Overview
- Musk wrote Thursday that he was “clearly wrong” about Anthropic, praised its Mythos and Fable models as best‑in‑class, and said he expects a follow‑up model, Mythos 2, soon.
- He publicly ruled out cutting off Anthropic’s access to SpaceX/SpaceXAI compute, saying he would not sever service in a way that “hurt them badly,” which Calms immediate worries about leverage over the company’s operations.
- Anthropic signed a May lease for the full Colossus 1 capacity — more than 300 megawatts and roughly 220,000 Nvidia GPUs — under terms reported as about $1.25 billion per month through May 2029 with 90‑day termination rights.
- U.S. agencies briefly restricted global access to Mythos and Fable for national‑security review before the Commerce Department lifted those limits, and the models have drawn attention for reported cybersecurity capabilities including mass vulnerability detection.
- The reversal matters for markets and customers because Anthropic’s near‑$1 trillion valuation and confidential June IPO filing rest on model performance, large leased compute, and rising federal scrutiny that will shape competition and product access.