Anthropic Locks In SpaceX's Colossus 1 Compute Before June IPO

Anthropic has signed SpaceX to supply the full compute capacity of the Colossus 1 data center — a deal that lands weeks before Anthropic's expected June IPO and signals the AI lab is racing to lock in infrastructure muscle before public markets weigh in on its $61.5B valuation.

An infinite dark data center hall with glowing blue server racks beneath a star-filled ceiling crossed by a rocket contrail.
When the rocket company becomes the cloud: Anthropic's Colossus deal rewrites who controls the AI compute stack.

Anthropic has secured SpaceX as the compute provider for the entirety of Colossus 1, a large-scale data center project, in a deal finalised ahead of the AI safety company's anticipated June 2026 IPO. The agreement hands SpaceX — better known for orbital launches than data-centre operations — full capacity rights over one of the most closely watched AI infrastructure builds of the year, and it arrives at a moment when Anthropic is under maximum scrutiny from institutional investors sizing up a debut that Bloomberg has previously pegged at a potential valuation north of $60 billion.

What's New on the Table

The Colossus 1 facility is not a conventional hyperscaler lease. According to people familiar with the arrangement, Anthropic structured the deal to give SpaceX the full compute footprint of the centre rather than a carved-out partition — a topology that differs sharply from the multi-tenant GPU-cluster agreements that have dominated AI infrastructure deals over the past two years. The practical implication is that SpaceX becomes, in effect, Anthropic's sole external compute counterparty for this facility, removing the vendor diversification buffer that most large AI labs maintain as a hedge against supply shocks.

SpaceX's infrastructure ambitions have been quietly expanding. The company's Starlink ground-segment investments and its internal supercomputer work for xAI — the Elon Musk-founded rival to Anthropic — have given it hands-on experience operating GPU-dense environments at scale. It is worth noting that xAI's own Colossus supercluster in Memphis, Tennessee, was publicly announced in late 2024 and reached a reported 100,000 Nvidia H100 equivalents in its first phase. The naming overlap between xAI's Colossus and Anthropic's Colossus 1 is, according to sources close to Anthropic, coincidental — but the competitive optics are anything but.

Exact financial terms of the compute agreement have not been disclosed in public filings as of press time. Based on comparable wholesale GPU-cluster contracts in the $1–2 billion per annum range for facilities of similar scale, the multi-year commitment could represent one of the largest single infrastructure expenditure lines in Anthropic's pre-IPO cost structure.

The Capital Picture

Timing is everything. Anthropic filed confidentially for a public offering earlier this year, with bankers — including leads at Goldman Sachs and Morgan Stanley, per prior Reuters reporting — targeting a June 2026 listing window. Locking in Colossus 1 capacity before the S-1 goes effective serves a dual purpose: it firms up the compute roadmap that investors will scrutinise line by line, and it demonstrates that the company can secure infrastructure at a moment when Nvidia H100 and H200 allocations remain constrained across the industry.

Anthropic's most recent disclosed funding round — a $2.75 billion raise co-led by Google and a consortium of strategic investors closed in early 2025 — valued the company at approximately $61.5 billion post-money. Whether public markets will sustain or expand that multiple depends heavily on the company's ability to show a credible path to the compute capacity needed to train and serve next-generation models beyond its current Claude 3 family. The Colossus 1 deal is, in that context, as much an investor relations document as it is an infrastructure contract.

The involvement of SpaceX — a private company with no public equity — also raises governance questions that IPO lawyers will need to address. SpaceX and Elon Musk's broader portfolio sit in a complex competitive relationship with Anthropic: Musk co-founded OpenAI, departed acrimoniously, and subsequently launched xAI, which competes directly with Anthropic's enterprise Claude products. Any material dependency on a Musk-affiliated entity will require Anthropic to disclose and explain the arrangement's arm's-length nature in its S-1 risk factors.

BlockAI News' Take

The surface reading of this deal is straightforward: a cash-rich AI lab buys compute to train bigger models. The deeper reading is more interesting. Anthropic choosing SpaceX — rather than defaulting to Microsoft Azure, Google Cloud, or Amazon Web Services — suggests the company is deliberately hedging its infrastructure dependency away from the hyperscalers that also happen to be its direct commercial rivals in the enterprise AI market. Google is both an Anthropic investor and a competitor; Amazon, via its own $4 billion Anthropic investment, is similarly entangled. A compute partner with no competing AI product line — even an unusual one like SpaceX — may have looked strategically cleaner to Anthropic's board as it prepared for public-market transparency.

There is also a signal here about the broader fragmentation of AI infrastructure. The era when every serious AI lab simply spun up capacity on AWS or Azure is ending. Custom silicon, proprietary data centres, and unorthodox compute partnerships are becoming markers of competitive differentiation. Meta has gone fully in-house. xAI built Colossus on owned hardware. OpenAI is co-investing in data centres with Microsoft under the Stargate banner. Anthropic's SpaceX move fits the same pattern: own — or at minimum exclusively contract — your compute destiny before someone else's cloud roadmap defines your model capabilities.

For Web3 and decentralised compute advocates, this deal is a useful stress test of the thesis that distributed GPU networks can compete for frontier AI workloads. The fact that Anthropic reached for a centralised, single-counterparty solution — even an unconventional one — rather than a decentralised compute marketplace suggests frontier model training still demands the latency, interconnect bandwidth, and contractual certainty that no decentralised protocol currently delivers at scale.

Watch for Anthropic's S-1 registration statement, expected to become public four to six weeks before the IPO date, for the first official disclosure of Colossus 1's capacity specifications, contract duration, and the financial commitments baked into the SpaceX arrangement — those numbers will be the real signal about how seriously investors should take the compute story underpinning a post-$60 billion valuation.

Keep Reading

SpaceX & xAI to Power Anthropic's Claude in Surprise Compute Pact

SpaceX & xAI to Power Anthropic's Claude in Surprise Compute Pact

In one of the most unexpected infrastructure partnerships of the current AI cycle, SpaceX and xAI — both controlled by Elon Musk — have agreed to provide compute infrastructure to Anthropic, the Amazon- and Google-backed AI safety company behind the Claude family of models. The deal, confirmed by sources familiar with the arrangement, underscores just how severe the compute bottleneck has become: even a well-capitalized frontier lab flush with hyperscaler backing is turning to a competitor's hardware empire to keep its models running at scale.

Read full story →

Stay Ahead of the Market

Daily AI & crypto briefings — straight to your inbox, your phone, and your timeline.