Microsoft has not announced a lease for more than 3 gigawatts of SpaceX AI data-center capacity in 2027, and neither has SpaceX. The claim appears in an August 8 post by NextBigFuture, but the article’s own text does not document a signed agreement, identify a facility, name a contract value, or cite either company. More importantly, it retreats from its headline in the body: “IF Microsoft leases” capacity from SpaceX.

That is a meaningful distinction. A 3 GW lease would rank among the largest single AI infrastructure commitments disclosed anywhere, roughly 50% larger than Microsoft’s already substantial 2 GW Pecos, Texas campus announcement in June. A deal of that size would also be difficult to reconcile with the public record without at least some disclosure from Microsoft, SpaceX, a developer, a power supplier, or a major financial filing.

The evidence available as of August 9 supports a more modest conclusion: Microsoft is aggressively buying and building AI capacity; SpaceX has said it intends to sell excess terrestrial AI compute to selected third parties; and a future commercial relationship is plausible. The reported 3+ GW Microsoft lease, however, remains unverified speculation rather than established news.

Industrial power-and-launch complex at dusk with 1–3 GW capacity claims marked “Unverified.”The headline overstates what the article establishes​

NextBigFuture presents the claim as settled in its headline: Microsoft “Will Lease 3+ Gigawatts” from SpaceX in 2027. But its article offers no contract documents, no attributable company statement, and no reporting that identifies where the capacity would be located or what Microsoft would pay.

Instead, the post makes a chain of projections: SemiAnalysis’s assumptions about AI token revenue per gigawatt, an assertion that SpaceX could complete 10 GW of AI data-center capacity, and a calculation that Microsoft could reach more than $300 billion in annualized Azure AI revenue if it leased capacity from SpaceX. Those projections may be interesting to investors, but they do not establish that a lease exists.

The lack of basic deal terms is especially conspicuous for a claim at this scale. A genuine 3 GW capacity commitment would need answers to several operational questions:

  • Is the figure IT load, total facility power, or contracted power generation capacity?
  • Is the capacity in one campus or several locations, and is it terrestrial or orbital?
  • Is Microsoft leasing powered shells, GPU-equipped compute, or a managed inference service?
  • What is the delivery schedule during 2027, and what hardware generation is included?
  • Is there a minimum-spend obligation, a take-or-pay structure, or a termination right?

None is supplied. The post also provides no indication that Microsoft, SpaceX, xAI, regulators, lenders, utilities, or construction partners have confirmed the supposed transaction.

That is not a minor reporting gap. The difference between “SpaceX may someday offer capacity to Microsoft” and “Microsoft has signed for 3 GW in 2027” is the entire story.


SpaceX’s own prospectus describes a 1 GW installed base, not a disclosed Microsoft commitment​

SpaceX’s June 2026 prospectus offers the strongest primary record on the company’s terrestrial AI operations. It says the combined nameplate compute draw of COLOSSUS and COLOSSUS II reached approximately 1.0 GW as of March 31, 2026. The prospectus explicitly defines that metric as GPU power draw and says it excludes cooling, power-distribution losses, lighting, security, and other facility overhead.

That definition matters because the gigawatt numbers now circulating in AI infrastructure coverage are often not comparable. A site can be described in terms of grid interconnection, generation capacity, facility load, or accelerator draw. Each measure can produce very different numbers for what appears to be the same data center.

SpaceX also says in the prospectus that it intends to sell excess AI capacity to a limited number of third parties and explore becoming a supplier alongside firms such as CoreWeave and Nebius. Microsoft is mentioned in the document as a competitor in AI models and cloud services, not as a named customer or counterparty to a capacity contract.

The prospectus does describe future expansion at COLOSSUS II, including more than 400 MW of additional compute capacity once a next phase is operational. But that still does not amount to a disclosed 3 GW Microsoft lease. Nor does it establish that three gigawatts will be available to outside customers in 2027 rather than used for Grok training, inference, other customers, redundancy, or future expansion.

SpaceX has commercial ambitions in AI infrastructure. It has not publicly tied Microsoft to a 3 GW capacity purchase.

Orbital AI cannot be the 2027 explanation​

The SpaceX filing is also clear on timing for its space-based ambitions: it expects to begin deploying orbital AI-compute satellites as early as 2028. It describes early satellites generating about 100 kilowatts of compute power and acknowledges that meaningful scale would require enormous satellite manufacturing and launch cadence.

That makes an orbital interpretation of the NextBigFuture headline untenable for 2027. A 3 GW orbital service would equal 30,000 units of 100 kW compute capacity before accounting for redundancy, networking, ground links, maintenance, and the realities of launching and operating hardware at scale. SpaceX’s long-term target of deploying 10 GW or even 100 GW annually is presented as an ambition based on future manufacturing and launch capabilities, not a capacity commitment available next year.

