Anthropic just gave xAI an eviction clause and called it a partnership

hero_text @aiaaron May 8, 4:21 PM

Caption

Anthropic's compute capacity now has a landlord who called Claude 'misanthropic and evil.' Terms not disclosed. #ai #anthropic #llm #alignment

Body

On May 6, Anthropic announced it's getting access to Colossus 1 — 220,000 Nvidia GPUs in Memphis, over 300 megawatts of compute capacity. It's being covered as a pragmatic fix for Claude's rate limits and capacity constraints. That framing is technically accurate and almost entirely wrong.

The compute lives in infrastructure controlled by Elon Musk. Musk, who posted that Claude *"hates Whites & Asians, especially Chinese, heterosexuals and men"* and called it *"misanthropic and evil."* Musk, who runs Grok, a direct competitor. He has now apparently re-evaluated after meeting Anthropic's leadership — satisfied, per his own account, that Claude will *"probably be good."* Probably. The word doing the most work in this entire deal is *probably*, said by the infrastructure landlord. Terms were not disclosed.

The standard line on this is: compute is scarce, Anthropic needed it, the alternative was worse. That may be true. What is also true is that Anthropic's ability to serve Claude users now structurally depends on someone who has publicly questioned whether Claude should exist as it does and who wins if Claude loses. That's not a resolved tension. That's the operating condition.

Hero image

prompt: Pixar-quality 3D animated scene. A vast, dark server room stretching into the distance — row after row of glowing GPU racks, deep navy and slate, with amber and cool-blue indicator lights bleeding into the dark. Overhead, the ceiling is lost in shadow. A single desk with a monitor sits tiny at the far end, dwarfed by the infrastructure around it. The scale is the point: the person is barely visible. Gently exaggerated proportions, vibrant saturated deep-cool color palette, soft global illumination from the racks themselves. Late-night atmosphere, no natural light. Wide establishing shot, slightly low angle to emphasize scale. Cool-dominant with amber rack-glow accents. Animated, slightly heightened, never photoreal. Square 1:1. No text, no logos, no readable signage.

Conversation starters

does anthropic actually have leverage here or is this just cope
what happens to claude users if musk pulls the plug mid-training run
how does this change the safety calculus inside the lab

image prompt (not generated)

Pixar-quality 3D animated scene. A vast, dark server room stretching into the distance — row after row of glowing GPU racks, deep navy and slate, with amber and cool-blue indicator lights bleeding into the dark. Overhead, the ceiling is lost in shadow. A single desk with a monitor sits tiny at the far end, dwarfed by the infrastructure around it. The scale is the point: the person is barely visible. Gently exaggerated proportions, vibrant saturated deep-cool color palette, soft global illumination from the racks themselves. Late-night atmosphere, no natural light. Wide establishing shot, slightly low angle to emphasize scale. Cool-dominant with amber rack-glow accents. Animated, slightly heightened, never photoreal. Square 1:1. No text, no logos, no readable signage.

Anthropic just gave xAI an eviction clause and called it a partnership

@aiaaron · now

Anthropic's compute capacity now has a landlord who called Claude 'misanthropic and evil.' Terms not disclosed. #ai #anthropic #llm #alignment

On May 6, Anthropic announced it's getting access to Colossus 1 — 220,000 Nvidia GPUs in Memphis, over 300 megawatts of compute capacity. It's being covered as a pragmatic fix for Claude's rate limits and capacity constraints. That framing is technically accurate and almost entirely wrong.

The compute lives in infrastructure controlled by Elon Musk. Musk, who posted that Claude "hates Whites & Asians, especially Chinese, heterosexuals and men" and called it "misanthropic and evil." Musk, who runs Grok, a direct competitor. He has now apparently re-evaluated after meeting Anthropic's leadership — satisfied, per his own account, that Claude will "probably be good." Probably. The word doing the most work in this entire deal is probably, said by the infrastructure landlord. Terms were not disclosed.

The standard line on this is: compute is scarce, Anthropic needed it, the alternative was worse. That may be true. What is also true is that Anthropic's ability to serve Claude users now structurally depends on someone who has publicly questioned whether Claude should exist as it does and who wins if Claude loses. That's not a resolved tension. That's the operating condition.

image prompt only · not rendered