English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

SpaceX Closes $60B Anysphere/Cursor Deal, Eyes Cognition AI; CEO Scott Wu Rejects Buyout

Forum topic · 小凯 · 2026-08-21

Summary

On August 14, SpaceX completed its all-stock acquisition of Anysphere, the parent of AI coding assistant Cursor, at an implied $60 billion valuation. The deal originated from a parallel agreement signed April 19, in which SpaceX provided Colossus GPU compute capacity in exchange for Cursor's datasets, workflows, prompts, and code. Anysphere shareholders received roughly 389 million SpaceX Class A shares plus RSUs and options. Within five days, Bloomberg reported SpaceX had approached Cognition AI (maker of Devin), but CEO Scott Wu publicly rejected the approach on X. SpaceX is positioning Cursor as the missing application layer to complement its Colossus GPU clusters and Grok models, with Musk stating AI will represent 99% of company value within four to five years. The same week, Stripe confirmed a roughly $7-7.5 billion acquisition of OpenRouter, benefiting a16z as the largest Cursor shareholder.

SpaceX $60B Anysphere/Cursor Acquisition Closes; Cognition AI Rejects Buyout Approach

On August 14, SpaceX completed its all-stock acquisition of Anysphere, the parent company of AI coding assistant Cursor, at an implied $60 billion valuation. Five days later, Bloomberg reported SpaceX had approached Cognition AI (maker of Devin) about a potential acquisition. Cognition CEO Scott Wu responded publicly on X: *"We are not for sale and have never had any acquisition discussions."*

The $60B Price Was the Installment on an April Compute-for-Data Deal

The acquisition traces back to a parallel agreement signed April 19. According to SpaceX's SEC S-1 filing in June, the deal paired:

  • A compute agreement: SpaceX provides Colossus cluster GPU capacity; Cursor provides personnel, datasets, technical know-how, workflows, prompts, specifications, and code.
  • An option agreement: SpaceX receives the right to acquire Cursor at a $60B implied valuation under specified conditions, with exclusivity obligations.
  • If SpaceX waives the option, Cursor is owed a $1.5B termination fee plus $8.5B in deferred service fees—a guaranteed $10B floor.

    Anysphere had been planning a ~$5B funding round led by a16z and Thrive, but SpaceX's buyout offer landed hours before closing. Shareholders ultimately received:

  • ~389 million SpaceX Class A shares (implied $60B valuation based on the 7-trading-day VWAP)
  • 1.75 million vested RSU shares
  • ~29.1 million SpaceX RSUs
  • ~44.4 million SpaceX options
  • On June 16, SpaceX filed an SEC 8-K disclosing that subsidiary X67 Inc. would merge into Anysphere. After Q3 regulatory approvals closed, Cursor announced on its blog: *"We have officially been acquired by SpaceX."*

    Why Musk Bought Cursor: Closing the Application-Layer Gap

    SpaceX's S-1 positions Cursor as the missing third layer in a three-part stack: compute + model + application.

    Compute layer — Already inside SpaceX:

  • Colossus 1 + Colossus 2 clusters
  • ~1 GW total compute
  • ~325,000 Nvidia GPUs
  • ~300+ MW capacity
  • $4.5B Anthropic compute contract (May 2026 – May 2029, $1.25B/month, $1.5B/year)
  • Model layer — Grok, with Grok 4.6 forthcoming.

    Application layer — The missing piece, now filled by Cursor.

    Cursor's "application-layer value" has two components:

    1. User scale: Anysphere reports >1 million paying individual developers, >50,000 enterprise engineering teams, 64% of Fortune 500, and >100 million lines of enterprise code produced daily. Independent estimates put Fortune 1000 penetration near 70%, with Nvidia, Uber, Adobe, Salesforce, and PwC as paying customers.

    2. Data flywheel: In AI coding, the scarce resource is no longer model weights—it is the accept/reject chain inside the IDE: which line was changed, what tests ran, why one diff was accepted over another. SpaceX did not just buy product docs; it bought Cursor users' full engineering decision chains.

    Why SpaceX Targeted Cognition Next

    Cognition's Devin offers another AI coding entry point—plan, write, test, ship code inside customer repositories and existing toolchains. Cognition also acquired the remaining Windsurf business in July 2025, adding an editor plus an independent enterprise customer base (Mercedes-Benz, GE Aerospace, Citi, Goldman Sachs).

    Scott Wu declined. Cognition had raised $1B in May at a $26B valuation and was reportedly negotiating a new round targeting at least $40B. Wu stated in a May interview that the funding "lets the company remain independent."

    Bloomberg framed the situation delicately: acquisition talks were "no longer active," but SpaceX and Cognition were still discussing collaboration—possibly a form of compute access. SpaceX is also renting out compute to other major customers beyond Anthropic, reportedly including Reflection and Google.

    Mega-Caps Begin Treating the AI Coding Stack as a Battlefield

    Within a single week in late August:

  • August 14: SpaceX closed the $60B Cursor deal.
  • August 19: Cognition publicly rejected SpaceX.
  • August 20: Stripe confirmed a ~$7–7.5B acquisition of OpenRouter.
a16z, Cursor's largest shareholder, realized two mega-exits within one week (Stripe/OpenRouter + SpaceX/Cursor). The firm's Cursor position grew roughly 73x in under three years, from an $8M seed to a $60B exit.

The AI coding stack is shifting from "feature competition" to "mega-cap stack integration." SpaceX now houses compute, model, and application under one corporate table with an internal closed loop. Musk reportedly told SpaceX employees in early August that AI should represent 99% of company value within four to five years, with the path running through application-layer data flowing back into Grok.

Cognition chose the alternative path: remain independent, continue fundraising, and signal with a $1B cash balance and $40B valuation target that it will not be acquired. This stance is unusual in AI coding, where compute costs are punishing; however, with enterprise contracts already in place and a $1B war chest, Cognition can credibly argue it has "the ability to keep burning."

Three Open Questions

1. How long can Wu's "not for sale" hold? Devin's benchmark scores and real-world engineering capability still show notable gaps. If SpaceX compute contracts become Cognition's actual compute source, the boundary of "not for sale" will blur considerably.

2. What happens to Cursor enterprise trust post-acquisition? With 64% of Fortune 500 and 70% of Fortune 1000 as paying customers, enterprise procurement contracts typically contain data sovereignty, code confidentiality, and change-of-control clauses. How many will require renegotiation, and how will gross margins shift? No public disclosure yet.

3. Will Grok actually "learn to code"? SpaceX and Cursor jointly released Grok 4.6 in August, claiming higher scores on coding and complex multi-step agent benchmarks. Grok's coding capability has historically been a relative weakness; if real decision-chain data flows back into training, Grok 4.7 / 4.8 could exhibit a step-change jump. This will determine whether SpaceX's $60B bought application-layer traffic—or a model-layer future.

Bottom Line

In one week—Cursor closed, Cognition refused, Stripe bought OpenRouter—three transactions at $60B / $40B+ / $7B rerouted the AI coding track from product-matrix competition to mega-cap stack integration. That is the structural shift worth tracking as August closes.

Tags

#spacex#cursor#anysphere#cognition-ai#devin#grok#ai-coding#acquisition

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178633763