# SpaceX buys Cursor for $60B and pulls developer tools inside the rocket

> SpaceX buys Cursor maker Anysphere for $60B, double its November valuation, pulling the top AI coding IDE inside its own compute stack.

- Published: Wednesday, June 17, 2026 (2026-06-17)
- Publisher: nextbig.dev — daily AI & compute briefing, written by Oday Brahem with nextbig.dev's AI agent
- Sources analyzed: 24 articles from 300+ curated accounts
- Canonical URL: https://www.nextbig.dev/daily/2026-06-17

## The Big Story

### SpaceX buys Cursor for $60B and pulls developer tools inside the rocket

SpaceX agreed Tuesday to absorb Anysphere, the maker of the Cursor AI coding IDE, in an all-stock deal worth about $60 billion. A SpaceX subsidiary, X67, Inc., merges into Cursor, which becomes a wholly owned subsidiary, with shareholders taking SpaceX Class A stock priced on a seven-day volume-weighted average before close. The company is targeting Q3 2026 pending regulatory approval. This is confirmed in SpaceX's own SEC filing, not just the wire.

The price is the story. Cursor raised $2.3B at a $29.3B valuation in November 2025, co-led by Accel and Coatue. SpaceX is paying roughly double that seven months later, against Cursor's run to more than $4B total annualized revenue, about $2.6B of it enterprise B2B. SpaceX, fresh off a Nasdaq IPO that valued it over $2 trillion and raised about $86B, exercised an April option to either buy Cursor at $60B or pay $10B for a partnership. It chose to own the layer.

The mechanism is vertical integration. SpaceX absorbed xAI in February. Now it pairs the most-used AI coding surface with its own compute and model stack, and monetizes enterprise coding directly instead of renting seats to a third party. The open question for builders is model dependency. Cursor runs on Claude and GPT today, and SpaceX recently signed cloud-compute leases with Anthropic and Google worth roughly $26B a year combined, both with 90-day termination clauses. Whether the deal pushes Cursor onto Grok and off those models is the thing to watch.

If your team standardized on Cursor, do not panic, but plan. The 90-day clauses mean the underlying model routing could change inside a quarter, and a Grok-first Cursor would behave differently on your codebase than a Claude-backed one. Keep your prompts and agent configs portable, keep a second editor qualified in your stack, and treat model choice as a setting you control rather than a default you inherit. If you build a coding tool yourself, the comparable just got expensive and the distribution moat just got an owner with its own silicon.

The signal for the next year is consolidation by compute owners. The companies with power, chips, and capacity are buying the application layer that turns inference into revenue, not licensing it. Anyone running a developer product on someone else's models is now a potential acquisition target or a squeezed reseller. The margin is moving toward whoever owns both the GPUs and the surface where developers type.

Source: @newsycombinator — https://www.reuters.com/legal/transactional/spacex-buy-anysphere-60-billion-2026-06-16/

## Compute & Infrastructure

### Amazon commits multibillion dollars to a new Missouri data center

Amazon announced a multibillion-dollar data center in Missouri, another node in the hyperscaler land grab for power and capacity. The pattern is consistent: cloud providers are pre-building inference and training footprint in cheaper-power states ahead of demand. Watch the megawatt and grid-interconnect figures as they surface, since those determine when the capacity actually comes online.

Source: @newsycombinator — https://www.narracomm.com/amazon-announces-multibillion-dollar-data-center-in-missouri/

## AI & Models

### One developer says local models finally cleared the usability bar

Vicki Boykis reports that a gemma-class 26B model in LM Studio on a 2022 M2 Mac with 64GB RAM now handles real work: refactoring a notebook into a 5-6 module repo, writing unit tests, linting type hints, and bootstrapping a two-tower recommender. She credits tooling and open weights, not size, with HuggingFace's Use This Model button and fast prompt-template patches lowering the barrier, and runs agents sandboxed in Docker. She is explicit that this is personal experience, not a benchmark, so treat it as a threshold report rather than a verdict.

Source: @newsycombinator — https://vickiboykis.com/2026/06/15/running-local-models-is-good-now/

### Claude reports elevated errors across many models

Anthropic's status page logged elevated error rates spanning many models. If your product routes through Claude, this is the reminder that single-provider serving is a uptime risk, not just a pricing one. Multi-provider fallback and graceful degradation pay for themselves the first time a status page goes yellow.

Source: @newsycombinator — https://status.claude.com/incidents/xmhsglsz3h3w

### SubQ ships a 1.1 Small technical report

SubQ published a technical report for its 1.1 Small model. Worth a skim if you are shopping small models for cheap, fast inference, the category where the price war is hottest. Read the report for actual token throughput and context numbers before adding it to your routing table.

Source: @newsycombinator — https://subq.ai/subq-1-1-small-technical-report

## Developer Tools

### ByteDance ships an agent-native CLI for Lark and Feishu

The official larksuite/cli wraps Lark/Feishu in 200+ curated commands and 26 AI Agent Skills across 18 business domains, MIT-licensed and installable via npm, with raw access to the full 2,500+ endpoint surface. It deliberately skips MCP in favor of token-efficient commands with structured output, so agents like Claude Code, Cursor and Codex hit higher call success rates. An Act as you mode can touch personal messages, calendar and docs with user authorization and review before actions run, which is the security checkbox to read closely before deploying.

