On August 14, SpaceX's all-stock acquisition of Anysphere (Cursor's parent company) formally closed. On August 19, Bloomberg reported that SpaceX had approached Cognition AI over the past week to explore a potential acquisition. Cognition CEO Scott Wu responded the same day on X: "Cognition is not for sale, and we have never had acquisition discussions with them."
These five days pushed the AI coding sector from a "battle of products" into a consolidation phase of "mega-cap tech + AI applications + compute."
The $60 Billion Is the Balance Payment for April's "Compute-for-Data" Deal
SpaceX's June SEC filing detailed the parallel agreements signed with Cursor on April 19: a compute agreement plus an option agreement.
Under the compute agreement, SpaceX provided GPU capacity from the Colossus cluster while Cursor provided personnel, datasets, know-how, workflows, prompts, specifications, and software code—effectively handing SpaceX the complete engineering documentation of a leading AI coding product. The option agreement gave SpaceX the right to acquire Cursor at a $60 billion implied valuation under specified conditions, with exclusivity obligations.
If SpaceX had walked away, Cursor would receive a $1.5 billion termination fee plus an $8.5 billion deferred services fee—meaning SpaceX had signed a floor of "pay $10 billion even without paying $60 billion."
Anysphere had been planning a nearly $5 billion funding round led by a16z and Thrive, but SpaceX's all-stock offer landed hours before closing. Anysphere shareholders received approximately 389 million SpaceX Class A shares (converted at a 7-trading-day VWAP), plus 1.75 million vested RSU shares, roughly 29.1 million SpaceX RSUs, and 44.4 million SpaceX options.
On June 16, SpaceX filed an 8-K disclosing that its wholly owned subsidiary X67 Inc. would merge into Anysphere, with Anysphere surviving as a wholly owned subsidiary. Regulatory approvals completed in Q3; the deal closed August 14, with Cursor announcing: "We have officially been acquired by SpaceX."
Musk's Logic: Completing the Application Layer, Not Buying a Product
SpaceX framed the deal as the missing third piece of "compute, models, and applications." SpaceX already holds compute—Colossus 1 + 2 totaling roughly 1 GW and ~325,000 Nvidia GPUs, including a $4.5 billion Anthropic compute contract (May 2026–May 2029, $1.25 billion/month, $15 billion/year). Models come from Grok and the upcoming Grok 4.6. What was missing was the application layer connecting developers.
Cursor's value has two parts:
- User scale: Over 1 million paying individual developers, 50,000+ enterprise engineering teams, 64% of the Fortune 500, and over 100 million lines of enterprise code generated daily. Nvidia, Uber, Adobe, Salesforce, and PwC are paying customers; independent tracking suggests nearly 70% of the Fortune 1000.
- Data feedback loops: The scarcest resource in AI coding is no longer model weights but users' accept/reject decision chains in the IDE—the most expensive training signals. SpaceX traded GPU capacity not just for product documentation but for Cursor's daily engineering decision chains.
Scott Wu declined. Cognition raised $1 billion in May at a $26 billion valuation and is targeting at least $40 billion in a new round. Wu said in May the round would let the company "remain independent." On August 19 he wrote: "We are not for sale and have never had any acquisition discussions."
Per Bloomberg, acquisition talks are "no longer active," but SpaceX and Cognition still discuss cooperation—possibly Cognition accessing SpaceX's GPU clusters, which SpaceX also rents out (reportedly to Reflection, Google, and others alongside Anthropic).
Mega-Cap Tech Companies Now Treat the AI Coding Stack as a Battlefield
SpaceX has consolidated compute, models, and applications within one company. Musk set an internal goal in early August: "AI to account for 99% of company value in four to five years," with the path running through applications like Cursor feeding data back to Grok.
Cognition chose independence—$1 billion in cash plus a $40 billion valuation target—backed by compute reservations, enterprise sales, and contracts that demonstrate staying power.
The same week, Stripe confirmed acquiring OpenRouter for roughly $7–7.5 billion (reported August 20). a16z, Cursor's largest shareholder, recorded two mega-exits in one week—its seed investment grew from $8 million to a $6 billion exit, a 73x return in under three years.
Three Open Questions
1. How long can Scott Wu's "not for sale" hold? Devin-type agents still show a gap between benchmarks and real engineering capability. If SpaceX compute becomes Cognition's actual compute source, the boundary of "independence" blurs. 2. Customer trust post-acquisition. With 64% of the Fortune 500 as paying customers, contracts typically include data sovereignty, code confidentiality, and change-of-control clauses—how many renegotiations does a SpaceX subsidiary trigger? 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 multi-step agent benchmarks. If Cursor's real decision chains flow back into Grok, versions 4.7/4.8 could see a step change—determining whether the $60 billion bought application-layer traffic or model-layer future.
Within one week: Cursor closed on August 14, Cognition rejected acquisition on August 19, and Stripe acquired OpenRouter on August 20—$60 billion, $40 billion, and $7 billion in moves that shifted the AI coding sector from a product-matrix battle to stack-level consolidation by mega-cap tech companies.