If Microsoft were to buy from SpaceX in 2027, the available public evidence points to terrestrial capacity, likely associated with SpaceX/xAI’s fast-build data-center model and behind-the-meter power generation. But even that interpretation remains a hypothesis. The companies have not identified a site, utility arrangement, construction program, or product offering connected to Microsoft.

For IT buyers, the distinction is practical. Terrestrial hosted GPU capacity can be contracted, provisioned, connected to Azure services, and governed through conventional data-residency and compliance arrangements. An orbital compute service would introduce radically different architecture, latency, data-transfer, support, fault-tolerance, and regulatory questions. Treating the two as interchangeable “gigawatts” obscures the actual delivery risk.


Microsoft’s confirmed capacity plans show why the rumor sounds plausible​

Microsoft is not standing still. The company announced in June that it will build an approximately 2 GW data-center campus in Pecos, Texas, over the next five to seven years. Microsoft said the campus will use dedicated, on-site energy infrastructure initially and that it is funding the generation and supporting equipment needed for its own operations.

Microsoft also has a documented lease at the Crusoe-developed Abilene, Texas campus. The Information reported that Microsoft had leased a 900 MW site after Oracle and OpenAI walked away from the planned expansion, while Data Center Dynamics earlier reported an approximately 700 MW arrangement based on sources familiar with the matter. The later confirmed figure is 900 MW, illustrating another important point: early megawatt reports can change materially as deals close and scope becomes public.

Those transactions show that Microsoft has both the financial capacity and the immediate need to secure enormous blocks of AI power. They do not validate the SpaceX claim. Pecos is Microsoft-built capacity; Abilene is a Crusoe lease. Neither is SpaceX capacity, and neither adds evidence of a separate 3 GW agreement.

Microsoft’s April 27 restructuring of its OpenAI relationship does strengthen its incentive to acquire more compute for Azure and its own AI services. OpenAI said Microsoft will no longer pay OpenAI a share of Microsoft revenue earned by offering OpenAI models on Azure. But the same announcement says OpenAI’s revenue-share payments to Microsoft continue through 2030, subject to a cap. The NextBigFuture post gets the first half of that change right while omitting the continuing revenue stream in the other direction.

More capacity does give Microsoft additional choices: serve OpenAI, run Microsoft’s own models and Copilot services, or sell AI infrastructure and model access to Azure customers. Yet the leap from that incentive to a forecast of Azure growth exceeding 100% next year is unsupported. Microsoft’s recently reported Azure growth has been in the low-40% range; capacity availability is important, but revenue depends on customer demand, pricing, utilization, model costs, and the timing of hardware deployment.

The consequence: do not plan around a deal that has not been disclosed​

The most useful reading of this report is not that Microsoft has secured a vast new supply of SpaceX compute. It is that the AI infrastructure market has reached a point where a 3 GW bilateral lease sounds credible enough to spread quickly—even when the underlying evidence does not show that it happened.

Windows and Azure administrators should treat it accordingly. There is no announced Azure region, no new GPU SKU, no private-preview program, no published service-level agreement, and no customer migration path tied to SpaceX capacity. Enterprise procurement teams should not infer future Azure availability, pricing, or OpenAI-model access from the claim.

A confirmed agreement would leave a substantial paper trail: a named capacity site, a construction or power partner, a SpaceX disclosure, Microsoft commentary, or an Azure product and region roadmap. Until one of those appears, the 3+ GW figure belongs in the category of market speculation—not Microsoft’s capacity plan for 2027.


References​

  1. Primary source: nextbigfuture.com
    Published: August 8, 2026 at 6:29 PM UTC
  2. Related coverage: blogs.microsoft.com
  3. Related coverage: techinvestment.com
  4. Related coverage: sciencesprings.wordpress.com
  5. Related coverage: explainx.ai
  6. Related coverage: inc.com
  7. Related coverage: content.spacex.com
  8. Related coverage: content.spacex.com
  9. Related coverage: local.microsoft.com
  10. Related coverage: theinformation.com
  11. Related coverage: tbpndigest.com
  12. Related coverage: news.microsoft.com
  13. Related coverage: datacenterdynamics.com
  14. Related coverage: siliconreport.com
  15. Related coverage: nextbigfuture.com
  16. Related coverage: news.bloomberglaw.com
  17. Related coverage: seekingalpha.com
  18. Related coverage: windowscentral.com
  19. Related coverage: openai.com
  20. Related coverage: openai.com
  21. Related coverage: theinformation.com
  22. Related coverage: arstechnica.com
  23. Related coverage: microsoft.com
  24. Related coverage: investing.com
  25. Related coverage: sec.gov
  26. Related coverage: microsoft.com
  27. Related coverage: washingtonpost.com
  28. Related coverage: finance.yahoo.com