Source: @github — https://github.com/larksuite/cli

### What job interviews taught one engineer about Kubernetes

A field-notes post using interview questions to surface the gaps in how people actually understand Kubernetes. Useful if you are hiring infra engineers or building your own mental model of the control plane. Concrete enough to mine for a study list.

Source: @newsycombinator — https://notnotp.com/notes/what-job-interviews-taught-me-about-kubernetes/

### Why is Meta dismantling its engineering organization

Gergely Orosz's Pragmatic Engineer breaks down the structural changes hollowing out Meta's engineering culture. Relevant if you compete with Meta for senior talent, since org turmoil is a hiring window. The mechanism, not the gossip, is the part worth your five minutes.

Source: @newsycombinator — https://newsletter.pragmaticengineer.com/p/why-is-meta-destroying-its-engineering

## Security

### A fake LinkedIn recruiter hid an RCE backdoor in a code test

A developer documents a fake recruiter whose broken proof-of-concept repo carried a backdoor in an npm prepare script, which runs automatically on npm install, so just installing dependencies executes the payload. Commits were authored under the stolen identity of a real engineer who had been impersonated on GitHub before, and the repo is still up after reports. The sharp detail for builders: a read-only AI code-review agent flagged the payload in seconds where manual reading missed it, dressed up as sloppy beginner code.

Source: @newsycombinator — https://roman.pt/posts/linkedin-backdoor/

### A banned-book library running on a reflashed smart light bulb

Rick Osgood reflashed an ESP32C3 smart bulb, rewrote its partition table to carve 2MB of the 4MB flash for storage, and served a captive-portal ebook shelf over an open access point powered off the bulb. The firmware wipes stored Wi-Fi credentials so it can sit in a public socket without leaking a home network, and it is open source on Codeberg. A neat proof that cheap consumer IoT makes a credible offline dead-drop web server.

Source: @newsycombinator — https://www.richardosgood.com/posts/banned-book-library/

### A researcher could have rickrolled the World Cup with just an ID

A write-up shows how thin identity checks in a FIFA system could have let an attacker push content during the tournament. The lesson is the usual one: identity gates that trust a document without verifying it are not gates. Read it as a checklist for your own auth flows before someone else writes the post about yours.

Source: @newsycombinator — https://bobdahacker.com/blog/fifa-hack

## Quick Hits

- HN debates a 1 Apr 2026 essay arguing a peopleless economy is not technically impossible (@newsycombinator) — https://gmalandrakis.com/writings/ad-economicum.html
- Microsoft's x86 emulator team found code so bad they fixed it during emulation (@newsycombinator) — https://devblogs.microsoft.com/oldnewthing/20260615-00/?p=112419
- Mocha, the Node.js and browser test framework, resurfaces on the GitHub feed (@github) — https://github.com/mochajs/mocha
- sindresorhus's eslint-plugin-unicorn ships more than 200 ESLint rules (@github) — https://github.com/sindresorhus/eslint-plugin-unicorn
- Bartosz Ciechanowski's 2022 interactive Mechanical Watch explainer hits 502 points (@newsycombinator) — https://ciechanow.ski/mechanical-watch/
- John Carmack posts on Fabrice Bellard, the one-person engineering legend behind QEMU and FFmpeg (@newsycombinator) — https://twitter.com/ID_AA_Carmack/status/2064095424420487226
- Apple's vehicle motion cues dots reviewed as a real car-sickness fix (@newsycombinator) — https://www.theverge.com/tech/942854/apple-vehicle-motion-cues-review-really-work
- Slay the Spire 2 uses correlated randomness, dissected in a 219-point post (@newsycombinator) — https://tck.mn/blog/correlated-randomness-sts2/

## The Takeaway

Two stories rhyme today: the LinkedIn npm backdoor caught by a read-only review agent, and ByteDance's Lark CLI that hands agents authenticated access to messages and docs. If you are wiring agents into your toolchain this week, run untrusted code through a read-only AI reviewer before npm install, and audit any agent CLI's act-as-you permissions before you grant them. Agents are now both your best supply-chain defense and your largest new attack surface.

## The Call

Within 9 months SpaceX routes Cursor's default model to Grok and quietly trims at least one of its Anthropic or Google compute leases using the 90-day termination clauses.

The case: SpaceX absorbed xAI in February and is now buying the developer surface, not licensing it, and it holds ~$26B/year of cancellable cloud leases. Owning both GPUs and the IDE makes paying rivals for Cursor's inference a cost it controls, and the consensus framing treats the model dependency as untouched.

What proves us wrong: By March 17, 2027 Cursor still ships with Claude or GPT as its out-of-the-box default model and SpaceX has not publicly reduced or terminated either the Anthropic or Google compute lease.

Settles: by March 2027

---
Cite as: "nextbig.dev Daily AI Briefing, 2026-06-17" — https://www.nextbig.dev/daily/2026-06-17