DEV Community

Cover image for SpaceX Buys Cursor for $60B to Build the World's Most Useful AI Models
Dave Kurian
Dave Kurian

Posted on • Originally published at otf-kit.dev

SpaceX Buys Cursor for $60B to Build the World's Most Useful AI Models

$60 billion for an editor. Read that again.

Yes, SpaceX just paid sixty billion dollars for Cursor. All stock. Largest software acquisition ever recorded. Take a second on that sentence, because the next one is the part that actually matters: at $60B, SpaceX is paying 20–30× Cursor's current revenue. That's not a price paid for what Cursor is today — that's a price paid for what the market believes Cursor will be at the end of 2026, when annualized revenue hits $6B (up from $2B today). The model layer just got repriced. Hard. Here's why that matters to anyone writing code on a Tuesday morning.

The receipt

The numbers, straight from the official filing:

  • $60B, all-stock. Every Cursor share converts to SpaceX Class A common at closing, at the volume-weighted average closing price of SPCX at the time. Cursor shareholders aren't cashing out — they're becoming SpaceX shareholders.
  • 3.4% dilution at SpaceX's IPO valuation. Big, but not crushing for a company that raised $75B in its IPO four days ago and is trading well above offer.
  • Cursor today: 1M+ paying customers, $2B ARR, built by Anysphere.
  • Cursor by EOY 2026, per the deal's own projected trajectory: $6B ARR.
  • Close: Q3 2026, subject to regulatory approval.

The comps are almost unfair to write. Microsoft bought GitHub for $7.5B in 2018, Salesforce paid $27.7B for Slack in 2021, Activision went to Microsoft for $69B. A software deal isn't supposed to look like this. SpaceX isn't paying for code-editor market share; it's paying for share of the next decade of model-and-software co-evolution — and the market put a price on it accordingly.

The thing that already started

Here's the part the headline buries. Per the announcement, SpaceXAI has been jointly training a model with Cursor for the past several months. That model is expected to ship inside both Cursor and Grok Build in the near term. This isn't a buy-and-figure-it-out-later deal. The integration runway is already in flight.

That's the actual story. Two teams converged on the same product bet before the corporate paperwork caught up. SpaceXAI wants to build "the world's most useful AI models." Cursor has the funnel where every model gets stress-tested by a paying developer within seconds of release. The 20–30× multiple is the market's price for that loop running at $6B ARR instead of $2B.

How to actually use Cursor today

This is the part most "Cursor coverage" skips. If you haven't tried it — and 1M people have, so statistically you probably have — here's the fastest way in:

# macOS / Linux
curl  -L | bash
Enter fullscreen mode Exit fullscreen mode

The default model is good. The move that changes everything is pointing Cursor at OpenRouter so you can swap models per-task without leaving the editor:

// ~/.cursor/settings.json
{
  "cursor.ai.primaryModel":    "openrouter/anthropic/claude-sonnet-4.5",
  "cursor.ai.fallbackModel":   "openrouter/minimax/minimax-m3",
  "openrouter.apiKey":         "${OPENROUTER_API_KEY}"
}
Enter fullscreen mode Exit fullscreen mode

Now Cmd-K opens a prompt whose base URL is yours. Frontier one moment, cheap one the next. The 1M paying customers aren't paying for tab completion — they're paying for the loop where Cursor reads the whole repo, proposes a diff, runs the tests, fixes the tests, and lands the patch.

That's the loop SpaceX is buying. Not the editor.

What the 20–30× signal means for builders

When the public market pays 20–30× revenue for a coding tool, it says one of two things. Either it's wrong and in 18 months Cursor trades at half that multiple — and we're writing the autopsy. Or it's right, and the price of AI-native software just re-rated above every prior SaaS benchmark in history.

Honest read: in 2026, AI-native tools earn a premium multiple when they own a workflow loop that compounds with the model. Cursor owns the loop from intent → diff → test run → landed PR. The moat is the user-trained completion corpus, the deep IDE hooks, and the paying-developer feedback loop that ships every improvement back into a billion-token training set. If you're building an AI tool, that's the line you have to clear — not "calls a model," but a loop that gets tighter the more paying users run it.

The part that doesn't move when the model does

Here's where it lands for the rest of us.

A year ago the AI-editor story was Cursor vs Copilot. Then Claude Code showed up and ate the terminal-and-prompt crowd. Then Lovable, Bolt, Rork, v0 took the no-code side of the same wave. Now SpaceX is putting $60B behind the loop. The model layer moves every quarter — the agents, the editors, the harnesses, the base URLs. Some ship 15× faster than their peers, then a peer ships 16× faster the next month. The harness is, by construction, disposable.

What compounds across that churn is what the loop writes into. Your components. Your shipping surface. Whether the button the agent just shipped renders the same on web, iOS, and Android without you pasting a third adapter file. That's the layer where a year of agent commits either holds up or doesn't.

[[COMPARE: the churning AI tool layer vs the durable component layer]]

The honest framing isn't "Cursor vs a cross-platform kit." It's: use Cursor — it's the best agent loop on the market and $60B just said so out loud — and pair it with a component layer where the AI's diff doesn't break three platforms at once. One API, same component, same behavior on web, iOS, Android. That's the bit that pays you back whether tomorrow's editor is Cursor, Claude Code, or something none of us have heard of yet.

What this gets us

A few things, concretely:

  • Cheaper frontier access, likely. A SpaceX-Cursor combo has every incentive to subsidize model access inside Cursor's editor to keep the loop competitive. Watch for Sonnet- and M3-class pricing inside Cursor dropping hard over the next two quarters. (Listed as likely, not confirmed — no terms have been published yet.)
  • Real co-trained tools. The joint model shipping in Cursor and Grok Build is the first frontier model trained on the inside of a paying-developer feedback loop. Expect it to be noticeably better at long-horizon coding tasks than models trained on scraped GitHub alone — that's the bet, at least.
  • A repriced market. 20–30× ARR for AI-native tools sets the floor for the next 18 months of funding rounds. Smaller AI-coding startups are now valued against this comp, whether their loops deserve it or not.
  • A reminder that loops beat wrappers. The Cursor story is "a loop that gets tighter with usage." If your AI tool doesn't have one, the 20–30× multiple was never going to be yours.

What to watch

Three signals will tell you whether the 20–30× bet holds.

One: what ships inside Cursor and Grok Build within 60 days of closing. If it's a clearly better coding model than the public frontier, the multiple was right.

Two: regulatory posture. A $60B all-stock deal in a sector the FTC has already come after GitHub and Activision for won't close quietly. Watch for remedies, not refusal.

Three: the cohort. When the deal closes, every Cursor employee is now a SpaceX shareholder with vested upside tied to SPCX. Retention is the real risk — and the first thing SpaceX should solve, because the loop is the asset.

Deal closes Q3 2026. By then the next Cursor competitor — probably already incubated inside SpaceXAI — will have shipped something. The model layer keeps moving. Build on something that doesn't.

Top comments (0)