Anthropic Signs $40B Compute Deal With xAI Through 2029
Key insights
- Anthropic is paying xAI $1.25B monthly for Colossus 1 access plus Colossus 2 rights, totaling over $40B through May 2029.
- xAI is monetizing its training infrastructure as a neocloud, with a 90-day mutual exit clause as the only off-ramp.
- The deal was disclosed in SpaceX's S-1 IPO filing, not by Anthropic, making it one of the largest surprise infrastructure revelations in AI history.
Why this matters
For AI founders and infrastructure strategists, this deal signals that frontier compute access has consolidated to a degree where even a $30B-valued safety-focused lab must route billions through a competitor's data centers, compressing the field's independence. The supplier relationship between Anthropic and xAI creates a structural conflict of interest that regulators, investors, and enterprise customers will scrutinize, particularly given Musk's public antagonism toward Anthropic's leadership and mission. At $15B annually, Anthropic's compute spend rivals or exceeds the entire AI infrastructure budgets of most sovereign nations, setting a new floor for what serious frontier model development costs and who can realistically compete.
Summary
Anthropic has committed to paying xAI $1.25 billion per month for access to Musk's Colossus data center infrastructure through May 2029, a contract disclosed in SpaceX's S-1 filing that totals more than $40 billion and ranks among the largest compute agreements ever made public.
The deal gives Anthropic exclusive access to the Colossus 1 facility in Memphis, Tennessee, plus rights to capacity at the forthcoming Colossus 2 expansion. xAI operates the arrangement as a neocloud play, monetizing spare training infrastructure it built for its own Grok model work. Either party can exit with 90 days' notice, which is the only structural escape valve in an otherwise long-horizon commitment.
Essentially: (Anthropic, xAI) have locked their infrastructure strategies together despite being direct competitors in the frontier model market.
- Anthropic confirmed the Colossus 2 expansion rights minutes before the SpaceX S-1 became public, suggesting coordinated disclosure.
- At $1.25B per month, Anthropic is committing roughly $15 billion annually to a single compute supplier controlled by Elon Musk.
- The deal surfaced through a third party's IPO filing, not Anthropic's own communications, raising questions about how much of its infrastructure posture remains undisclosed.
The contract reframes the AI compute race as a supplier-dependency problem as much as a model-capability one.
Potential risks and opportunities
Risks
- Anthropic's safety mission and model roadmap could be operationally disrupted if Musk invokes the 90-day exit clause as leverage during a competitive or regulatory dispute between the companies.
- Enterprise and government customers with Anthropic contracts, particularly those with data-sovereignty requirements, may face compliance exposure if Colossus facility terms give xAI any access or visibility into workloads.
- If xAI's Colossus 2 buildout is delayed or underdelivers on capacity, Anthropic has no disclosed fallback supplier at the scale needed, potentially stalling its next-generation model training timeline into late 2026 or beyond.
Opportunities
- Hyperscalers with spare GPU capacity at scale (CoreWeave, Lambda Labs, Oracle Cloud Infrastructure) gain significant negotiating leverage with Anthropic and other frontier labs now visibly exposed to single-supplier concentration risk.
- Anthropic competitors including Google DeepMind and Meta AI can use this disclosure in enterprise sales cycles to highlight Anthropic's infrastructure dependency on a rival's hardware as a reliability and conflict-of-interest concern.
- Sovereign wealth funds and governments pursuing domestic AI infrastructure (UAE's G42, Saudi Aramco's Humain, the UK's AI Energy Taskforce) can point to this deal as evidence that frontier labs will pay premium rates for committed compute capacity, improving the business case for state-backed data center buildouts.
What we don't know yet
- Whether Anthropic has negotiated any pricing protections or capacity guarantees if xAI prioritizes Grok training runs over contracted Colossus access during peak demand periods.
- What due diligence Anthropic's board and investors conducted on counter-party risk given Musk's history of unilaterally altering or exiting commercial agreements.
- Whether the 90-day exit clause is practically exercisable, given that Anthropic would need an alternative multi-gigawatt compute destination secured before it could realistically trigger termination.
Originally reported by techcrunch.com
Read the original article →Original headline: Anthropic Will Pay xAI $1.25 Billion Per Month for Compute Through 2029 — $40B+ Deal Disclosed in SpaceX S-1