Anthropic raises Colossus 2 GB200 capacity in $1.25B-per-month SpaceX compute deal
Anthropic said it is expanding onto Colossus 2 GB200 capacity, while a SpaceX S-1 excerpt pegged the contract at $1.25 billion per month through May 2029. The deal shows long-term chip and datacenter supply is becoming a gating factor for frontier-model deployment.
TL;DR
- Anthropic said it is scaling up on GB200 capacity in Colossus 2 throughout June, with nottombrown's post framing the move as an expanded SpaceX partnership.
- Simon Willison's SpaceX S-1 excerpt says SpaceX and Anthropic signed Cloud Services Agreements worth $1.25 billion per month through May 2029, with lower fees during the May and June 2026 ramp.
- The same SpaceX S-1 excerpt says either side can terminate on 90 days' notice, a clause that pitdesi's follow-up highlighted because the deal is large relative to SpaceX's recent revenue base.
- eliebakouch's filing summary ties the Anthropic contract to a much larger buildout, including a first Colossus II cluster of about 110,000 GB200s and a second cluster of about 110,000 GB300s.
- Complaints from bridgemindai's usage screenshot and bridgemindai's review thread suggest xAI's own Grok Build users were already feeling compute scarcity before the Anthropic expansion became public.
Anthropic's own post gives the top-line confirmation, while Simon Willison's note on the filing supplies the contract language. The interesting extra detail is in eliebakouch's inventory of the S-1 numbers, which sketches the size of Colossus II, and in bridgemindai's Grok Build review, which landed two days earlier with complaints about tight rate limits on xAI's $300 plan.
Deal terms
Anthropic confirmed the partnership expansion itself. The pricing and duration came from the SpaceX S-1 excerpt, not from Anthropic's announcement.
Quoting SpaceX S-1
We have the ability to use compute resources to support our proprietary AI applications (such as Grok 5, which is currently being trained at COLOSSUS II), while also providing access to select compute capacity to third-party customers. For example, in May 2026, we entered into Cloud Services Agreements with Anthropic PBC (“Anthropic”), an AI research and development public benefit corporation, with respect to access to compute capacity across COLOSSUS and COLOSSUS II. Pursuant to these agreements, the customer has agreed to pay us $1.25 billion per month through May 2029, with capacity ramping in May and June 2026 at a reduced fee. The agreements may be terminated by either party upon 90 days’ notice. — SpaceX S-1, highlights mine Tags: anthropic, grok, generative-ai, ai, llms
The filing language quoted by Simon Willison's SpaceX S-1 excerpt says the agreements cover capacity across Colossus and Colossus II, start ramping in May and June 2026 at a reduced fee, run through May 2029, and can be ended by either party with 90 days' notice.
Colossus 2 ramp
The new piece here is not just "Anthropic uses SpaceX compute." It is that Anthropic is moving onto GB200 capacity in Colossus 2, and doing it during the same month SpaceX says the ramp starts.
According to eliebakouch's filing summary, SpaceX described Colossus II in three chunks:
- first cluster: about 110,000 GB200s, 210 MW
- second cluster: about 110,000 GB300s, 220 MW
- next expansion: at least 220,000 additional GB300s, more than 400 MW
That makes Anthropic's June ramp look less like a one-off lease and more like a tenant inside a fast-growing external GPU utility.
Revenue concentration
The dollar figure is big enough to matter to SpaceX on its own. pitdesi's post did the back-of-the-envelope math at roughly $15 billion in next-twelve-month revenue, while the S-1 excerpt says SpaceX reported $18.7 billion in total 2025 revenue.
Because the filing excerpt also gives both parties a 90-day termination right, this is contracted revenue with a short escape hatch, not three years of locked-in backlog.
Grok Build complaints
One odd side thread is that public complaints about xAI product limits showed up before the Anthropic expansion did. bridgemindai's usage screenshot said a $300-per-month Grok Build plan burned 8% of a two-week quota in about an hour, and bridgemindai's review thread said 50 subagents crashed, rate limits hit quickly, and a Remotion job spent 25 minutes planning.
Those posts do not prove the Anthropic contract caused the limits. They do show that outside demand for Colossus capacity was becoming visible to end users at roughly the same moment SpaceX disclosed Anthropic as a multibillion-dollar compute customer.