Why Anthropic Keeps Backing Cursor While OpenAI Cuts It: Compute Leverage

OpenAI made headlines by announcing it would sever Cursor's direct access to its models in November, citing SpaceX's $60 billion acquisition of the AI coding tool and Elon Musk's track record of violating AI licensing agreements. Yet Anthropic's response—doubling down publicly as a "trusted partner" and pledging more Claude compute to Cursor—has raised eyebrows. The contrast exposes a tension in how AI labs manage competitive relationships when compute dependency complicates their leverage.

A Pattern Broken#

Anthropic has form in cutting off companies that start to look like rivals. When OpenAI was rumored to be acquiring Windsurf in May 2025, Anthropic acted swiftly, revoking Windsurf's Claude access within weeks—before any deal was even done. Co-founder Jared Kaplan later remarked it would be "odd" for Anthropic to supply Claude to a company potentially owned by its arch-rival. (The OpenAI deal fell through; Google ultimately hired Windsurf's team for $2.4 billion.)

Anthropic also cut xAI staff off from Claude when accessing through Cursor in January, citing what it called "a new policy" targeting "major competitors." Yet now, with SpaceX having actually acquired Cursor—and owning xAI as well—Anthropic is publicly amplifying its support for the platform.

Industry observers have noted the apparent double standard. Gergely Orosz, author of the Pragmatic Engineer newsletter, highlighted the shift: "They cut them off ruthlessly when RUMORS of OpenAI potentially buying Windsurf surfaced. Now, SpaceX, an Anthropic competitor bought Cursor — it's still a trusted partner?" Replit CEO Amjad Masad put it more bluntly, suggesting Anthropic's hands are tied: "More likely answer is that you can't do that here because you need the compute."

The Compute Bind#

That assessment hinges on a single, enormous deal: in May, Anthropic secured the entire output of SpaceX's Colossus 1 data center in Memphis—over 300 megawatts and more than 220,000 Nvidia GPUs. The agreement, revealed in SpaceX's IPO filing, commits Anthropic to pay $1.25 billion monthly for compute across Colossus I and II through May 2029, covering roughly 325,000 GPUs combined.

For Anthropic, the arrangement solved an acute constraint. The AI lab had been hampered by compute scarcity; the SpaceX deal let it raise usage limits for Claude Pro and Max subscribers almost overnight. For SpaceX, it unlocked a new revenue stream while preserving capacity for its own AI training and competing inference.

The asymmetry is stark: SpaceX now owns two direct Anthropic rivals—xAI (valued at $250 billion in February) and Cursor. Yet it also holds the keys to the compute infrastructure Anthropic depends on to scale Claude. That dependency creates leverage no earlier target of Anthropic's access cuts possessed.