Musk: "I Was Clearly Wrong About Anthropic" — The $1.25B/Month Compute Deal Made a Critic Into a Backer
Elon Musk reversed his public stance on Anthropic this week, writing on X that he was “clearly wrong about Anthropic” and calling the company the industry’s leader. The statement is a complete reversal from his September 2025 post in which he described Anthropic as “hypocritical,” “woke,” and an “inevitable loser” in the AI race.
The pivot came during the aftermath of the Fable 5/Mythos controversy, after users on X raised the possibility that Musk could terminate Anthropic’s SpaceX compute access as a competitive move. Musk rejected the suggestion and posted praise for the company instead.
The Numbers Behind the Reversal
SpaceX’s Colossus data center earns $1.25 billion per month from Anthropic. Anthropic took all 220,000 GPUs at Colossus 1 and is now expanding to Colossus 2 with GB200 hardware. The arrangement, confirmed in May 2026, makes Anthropic one of SpaceX’s largest single commercial revenue lines.
Terminating the compute contract would have cost SpaceX $15 billion in annualized revenue — roughly equivalent to a significant fraction of its non-Starlink commercial income. Whatever Musk’s competitive preferences, the financial logic of the relationship is not ambiguous.
What He Said Before
Musk’s September 2025 criticism was pointed. He called Anthropic “hypocritical” for receiving funding from Google while positioning itself as an independent safety-focused lab. He predicted the company would fail to compete against OpenAI and said its culture was antithetical to building aggressive, fast-moving technology.
Those criticisms tracked closely with the public competitive narrative at the time: Anthropic was viewed by many observers as a distant third behind OpenAI and Google in model quality. That view has since been substantially revised. Claude Fable 5 reached 95% on SWE-bench Verified, took the top spot on the Artificial Analysis Intelligence Index at 64.9, and led Agent Arena.
The Distillation Subtext
Microsoft CEO Satya Nadella this week offered a parallel observation. Nadella said AI labs complaining about distillation from their models are being hypocritical, because all major labs have trained on outputs from other systems. The remark landed in the context of Anthropic naming Alibaba in a distillation attack involving 28.8 million Claude exchanges and 25,000 fake accounts.
Nadella’s framing suggests the industry’s rhetorical posture on competitive practices is drifting toward more open acknowledgment of what has long been happening in practice. Musk’s reversal on Anthropic reads in the same register: competitive antagonism is hard to sustain when compute economics and market position have both moved against the original narrative.
What It Does Not Mean
Musk’s change of tone does not signal a strategic alignment between xAI and Anthropic. xAI’s Grok 4.5 competes directly with Claude in coding agents and agentic benchmarks. Grok is now a SpaceX product following xAI’s dissolution into the parent company.
What it does signal is that the rhetorical war between AI labs in 2025 — when positioning mattered because market outcomes were still uncertain — has given way to a period where the standings are clearer. Fable 5 is the current benchmark leader. GPT-5.6 Sol is the cost-performance leader. The competitive map has settled enough that Musk can acknowledge Anthropic’s performance without conceding the fight.
The compute dependency does the rest.