SpaceX Acquires Cursor for US$60 Billion Just Days Following Its IPO

On June 16, 2026—just 4 days after its blockbuster NASDAQ debut that valued the company at over US$2 trillion—SpaceX announced a merger with Cursor in an all-stock transaction valued at US$60 billion. The deal is expected to close in 3Q 2026, subject to antitrust clearance, with a US$10 billion breakup fee for SpaceX included in the agreement. This deal’s history initially made headlines back in April 2026, when SpaceX and Cursor entered into an agreement granting SpaceX the exclusive right to either acquire Cursor or pay the US$10 billion breakup fee. SpaceX had 30 days following its Initial Public Offering (IPO) to activate this exclusive acquisition agreement. At the same time, it also announced a compute agreement to train Cursor’s Composer 2.5 model on SpaceX’s Colossus supercluster using NVIDIA H100 Graphics Processing Units (GPUs).

Cursor, founded in 2022, had grown to a reportedly US$4 billion Annual Recurring Revenue (ARR) business by June 2026 and had executed a US$2.3 billion Series D back in November 2025, when the company was valued at US$29.3 billion. Competing against products like GitHub Copilot, Claude Code, and OpenAI’s Codex, it has market differentiation through its model-agnosticism, as well as its own Composer model family, which it trained for agentic coding workflows.

This acquisition/merger has direct consequences for every compute provider currently in Cursor’s stack. As an independent company, it was a named customer for players like CoreWeave, Fireworks AI, Together AI, and Baseten—a multi-cloud footprint due to the sheer scale of its operation and infrastructure necessities. However, as a subsidiary of SpaceX, there could be a shake-up on this layer of its stack. Third-party GPU providers like Fireworks AI, Baseten, Together AI, etc. face more direct displacement risk as SpaceX’s Colossus data center has GPU capacity abundance and is already being leveraged by Cursor for model training. Break clauses in SpaceX’s agreements with Anthropic and Google Cloud could see over 200 Megawatts (MW) of GPU capacity become available again within the Colossus data center. Hyperscalers are likely to be less affected, as it is reported that Cursor was using Central Processing Unit (CPU) racks from them, and given their necessity for agentic systems, this is not going away.

On the Cursor platform itself, model agnosticism is likely to remain a commercial differentiator. However, the merger announcement also confirmed that the two companies have been jointly training a model for agentic coding. Therefore, SpaceX’s Grok models are highly likely to be elevated to first-class citizens within Cursor’s platform. The commercial implication extends to the compute layer as well, as SpaceX can optimize all Cursor-generated Grok traffic on its data centers to minimize latency to drive sustained utilization of Grok, a requirement only growing in importance as agentic systems and multi-turn reasoning workloads scale.

The announcement of the jointly trained coding model also points to a larger market development in 2026—frontier models continue to be more heavily trained on coding-related data to be frontier coding models, rather than simply frontier general Large Language Models (LLMs). Anthropic’s Opus, Fable, and Mythos models have advanced significantly in coding capabilities (especially for cybersecurity), OpenAI’s GPT-5.5 has also pushed OpenAI to the top of many coding-related benchmarks, and almost all Chinese open-source Artificial Intelligence (AI) labs like DeepSeek, Z AI, and MiniMax have aligned their recent model releases with agentic and coding use cases. The result is that the model landscape for Agentic AI will continue to grow as labs begin to specialize and optimize their models for specific use cases and workloads. This will result in sustained training demand growth over the coming years, even as inference proliferates.

This acquisition by SpaceX gives it a stronger full-stack AI value proposition, positioning it to better compete against its competitors like OpenAI and Anthropic. Should it continue to lease capacity from its data center portfolio to other players in the AI ecosystem, SpaceX will have a unique position between frontier AI lab and cloud provider. For OpenAI and Anthropic, the long-term strategic challenge will be in the strong enterprise channel that Grok will now have via the Cursor platform, especially due to the credibility the platform has built with developers as a model-agnostic agentic coding solution. The performance of the Grok model relative to the other models available on Cursor, due to inference optimizations SpaceX will be able to make for Grok from Cursor workloads on its own data centers, will be something that developers will actively be evaluating.

Should this result in developers shifting to Grok models in the future, it could change the business model for OpenAI and Anthropic. OpenAI is already developing custom chips for inference via Broadcom, and Anthropic is exploring the possibility of adopting a similar strategy, so we could see them finding more value in the future in owning and operating some of their own data center facilities for workloads via their own platforms like Codex and Claude Code.

Inference providers that may lose some of their Cursor capacity demand as SpaceX moves internal workloads in-house have several strategies that they can adopt. First of all, providing compute for niche AI labs that have specialized model families and require a compute channel is a great stopgap for capacity demand, and helping them optimize model performance to grow demand would work. We have already seen Baseten launch its Frontier Gateway solution back in May 2026 to help labs launch inference Application Programming Interfaces (APIs) on top of Baseten’s inference compute, and while targeting smaller niche players, there is a lot of opportunity to support their growth.