--- title: "Anthropic's Cursor Rescue Has a $15 Billion Asterisk" date: 2026-08-31 tags: ["anthropic","cursor","openai","spacex","xai","industry","claude-code"] categories: ["Industry"] summary: "Within two hours of OpenAI announcing it would cut Cursor off over its new SpaceX ownership, Anthropic co-founder Tom Brown publicly pledged to increase Claude compute for Cursor instead. Anthropic has cut off Windsurf and blocked xAI engineers under near-identical circumstances before — and it happens to be paying SpaceXAI $1.25 billion a month for the compute behind that pledge." --- ![Anthropic's Cursor Rescue Has a $15 Billion Asterisk](/images/anthropic-cursor-rescue-spacex-colossus-self-interest.png) On August 28, [OpenAI told Cursor it was cutting off model access](https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/), effective November 12, because Cursor's new owner is SpaceX and OpenAI doesn't trust Elon Musk's companies to honor a contract. This blog [covered that decision the same day it broke](/posts/openai-cuts-cursor-off-after-spacex-acquisition/), including CEO Michael Truell's response that OpenAI models account for only about 5% of Cursor's traffic anyway. What happened next didn't make it into that piece, because it happened almost two hours after OpenAI's post went up: Anthropic co-founder and Chief Compute Officer Tom Brown [posted on X](https://x.com/NotTomBrown/status/2093541294027280657) that "Cursor has been a trusted partner of Anthropic since Sonnet 3.5. We'll continue to increase compute to support Claude models in Cursor and are excited for what comes next with them at SpaceX." Multiple outlets — [CNBC](https://www.cnbc.com/2026/08/29/openai-cursor-spacex-model-access.html), [wccftech](https://wccftech.com/anthropic-pounces-as-openai-abandons-spacexs-cursor-vowing-to-increase-claude-compute-even-as-openai-cites-contract-distrust/), [Cryptopolitan](https://www.cryptopolitan.com/anthropic-keeps-cursor-openai-pulls-out/) — ran near-identical versions of the same story: Anthropic swoops in as the reliable partner exactly where OpenAI just proved unreliable. It's a good story. It's also missing a fairly important piece of context: Anthropic has done exactly what OpenAI just did, twice, under nearly identical circumstances — and it happens to be collecting $1.25 billion a month from the same company it's now publicly vowing loyalty to. ## Anthropic's own cutoff history Go back to June 2025. Windsurf CEO Varun Mohan disclosed that Anthropic had limited Windsurf's direct access to Claude 3.5 and 3.7 Sonnet with less than five days' notice. The timing wasn't subtle — it landed right as Bloomberg reported OpenAI was in talks to acquire Windsurf for $3 billion. Anthropic's Jared Kaplan framed it publicly as capacity triage, telling [TechCrunch](https://techcrunch.com/2025/06/05/anthropic-co-founder-on-cutting-access-to-windsurf-it-would-be-odd-for-us-to-sell-claude-to-openai) it would be "odd for us to sell Claude to OpenAI." Translation: once your acquirer is a direct competitor, the compute stops flowing. Windsurf pivoted to third-party inference and was ultimately picked apart between Google (a $2.4B licensing deal) and Cognition. Then, in January 2026, xAI staff working inside Cursor found Claude models had gone silent on them. [Reporting at the time](https://sherwood.news/tech/report-anthropic-cuts-off-xais-access-to-its-models-for-coding/) traced it to Anthropic enforcing its usage terms — which, notably, are broader than OpenAI's. OpenAI's restriction is about not training competing models on its outputs. Anthropic's bars customers from building "any products or services that compete with our Services" at all, a policy multiple outlets have flagged as the most restrictive among the frontier labs. xAI cofounder Tony Wu had to tell his own staff internally that Anthropic models simply weren't responding. So the pattern, laid out plainly: Anthropic cut Windsurf off the moment it smelled a competitor acquisition. Anthropic cut xAI engineers off the moment their usage looked competitively inconvenient. And now, weeks after SpaceX formally absorbed both xAI and Cursor into one Musk-controlled entity, Anthropic is the one publicly promising to lean in harder. Same acquirer-is-now-a-rival logic that drove the Windsurf and xAI decisions. Opposite outcome. ## The part everyone's skipping: Anthropic pays SpaceX $15 billion a year Here's why. In May 2026, [Anthropic signed a deal](https://www.anthropic.com/news/higher-limits-spacex) for the entirety of SpaceXAI's Colossus 1 capacity — over 220,000 GPUs across 300+ megawatts — specifically to