SpaceX Has 1.4 Gigawatts of AI Capacity Online and Wants 10 Gigawatts by Next Year. Here's Why Microsoft Could Be the One Writing the Check.
SpaceX has 1.4 GW of AI capacity online and aims for 10 GW by next year, driven by $15.8B Q2 capex. Major clients include Anthropic ($40B deal), Google Cloud ($30B deal), and Reflection AI. SemiAnalysis suggests Microsoft could be the next partner to bridge its compute gap until late 2027, leveraging SpaceX's rapid deployment capabilities and flexible cancellation policies.
Over the last year, Space Exploration Technologies (SPCX -2.64%) -- more commonly known as SpaceX -- has quietly made itself into a critical supplier of artificial intelligence (AI) infrastructure. The company has channeled enormous sums into capital expenditures, building out data center capacity rapidly.
During the second quarter alone, SpaceX allocated $15.8 billion of capex toward AI compute. The company's aggressive investments have already delivered 1.4 gigawatts (GW) of nameplate capacity online, and it has ambitions to reach 10 GW by the end of next year.
The scale of this expansion raises the question of which companies might help underwrite SpaceX's data center expansion plans. According to a new report from research firm SemiAnalysis, Microsoft (MSFT +1.55%) emerges as a potential partner. Here's why it's a logical candidate to be SpaceX's next AI data center customer.
Image source: The Motley Fool.
SpaceX has already teamed up with AI's largest developers
SpaceX has secured a few landmark agreements that underscore the viability of its AI infrastructure ambitions. A couple of months ago, Anthropic committed to pay $1.25 billion per month for access to over 300 megawatts (MW) of capacity at SpaceX's Colossus facility. The multiyear agreement is valued at more than $40 billion through 2029.
Google Cloud swiftly followed with a deal valued at $920 million per month. That partnership is scheduled to begin later this year, with Google Cloud accessing 110,000 Nvidia graphics processing units (GPUs) and supporting hardware. The contract represents more than $30 billion in spend over its full term.
Lastly, an Nvidia-backed start-up called Reflection AI has contracted with SpaceX to pay $150 million per month for compute in a deal that has a total value of $6.3 billion through 2029.
These deals are central to the SpaceX investment thesis because they prove that the company can convert its underutilized data center capacity into high-margin recurring revenue -- validating secular demand from both frontier model developers and cloud hyperscalers. By locking in large-scale offtake, SpaceX puts itself in a better position to fund further compute build-outs while commanding attractive data center economics.
NASDAQ: SPCX
Key Data Points
What are SpaceX's AI infrastructure ambitions?
During SpaceX's second-quarter earnings call, CEO Elon Musk outlined a detailed infrastructure roadmap that goes beyond incremental growth. He explained that SpaceX expects to finish 2026 with more than 2 GW of compute online and that cumulative capacity by the end of next year may be "closer to 10 GW of compute than 5 GW." Musk stressed that the company is really aiming for much higher capacity at the power and cooling level, targeting a series of projects that collectively reach 20 GW.
SpaceX intends to achieve this infrastructure expansion by focusing exclusively on Nvidia's Vera Rubin architecture. The company's goal is not merely internal training for its AI model, Grok, but shifting toward renting additional capacity for external training and inference applications.
Why Microsoft makes a logical partner for SpaceX
Analysts at SemiAnalysis suggest that Microsoft could be SpaceX's next major hyperscaler customer. Of note, Microsoft has already contracted for 10 GW of capacity elsewhere for a cumulative total of $300 billion. The subtle detail here is that the compute capacity to fulfill those agreements won't be fully online until late 2027 or possibly early 2028. With that in mind, there's an obvious question: How will Microsoft bridge the gap and meet its needs in the meantime?
SpaceX could be an ideal fit for two reasons. First, the company includes a 90-day cancellation policy in its capacity agreements. That provides its customers with financial flexibility at little risk to its balance sheet. Moreover, SpaceX has already proven that it can bring massive compute clusters online within a matter of months. This would allow Microsoft to secure large capacity blocks almost immediately.
Admittedly, for SpaceX to multiply capacity from 1.4 GW to 10 GW within a year and a half is quite an ambitious goal. That said, it's hard to ignore the company's demonstrated successes in execution. On-site natural gas power generation, modular power systems, and access to secondary turbine markets have already enabled SpaceX's data center facilities to come online in mere months rather than several years.
Ultimately, I agree that Microsoft is a logical candidate for a SpaceX partner, given that its operational needs may exceed those of frontier AI labs, while its scale and the urgency of its compute needs align with SpaceX's delivery speed.
The combination of proven hyperscaler demand, engineering advantages, and complementary business needs makes SpaceX's visionary infrastructure expansion both strategically coherent and commercially plausible. Against this backdrop, investors may want to consider buying the dip in SpaceX stock as its AI business looks poised for a potential breakout that few seem to be anticipating.
