Cursor Is Closed — SpaceX Owns Your Model Router Now

SpaceX closed its $60 billion all-stock acquisition of Anysphere on August 14, confirmed by SEC Form 8-K. Cursor shares converted into 389,289,254 SpaceX Class A shares. The startup now operates inside a new SpaceXAI division — the same organizational unit that builds Grok, runs Colossus, and had its Claude API access revoked by Anthropic seven months ago. Three days after the close, Cursor shipped Origin, a GitHub competitor, to all paid plans. No published data retention policy. No subprocessor disclosures. The speculative risk I wrote about in June is now operational risk sitting inside your CI pipeline.

TL;DR

• Closed: SpaceX completed the Cursor acquisition August 14; Cursor is a wholly owned SpaceXAI subsidiary
• Grok 4.6 shipped to Cursor August 12 — two days before close, $2/$6 per million tokens, 500K context
• Origin launched August 17 — code hosting, PRs, GitHub sync, zero published data terms
• Action: Audit your model routing dependency now — Claude access through Cursor Router is a closed internal decision you do not control

SpaceX Owns Your Model Router — What Happened

The merger agreement was signed June 16. Eight weeks of regulatory process later, the deal closed August 14. What happened in the gap matters more than the close itself.

On August 12 — two days before regulatory clearance was even confirmed — SpaceXAI pushed Grok 4.6 live across all Cursor plans. The model ships at $2 per million input tokens and $6 per million output tokens with a 500,000-token context window. Day-one availability in Cursor, not day-one availability on the xAI API with Cursor integration coming later. Cursor got it first. That sequencing tells you everything about how product decisions flow inside SpaceXAI: the editor is the distribution channel for the model, not the other way around.

Then on August 17, three days post-close, Origin went live in early beta across Pro, Teams, and Enterprise plans. Repositories, pull requests, GitHub sync, integrations with Vercel, Depot, and Buildkite. A full code hosting platform, not a feature experiment. Cursor has not published data retention terms, training-use policies, or subprocessor disclosures for code hosted natively on Origin. Your source code is now inside a platform owned by SpaceX with no contractual clarity on what happens to it.

Origin launched with zero published data retention or residency policy. If your team enables it on an Enterprise plan, your code is hosted inside SpaceX infrastructure with no documented governance guarantees. Check before you sync.

Why This Matters

The deal announcement in June was a pricing event — Cursor’s market share on Ramp’s AI spending index had already slid from roughly 41% in mid-2025 to 26% by May 2026, reflecting developer skepticism about where this was heading. The close converts that skepticism into a structural reality: SpaceX now controls the model router for tens of thousands of engineering teams, and SpaceX builds a competing model.

The precedent is already set. In January 2026, Anthropic cut Claude API access for xAI engineers after discovering they were using Cursor to tap Claude for competitive model development work. That was Anthropic enforcing its commercial terms against xAI. Now xAI is Cursor. The entity that lost Claude access seven months ago now owns the routing layer that determines which model your Cursor instance calls by default.

I do not think SpaceX will flip a switch and kill Claude access overnight. That would cause immediate churn and hand GitHub Copilot and Claude Code a migration wave on a silver platter. The playbook is subtler. Grok 4.6 was live in Cursor before the deal even closed. Origin ships with tight SpaceX infrastructure integration and no third-party model guarantees. Reports indicate a jointly trained SpaceXAI-Cursor coding model is in development on Colossus — roughly 200,000 NVIDIA GPUs with plans to scale toward one million — though no official announcement has confirmed the project’s scope or timeline. When that model launches, the commercial incentive to make it the default is not hypothetical. It is the entire business case for a $60 billion acquisition.

The slow squeeze works like this: Grok gets top billing in the model picker. Claude stays “available” but gets a separate usage pool with a lower allocation. Latency for third-party models creeps up because routing priorities shift. The first jointly trained model gets a dedicated fast path. By the time Claude is quietly de-defaulted six months from now, most teams will not notice until they check their token bill and realize they have been running Grok for weeks.

Check whether your team accesses Claude through Cursor Router or through direct API keys. If it is Router-only, you have zero visibility into routing changes. Direct API keys give you model control independent of Cursor’s product decisions.

This matters differently depending on your stack. If you use Cursor as a smart autocomplete layer and route all agentic work through Claude Code or OpenCode with your own API keys, the acquisition changes nothing about your model access. If your entire workflow — editing, agentic coding, code hosting — runs through Cursor, you just handed a SpaceX subsidiary control over your model selection, your code storage, and your CI integration layer simultaneously.

The comparison that should concern you is not Cursor vs. GitHub Copilot. It is Cursor-under-SpaceX vs. model-agnostic alternatives like Claude Code, OpenCode, and Antigravity. Those tools let you pick your model and your hosting independently. Cursor used to offer that flexibility too. Whether it still will in twelve months is a question only SpaceX can answer, and SpaceX has not answered it.

Factory Router and Augment Code’s Prism already offer independent model routing — Factory Router claims 20–25% token cost savings with 99% of Claude Opus 4.7 performance on Terminal-Bench 2, and Prism routes across provider pools at 20–30% lower cost per task. Neither is owned by a model vendor. That structural independence is now a feature, not a footnote.

The Take

The June announcement was a signal to start planning. The August close is a deadline to finish planning. Here is the audit your team needs to run this week, not this quarter:

Which workflows are hardcoded to Cursor vs. model-agnostic? Map every place your engineering org depends on Cursor Router for model access. Any workflow that cannot switch to a direct API key within a sprint is a single point of failure controlled by SpaceX.

Does your data governance policy account for Origin? If any engineer on your team enabled Origin’s GitHub sync, your source code is inside SpaceX infrastructure with no published data retention terms. Your security team needs to know this before legal does.

What is the migration path? Claude Code, OpenCode, and Antigravity all offer model-agnostic editing or agentic coding. None of them are owned by a model vendor. If Cursor’s defaults shift toward Grok over the next two quarters, can your team move without losing a sprint?

I am not telling you to drop Cursor today. Cursor is still a strong editor with real workflow advantages. But the company that builds your editor now also builds the model it wants you to use, hosts the code it wants you to store, and controls the router it wants you to route through. That is not a tool — it is a platform with aligned incentives pointing away from model choice. Start the audit now. By the time the default model changes, the switching cost will be higher than it is today.

Cursor’s Ramp spending share dropped from 41% to 26% between mid-2025 and May 2026 — before the deal even closed. The market is already voting. The question is whether your team has voted yet or is waiting to be told.