Three Days After Closing, Cursor Shipped Code Hosting

The SpaceX acquisition of Anysphere — the company behind Cursor — legally closed on August 14, 2026, per a Form 8-K filed by Space Exploration Technologies Corp. the same day. Three days later, on August 17, Cursor’s changelog announced Origin code hosting “rolling out today in early beta on all paid plans.” Two dated events, two separate primary sources, and a three-day gap between them.

This post keeps them separate, because the separation is the useful part. The close is a legal event with exact, checkable mechanics — share counts, an Effective Time, an implied equity value, a pricing formula. The Origin beta is a product event with its own scope and its own fine print. What sits between them is the question that actually matters to a team whose IDE vendor just became a subsidiary of a launch-and-AI conglomerate: what changed in the documents you rely on, and what did not.

So this is a buyer-side read, argued only from what is documented: the SEC filing itself, Cursor’s own legal and trust pages with their date stamps, and the changelog. Where a question has no documented answer, we say so instead of speculating about intent.

Get the dates right first. The merger agreement was signed on June 16, 2026 — a fact the August filing restates in its opening sentence. The merger’s “Effective Time,” the filing’s own defined term for the moment the deal legally completed, was August 14, 2026, a Friday. And Origin’s early beta began rolling out on August 17, a Monday, per Cursor’s own changelog. The close and the beta are three days apart — one weekend, in practice — and they live on two entirely separate primary documents.

We covered the announcement, the deal economics, and the rationale when the agreement was signed — that analysis lives in our June breakdown of the $60B deal and is not re-told here. What is new as of this week is the completion itself, and the product release Cursor’s changelog dates to the Monday after the Effective Time: Origin.

On Origin itself, precision matters as much as it does on the filing. What shipped is repos, pull requests, code browsing, and two-way GitHub sync through a new Codebase tab — and Cursor’s own release notes state plainly that “pushes keep going to GitHub, which stays the source of truth for anything started there.” No separate price was disclosed; the beta is bundled into existing paid plans. The full teardown of what shipped, what did not, and what the GitHub-sync mechanics mean in practice is in our companion post on the Origin beta — this post stays on the ownership question.

One more timing observation, stated carefully. A beta that ships three days after a close was not built in three days. Origin was first shown publicly as a waitlisted concept in mid-June 2026 — around the same time the merger agreement was signed, per secondary coverage from the time. So the Monday release tells you the pre-close roadmap kept moving through the deal, which is itself informative: it is evidence of product continuity through the ownership change, not evidence of the new owner’s direction. What the new owner changes, if anything, will show up later — in documents, which is where the rest of this post looks.

The primary legal record of the completion is a Form 8-K filed by Space Exploration Technologies Corp. on August 14, 2026, signed by its CFO. Item 2.01 carries the mechanics, and every line of the disclosure is checkable. The table below reduces the filing to a single ledger — the numbers are the filing’s own, quoted at its stated precision.

The question “so does xAI own Cursor now?” deserves one precise answer. The registrant and legal acquirer in the filing is Space Exploration Technologies Corp., a Texas corporation; the merger vehicle was X67 Inc., a wholly owned SpaceX subsidiary; and no entity named “xAI” appears anywhere in the filing’s text. Separately and earlier, SpaceX and xAI merged into one company on February 2, 2026, and the combined AI and compute arm is publicly branded SpaceXAI. So “xAI now owns Cursor” is a defensible plain-English description of the corporate parent — but it is not the filing’s language, and this post uses the filing’s names for anything sourced to the filing.

Cursor’s own framing, published on the day of the close, is simpler: “Cursor has officially been acquired by SpaceX.” The same post dates the relationship to April 2026, “when we announced our partnership with SpaceXAI to accelerate our model training efforts” — an earlier, separate milestone that predates the June 16 merger agreement. Cursor’s post gives no specific day in April, so neither do we. That post also ties one shipped artifact to the new relationship: Grok 4.6 — released the Wednesday before the close, August 12 — which Cursor calls “an early look at what we can now build together.” The entanglement between Grok distribution and Cursor’s account plumbing predates the close, too; we documented it in our August 11 post on Grok Bot’s Cursor-tier gating.

