Anthropic Rents Colossus 1 – The Surprising Compute Deal with SpaceX/xAI

On May 6, 2026, Anthropic announced a deal at its Developer Day in San Francisco that few had anticipated: the company is renting the entire compute capacity of SpaceX's Colossus 1 data center in Memphis, Tennessee – from a company that until recently was considered a direct adversary.

What Happened

Anthropic has signed an agreement with SpaceX to use all compute capacity at the Colossus 1 data center in Memphis – over 300 megawatts and more than 220,000 NVIDIA GPUs, available within a month.¹ Colossus 1 features dense deployments of H100, H200, and GB200 accelerators and is considered one of the world's largest and fastest-deployed AI supercomputers.²

Why Colossus 1 Was Available

Musk explained he was comfortable leasing Colossus 1 to Anthropic because SpaceXAI had already moved its training entirely to Colossus 2.³ Colossus 2 came online in January 2026, with further data centers in Southaven and on adjacent parcels in planning or under construction.⁴ Anthropic is not renting strategically active infrastructure – it's renting capacity that xAI had already internally superseded.

A Note on "All" the Capacity

Both official statements – from Anthropic and SpaceXAI – use the phrase "all of the compute capacity." However, xAI had already announced in April that Colossus 1 capacity would also be used by Cursor, the AI code editor that SpaceX has agreed to acquire for up to $60 billion. The number of GPUs used by Cursor is not publicly known.⁴ Whether Cursor's allocation was moved before the Anthropic deal or runs in parallel cannot be conclusively determined from public sources.

The Problem Being Solved

Anthropic had acknowledged last month that demand for Claude had led to "inevitable strain on our infrastructure," impacting reliability and performance – especially during peak hours.⁵ The context: Dario Amodei revealed at the Developer Day that Anthropic grew 80x in Q1 2026 on an annualized basis – against an internal projection of 10x. Annualized revenue has crossed $30 billion, up from $9 billion at end of 2025.⁶ This growth trajectory explains why new capacity was urgently needed – and why Anthropic couldn't wait for its longer-term cloud deals to come online.

Immediate Changes for Users

Three changes effective immediately were tied to the deal:¹

• Claude Code's five-hour rate limits doubled for Pro, Max, Team, and Enterprise
• Peak-hours limit reductions on Claude Code removed for Pro and Max accounts
• API rate limits for Claude Opus models raised significantly

Context: Colossus 1 as a Fast Bridge

The SpaceX deal is part of a broad compute procurement push – but with one key distinguishing feature: it delivers capacity within a month, while the larger structural deals take much longer. Running in parallel:¹

• An up-to-5 GW agreement with Amazon (nearly 1 GW by end of 2026)
• A 5 GW agreement with Google and Broadcom (online from 2027)
• A Microsoft-NVIDIA strategic partnership covering $30 billion of Azure capacity
• A $50 billion investment in US AI infrastructure with Fluidstack

Colossus 1 is a tactical capacity boost – the structurally large volumes come later.

Geopolitics and an Unlikely Pairing

Musk had called Anthropic a company that "hates Western civilization" in February and publicly asked whether there was "a more hypocritical company than Anthropic."⁵ Now he's a supplier. After reportedly spending the past week with senior Anthropic staff, he posted on X that he was "impressed."

The timing is hard to ignore: SpaceX filed confidentially with the SEC in April for an IPO targeting a valuation of $1.75 to $2 trillion – with the public S-1 expected by late May and the roadshow set for the week of June 8.⁷ Anthropic as a named compute customer strengthens SpaceX's pitch as an AI infrastructure provider, with AI infrastructure now a disclosed revenue line in the prospectus.

At the same time, Anthropic is engaged in litigation with the US government: the Pentagon declared Anthropic a supply chain risk and barred it from military contracts. Anthropic sued, and the case remains open.⁵ While Anthropic buys compute from SpaceX, it is simultaneously fighting for the right to do business with the US government at all.

International Expansion

Enterprise customers in regulated industries like financial services, healthcare, and government increasingly require regional infrastructure. The Amazon partnership therefore includes additional inference capacity in Asia and Europe.¹

Outlook: Orbital Compute

SpaceX filed with the FCC in January for approval to deploy a one-million-satellite orbital AI data center constellation – a project that still faces significant hurdles.⁴ As part of the agreement, Anthropic expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity.¹ No concrete plans or timelines have been published – this remains a letter of intent.

What Nobody Says

The Musk kill-switch. Musk explicitly reserved the right to reclaim compute capacity if Anthropic's AI "engages in actions that harm humanity."³ This is not a legal footnote – it's a unilateral kill-switch clause in an infrastructure contract with a direct competitor. Who decides what "harms humanity"?

The utilization question. According to The Information, xAI's GPU utilization stood at around 11 percent – far below the 40 percent achieved by rivals.⁴ Leasing Colossus 1 to Anthropic is therefore also simply a monetization of underutilized hardware. Financial terms of the deal were not disclosed.

The Cursor ambiguity. Until SpaceX makes an explicit exclusivity statement, "all of the compute capacity" should be read as a commercial formulation – not a technically verifiable guarantee.