Version 1

SpaceX's S-1 IPO filing, published in May 2026, disclosed a landmark $1.25 billion-per-month compute deal with Anthropic for capacity on its COLOSSUS and COLOSSUS II data centers, running through May 2029 and totaling approximately $45 billion in contract value [1][2].

The filing also revealed that SpaceX formally acquired xAI, folding Grok AI under a new SpaceXAI division, and repositioned its traditional space launch and satellite business as a support layer for its AI ambitions [3].

SpaceX claims a $26.5 trillion total addressable market and frames its ability to rapidly build compute infrastructure at lower cost — citing SemiAnalysis research showing data center construction accounts for ~30% of AI compute capex for most companies — as its structural competitive advantage [4][6][3].

The Anthropic deal makes SpaceX one of the largest compute infrastructure players in AI overnight, while simultaneously creating a paradox: SpaceX is a critical supplier to Anthropic, one of its primary competitors in the AI model market. The scale of Anthropic's commitment — $15 billion per year — signals how aggressively frontier AI labs are locking in compute capacity, and the 90-day termination clause on either side adds a fragility layer to a deal of extraordinary size [2].

SpaceX's S-1 IPO filing landed in May 2026 and immediately drew intense scrutiny for the extraordinary scale of a single commercial disclosure: a Cloud Services Agreement with Anthropic PBC committing $1.25 billion per month for compute access on the COLOSSUS and COLOSSUS II data centers through May 2029 [1][2]. At roughly $15 billion per year, the deal's total contract value reaches approximately $45 billion — making it one of the largest AI infrastructure commitments ever made public. Capacity is ramping across May and June 2026 at a reduced rate before reaching full fees, and either party retains the right to exit with 90 days' notice, a clause that adds legal flexibility but also counterparty risk to a deal of this size [2].

The filing simultaneously revealed that SpaceX has completed its acquisition of xAI, integrating Elon Musk's AI startup into a new SpaceXAI division that now oversees the Grok family of models and the Grok chatbot [3]. In a strategic framing that analysts flagged as striking, SpaceX described its legacy space launch and satellite businesses — including Starlink — as playing a supporting role to its AI ambitions, not the other way around [3]. The company claims a total addressable market of $26.5 trillion, a figure that approaches US nominal GDP and that SpaceX attributes primarily to AI [3].

A key pillar of SpaceX's AI infrastructure argument — cited directly in the S-1 — is research from SemiAnalysis showing that for most AI companies lacking a construction cost advantage, approximately 30% of total compute infrastructure capital cost derives from data center construction alone [4]. SpaceX frames its internal ability to build and operate data centers at scale as a durable structural advantage. COLOSSUS II, which also hosts Grok 5 training workloads, is equipped with NVIDIA GB200 NVL72 rack systems [5].

Skeptics, led by Ars Technica's coverage, note a fundamental tension in SpaceX's positioning: Grok is currently losing ground to the very AI labs — OpenAI and Anthropic — that SpaceX hopes to displace, while simultaneously depending on Anthropic's massive compute payments to fund the infrastructure buildout [3]. SpaceX's S-1 strategy section is titled 'Our AI Compute Infrastructure Advantage and Growth Strategy' and asserts that 'AI leadership will be defined by the ability to rapidly scale compute capacity to support exponential usage growth and frontier intelligence' [6] — a claim that reads as both aspiration and market positioning ahead of the IPO.