The rest of Cursor’s day-of-close post is vendor statement, and should be read as exactly that. “We will have access to the largest fleet of GPUs in the world, giving us the compute to build stronger models that are also more economical to run” is a claim about capacity; the sentence that follows — “This means we can provide customers with more capable models at lower cost” — is a stated intention about future pricing and capability, not a demonstrated outcome. Nothing in the documents reviewed for this post evidences either result yet. Hold both claims to the same standard you would hold any vendor’s post-acquisition messaging: check the terms and the prices later, not the launch-day prose.

Here is what the date stamps on Cursor’s own legal pages show. Cursor’s Master Services Agreement — the commercial contract between Cursor and its business customers — is dated “Last Updated: August 13, 2026,” one calendar day before the merger’s Effective Time. Its Data Processing Addendum is dated November 5, 2025 — over nine months before the close. Its separate Data Use policy is dated July 15, 2026 — after the June 16 signing, about a month before the close. Those are the facts; the documents say nothing about why any of them carries the date it does, and neither do we.

The clause that governs what just happened sits in MSA §14(f). Ordinary assignments of the contract require prior written consent from both sides — but there is a carve-out for exactly the event of August 14:

Read as a buyer: the new owner did not need your consent to take over your Cursor contract, provided the successor is not “a direct competitor” of yours. The clause does not define “direct competitor,” and no document reviewed for this post clarifies how that term would be interpreted — worth noting for any customer whose own business now overlaps with a launch-and-AI conglomerate’s portfolio.

The asymmetry is the story. The commercial document that governs who holds the contract was refreshed the day before the ownership change. The two documents that govern what happens to your code and data — the DPA and the Data Use policy — were last touched long before it, and neither contains any change-of-control language as retrieved. That is not necessarily a problem: it means the data terms you agreed to are, on paper, exactly the terms you have today. But it also means the ownership change happened entirely outside the data documents, which is precisely the scenario a procurement checklist should have contemplated. How Cursor’s data terms compare to the rest of the coding-agent field on this axis is the subject of our data-terms census published the same day as this post.

The most common buyer worry after an acquisition like this is model exclusivity: does the parent’s model displace the rest? That question has a directly documented, checkable answer — Cursor’s own Trust Center subprocessor list. As retrieved at the time of writing, it names SpaceXAI as an inference provider alongside — not in place of — OpenAI, Anthropic, and Google Gemini, all listed as active inference-provider subprocessors.

Two caveats keep this honest. First, a subprocessor list is a snapshot, not a commitment — it documents today’s suppliers, not tomorrow’s roadmap. Second, the DPA itself defines how that list can change: 30 business days’ advance notice before adding or replacing a subprocessor, during which a customer may object on documented grounds, with urgent cases (such as security incidents) proceeding on 10 days’ retroactive notice instead. That notice window is the buyer’s actual instrument here. If model plurality matters to your team, the operational move is not to trust the snapshot — it is to make sure someone is subscribed to, and actually reads, the subprocessor-change notifications.

This is the matrix we would want handed to us if our IDE vendor changed owners over a weekend. Every “documented” cell traces to a primary Cursor or SEC document with a date; every “open” cell is a genuine gap in the record as reviewed for this post — not a prediction. It is deal-specific by design; for the general signing-stage version of these questions, see our AI procurement checklist.

None of the above argues for panic, and none of it argues for complacency. It argues for turning a news event into a monitoring posture. Four concrete moves, in rough priority order:

The concentration point deserves the extra paragraph, because it is the durable one. The trend this deal continues is vertical consolidation of the developer toolchain: model, editor, agent, and now hosting, converging under single owners across the industry. Each convergence is individually convenient and collectively risky — the failure modes correlate, the contract terms travel together, and the exit cost compounds quietly. The playbook for deliberately maintaining a second source — including when it is not worth the overhead — is in our vendor-resilience playbook. If your team wants help turning this into an actual governance posture — vendor inventories, data-term reviews, concentration ceilings, and the eval work to keep a second source genuinely warm — that is exactly the shape of our AI transformation engagements.

Looking forward, the checkable tells are already enumerated. Watch the DPA’s date stamp — a post-close revision to the data terms would be the first substantive governance change, and it would arrive as an edit to a page dated November 5, 2025. Watch the subprocessor list and its 30-business-day notice window — a removal of a competing model vendor would be the first substantive product-direction change. And watch the pricing page — the vendor’s stated intention of “more capable models at lower cost” is a testable claim, and its test is a price table, not a blog post. Any of those changes would justify a follow-up; none of them has happened as of publication.