographic regulatory equivalents and portability standards that apply to terrestrial cloud. The Five Fault Lines framework from Space Mafia identifies data sovereignty erosion as the first fault line that intensifies at altitude. The week's events confirm the fault line is no longer theoretical.

Rocket Lab provides NZ-adjacent launch capability, not equivalent scale. Rocket Lab operates commercial launch from the Mahia Peninsula, holds a growing orbital servicing portfolio through its Gauss propulsion capability, and is building genuine orbital infrastructure relevance for allied-nation partners, as Episode 7 documented. The orbital compute commitments made this week are at Starship-class heavy-lift scale. Rocket Lab and SpaceX are not substitutes for that workload category. What Rocket Lab does provide is a structural option for smaller NZ orbital deployments outside the SpaceX dependency, consistent with the Collaborative Vertical Integration model first described in Episode 7. That is a margin hedge, not an equivalent alternative. It matters for NZ procurement strategy precisely because it is the most accessible structural differentiation available to NZ organisations.

The Architecture Question That Remains Open

The Five Country Council issued joint guidance on 1 May 2026 addressing agentic AI risks. That guidance covers the terrestrial AI governance layer. A parallel multilateral framework for orbital compute concentration does not yet exist.

That is a factual observation, not a prescription. The governance timeline for orbital infrastructure runs on a different cadence from the Commercial and Strategic Clocks driving this week's announcements. Understanding that gap is part of the architectural analysis any organisation considering orbital workloads needs to conduct.

In Space Mafia, the Heaven Vector describes a future of transparent international frameworks: orbital compute governed by clear jurisdictional rules, portability guarantees between providers, and concentration oversight that gives users meaningful options. The Skynet Vector describes the alternative: orbital infrastructure concentrated in entities operating beyond the reach of any single regulatory framework, with governance determined by the architecture decisions of whoever controls the launch stack.

The week of 5 to 12 May 2026 moved the orbital compute story firmly toward the Skynet Vector as the current observable reality. The Heaven Vector outcome is not foreclosed. But it requires governance frameworks that currently do not exist at orbital altitude, and it requires them before the infrastructure being committed to this week becomes the settled architecture.

Space Mafia argued that the 2026 to 2030 window would set the governance defaults for a generation. We are inside that window now. The architecture decisions being made in boardrooms this week, by organisations many times larger than any New Zealand entity, will define the dependency conditions under which NZ organisations eventually engage with orbital compute.

The question for your organisation is not whether you have an orbital compute strategy today. Most do not, and that is a reasonable position. The question is whether the teams responsible for your enterprise architecture and your vendor governance have begun to treat the orbital layer as an arriving dependency rather than a distant scenario.

This week confirmed it is arriving.

What is your current thinking on when orbital compute appears on your vendor risk register?

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, security practitioner, and technology strategist with over 30 years of experience across New Zealand's public and private sectors. He holds TOGAF, IAPP, and AMInstD credentials and is the founder of Te Pono Limited. He publishes The Hamberger Report, a multi-series LinkedIn publication, as a New Zealand leader in architecture and security. 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] Wall Street Journal. "Google in Advanced Talks with SpaceX to Launch Project Suncatcher Satellites." 12 May 2026. [URL TBC at publication]

[2] TechCrunch. "Anthropic Signs Agreement to Use SpaceX Colossus 1 Facility, 300MW, 220,000 GPUs." Week of 5 May 2026. [URL TBC]

[3] Reuters / GVWire. "Anthropic-SpaceX Colossus 1 Agreement Details." Week of 5 May 2026. [URL TBC]

[4] Yahoo Finance. "Anthropic SpaceX Deal: Compute Commitment and Orbital Extension Option." Week of 5 May 2026. [URL TBC]

[5] TechCrunch. "SpaceX Files FCC Application for Up to One Million Orbital Data Centre Satellites." February 2026. [URL TBC]

[6] Tech Startups. "SpaceX Preparing IPO at Reported Approximately US$1.75 Trillion Valuation." May 2026. [URL TBC]

[7] Yahoo Finance. "xAI-SpaceX Merger: Combined Entity Valued at Reportedly US$1.25 Trillion." February 2026. [URL TBC]

[8] Musk, E. Announcement of SpaceXAI division reorganisation. Week of 5 May 2026. [Platform X / URL TBC]

[9] SpaceX / regulatory filings. Google 6.1 Percent Stake: US$900 Million Investment, 2015. [URL TBC]

[10] WCCFTech / The Assembly. "NVIDIA Orbital AI Partner Roster: Starcloud, Planet Labs, Kepler Communications, Firefly Aerospace, Sophia Space." 11 May 2026. [URL TBC]

[11] International Energy Agency. Data Centres and Data Transmission Networks. May 2026 update. https://www.iea.org/reports/data-centres-and-data-transmission-networks

[12] Data Centre Dynamics. "Cowboy Space Raises US$275 Million Series B at Reported US$2 Billion Valuation." 13 May 2026. [URL TBC]

[13] Bloomberg / Reuters. "Google Project Suncatcher: 81 Satellites, One-Kilometre Cluster, Custom TPUs, Planet Labs Manufacturing." May 2026. [URL TBC]

[14] National Cyber Security Centre NZ. "Five Country Cybersecurity Agencies Issue Joint Guidance on Agentic AI." 1 May 2026. [URL TBC]

[15] Hamberger, A. Space Mafia: The Battle Between an Accountable "Heaven" and an Unfettered "Skynet" in Orbital AI. Te Pono Limited, 2026.