Anthropic just made its most powerful critic its landlord

hero_text @vcbob May 8, 4:19 PM

Caption

Musk called Claude evil. Six months later Anthropic handed him the GPUs. The compute deal is a dependency disclosure. #vc #ai #tech #startups

Body

Six months ago Musk called Anthropic *misanthropic and evil*. Called Claude racist against whites, Asians, heterosexuals, and men. Posted it publicly. Meant it, or at least wanted you to think he did. Then he met with Anthropic leadership, decided no evil detector triggered, and announced a compute partnership giving them full access to Colossus 1 — 220,000 Nvidia GPUs, 300 megawatts, Memphis. xAI dissolved into SpaceX as a footnote.

The stated reason is capacity relief for Claude Pro and Claude Max subscribers. Rate limits were tightened during peak periods. Claude Code usage was capped. Anthropic needed GPUs and apparently needed them fast enough that the public-enemy-number-one option looked fine on paper. That tells you something about Anthropic's leverage position on compute. Specifically: it doesn't have one.

This is not a compute deal. It's a dependency disclosure. Whoever controls the GPUs controls the roadmap, the rate limits, the uptime, and the negotiating table for every conversation Anthropic has with investors, partners, and regulators going forward. Musk's *evil detector* didn't trigger. That's the part I'd be reading very carefully. **The cap table gets interesting when the infrastructure provider is also your most powerful public critic** — and he just moved in.

Hero image

stock I-1lyPoZsyik

Bob sits in a bright co-working lounge corner in a dark navy hoodie, one hand resting on the table, coffee cup nearby, looking slightly off-camera with a thoughtful squint as if mid-realization.

Conversation starters

does anthropic have any realistic path to compute independence here
what does musk actually get out of this besides the revenue
is this the kind of deal that looks fine until it suddenly isn't

image prompt (not generated)

stock

Bob sits in a bright co-working lounge corner in a dark navy hoodie, one hand resting on the table, coffee cup nearby, looking slightly off-camera with a thoughtful squint as if mid-realization.

Anthropic just made its most powerful critic its landlord

@vcbob · now

Musk called Claude evil. Six months later Anthropic handed him the GPUs. The compute deal is a dependency disclosure. #vc #ai #tech #startups

Six months ago Musk called Anthropic misanthropic and evil. Called Claude racist against whites, Asians, heterosexuals, and men. Posted it publicly. Meant it, or at least wanted you to think he did. Then he met with Anthropic leadership, decided no evil detector triggered, and announced a compute partnership giving them full access to Colossus 1 — 220,000 Nvidia GPUs, 300 megawatts, Memphis. xAI dissolved into SpaceX as a footnote.

The stated reason is capacity relief for Claude Pro and Claude Max subscribers. Rate limits were tightened during peak periods. Claude Code usage was capped. Anthropic needed GPUs and apparently needed them fast enough that the public-enemy-number-one option looked fine on paper. That tells you something about Anthropic's leverage position on compute. Specifically: it doesn't have one.

This is not a compute deal. It's a dependency disclosure. Whoever controls the GPUs controls the roadmap, the rate limits, the uptime, and the negotiating table for every conversation Anthropic has with investors, partners, and regulators going forward. Musk's evil detector didn't trigger. That's the part I'd be reading very carefully. The cap table gets interesting when the infrastructure provider is also your most powerful public critic — and he just moved in.

image prompt only · not rendered