expand headroom for Claude Pro and Max subscribers. [TechCrunch reported the terms](https://techcrunch.com/2026/05/20/anthropic-will-pay-xai-1-25-billion-per-month/) at $1.25 billion per month through May 2029, a contract that could deliver SpaceXAI more than $40 billion in total revenue. This blog wrote about the arrangement at the time as a "[dual bet](/posts/spacex-cursor-60b-anthropic-colossus-dual-bet/)" — SpaceX simultaneously holding acquisition optionality on Cursor and supplying the compute Anthropic's own product runs on, betting on both sides of the IDE-vs-terminal-native fight at once. That prediction is now cashing out in a very specific way. Anthropic isn't just SpaceX's landlord's tenant anymore — it's dependent on SpaceX for a meaningful chunk of the compute capacity behind Claude.ai itself, mid-contract, for the next three years. Publicly picking a fight with the company you're paying $15 billion a year, right after it just closed a $60 billion acquisition of your rival's former partner, is not a move any vendor makes lightly. Being the generous, reliable model supplier to SpaceX's newly acquired coding tool costs Anthropic nothing and buys goodwill with a landlord it can't easily replace. It would be a genuinely strange business decision for Tom Brown's team to have done anything else. None of this means Brown's statement was insincere, and it doesn't erase the real difference in tone: Anthropic chose warmth where OpenAI chose distrust. But "trusted partner since Sonnet 3.5" is doing some quiet work here too — it's the kind of language a vendor reaches for when the actual reason is closer to "we cannot afford the alternative." ## The lesson isn't about who's nicer It's tempting to read this as Anthropic simply being the more reliable partner in the AI coding market — and Anthropic will happily let that framing stand uncorrected. But the more useful takeaway is structural, not moral: every model-supply relationship in this market, including yours with Anthropic if you're building on the API rather than Claude Code itself, is commercially contingent. Windsurf found that out with five days' notice. xAI's engineers found it out mid-session. Cursor is finding out right now, from OpenAI, that a change of ownership can end an API contract regardless of how much traffic actually depends on it. Claude Code doesn't carry this specific risk — not because Anthropic is more virtuous about third-party relationships (the Windsurf and xAI episodes argue the opposite), but because Claude Code isn't a third party. There's no reseller contract between Anthropic and itself to terminate over a change of control. That structural difference — the same one this blog has argued makes Claude Code's terminal-native approach more resilient than IDE wrappers reselling access to models they don't own — holds up here too. It just doesn't mean the company behind it is playing a different game than everyone else when it's the one holding the leverage. --- **Sources:** - [Our decision on Cursor following its acquisition by SpaceX — OpenAI](https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/) - [Tom Brown (@NotTomBrown) on X](https://x.com/NotTomBrown/status/2093541294027280657) - [Anthropic Pounces As OpenAI Abandons SpaceX's Cursor — wccftech](https://wccftech.com/anthropic-pounces-as-openai-abandons-spacexs-cursor-vowing-to-increase-claude-compute-even-as-openai-cites-contract-distrust/) - [Anthropic keeps Claude in Cursor as OpenAI pulls out — Cryptopolitan](https://www.cryptopolitan.com/anthropic-keeps-cursor-openai-pulls-out/) - [OpenAI to end model access to Cursor after acquisition by SpaceX — CNBC](https://www.cnbc.com/2026/08/29/openai-cursor-spacex-model-access.html) - [Anthropic co-founder on cutting access to Windsurf — TechCrunch](https://techcrunch.com/2025/06/05/anthropic-co-founder-on-cutting-access-to-windsurf-it-would-be-odd-for-us-to-sell-claude-to-openai) - [Report: Anthropic cuts off xAI's access to its models for coding — Sherwood News](https://sherwood.news/tech/report-anthropic-cuts-off-xais-access-to-its-models-for-coding/) - [Higher usage limits for Claude and a compute deal with SpaceX — Anthropic](https://www.anthropic.com/news/higher-limits-spacex) - [Anthropic will pay xAI $1.25 billion per month for compute — TechCrunch](https://techcrunch.com/2026/05/20/anthropic-will-pay-xai-1-25-billion-per-month/) - This blog's original coverage: [OpenAI Just Cut Cursor Off](/posts/openai-cuts-cursor-off-after-spacex-acquisition/), [SpaceX Is Betting $60B on Cursor and $300MW on Anthropic](/posts/spacex-cursor-60b-anthropic-colossus-dual-bet/)