Musk Shuts Down xAI and Hands 220,000 GPUs to Claude – What It Means for the AI Race

Today Elon Musk publicly declared that xAI will be dissolved and folded into SpaceX as an internal AI product, effectively ending the independent company that was founded in 2023 to challenge OpenAI with its Grok model.

At the same time, SpaceX and Anthropic announced a joint deal to lease the full compute capacity of Colossus 1, the world’s largest super‑computer built in Memphis over 122 days, comprising 220,000 GPUs and more than 300 MW of power, to run Claude inference within days.

Financially, xAI’s valuation surged from $0 to $2.5 billion in a year, later reaching $250 billion after a $200 billion financing round, but the company burned roughly $10 billion in 2025 and projected a $130 billion loss for the full year, highlighting a severe cash‑flow hole.

Performance‑wise, Claude Opus leads the 2026 AI model rankings in developer‑tool ecosystems with an 80.8 % SWE‑bench score, while Grok’s scores remain modest and its flagship SuperGrok Heavy costs $300 per month for real‑time X (Twitter) data, limiting its impact.

The strategic rationale, as outlined in the article, is that Musk is using compute as a lever: by renting Colossus 1 to Claude, he both monetizes an idle asset and strengthens OpenAI’s biggest rival, effectively “betting on the infrastructure” rather than on a single model’s superiority.

Comparatively, Google previously invested up to $400 billion and 5 GW of TPU capacity into Anthropic, yet still chose to back Claude, illustrating a broader industry trend where control of compute resources outweighs raw model performance.

Looking ahead, the outcome of Musk’s lawsuit seeking to remove OpenAI’s board members, the upcoming Anthropic IPO pricing, and OpenAI’s response to this compute‑boosting move will shape the next phase of the AI competition, shifting the focus from “who has the best model” to “who controls the underlying compute infrastructure.”