SpaceX is in an “enviable position”: why Anthropic is sticking with Cursor as OpenAI cuts access
The New Stack Paul Sawers ● Covered by 2 sources
OpenAI says it’ll cut Cursor off from its models in November over Musk-linked contract disputes. Anthropic is doing the opposite — and that may say more about compute leverage than loyalty.
Based on reporting by The New Stack, Paul Sawers — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI kicked off the weekend with a sharp move: it said Cursor will lose direct access to OpenAI models in November. The company tied the decision to Elon Musk’s businesses, citing Twitter’s breach of a data-licensing deal after Musk’s 2022 takeover and Musk’s sworn admission earlier this year that xAI had partly used OpenAI models through distillation. OpenAI said it no longer trusted SpaceX to stay within its terms of service.
What stood out was the reply from Anthropic. Tom Brown, one of the company’s co-founders and its chief compute officer, moved quickly on X to say Cursor remains a “trusted partner” and that Anthropic would keep increasing compute to support Claude in Cursor. That is a very different tone from the one Anthropic used when Windsurf started looking like an OpenAI target.
In May 2025, reports said OpenAI was in talks to buy Windsurf for $3 billion. Anthropic did not wait for anything to close. Within weeks, Windsurf said Anthropic had cut off direct access to Claude 3.5 Sonnet and Claude 3.7 Sonnet. Jared Kaplan later said it would be “odd” for Anthropic to be selling Claude to OpenAI. The OpenAI deal fell apart anyway, and Google later paid $2.4 billion to bring Windsurf’s founders and R&D staff into DeepMind.
That history is why the Cursor response raised eyebrows. Gergely Orosz and Replit CEO Amjad Masad both pointed to the contrast: Windsurf got cut off when it only looked like an OpenAI target, while Cursor is getting public support after SpaceX actually bought it. The difference, they argued, is compute.
And the numbers back that up. On May 6, Anthropic said it had secured the entire output of SpaceX’s Colossus 1 data center near Memphis, Tennessee — more than 300 megawatts and over 220,000 Nvidia GPUs. Two weeks later, SpaceX’s IPO filing put the deal at $1.25 billion a month across Colossus and Colossus II, with about 325,000 Nvidia GPUs combined, running through May 2029 unless terminated. SpaceX said the setup helps it monetize spare capacity while keeping enough for its own needs.
That leaves SpaceX in a very unusual spot. It now owns xAI and Cursor, which puts it against Anthropic in both foundation models and coding tools. But Anthropic is also paying SpaceX billions for compute. So when Anthropic calls Cursor a trusted partner, that is not just a friendly message. It is a company speaking carefully to a supplier it does not want to upset.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI that keeps getting dressed up as principle and is really about infrastructure. If a lab depends on your GPUs, your “policy” suddenly gets a lot more flexible. The industry keeps pretending model wars are about safety and loyalty; mostly they’re about who owns the pipes.
Read more about this at: The New Stack
