Latent SpaceJun 19, 2026, 01:30 UTC+8

Anjney Midha on AI compute waste: frontier labs run sub-10% MFU, AMP plans an independent compute grid

The Professor of Outputmaxxing — Anjney Midha, AMP

Anjney Midha discusses hidden AI infrastructure waste on Latent Space. xAI's training MFU is under 10%, while Google treated 95% utilization as an outage; best-in-class today is 60–70%. He invested in Anthropic, Mistral, and Black Forest Labs, and now runs AMP, aiming for a 1.2 GW base-load compute grid with 6 GW spike capacity. He also flags DeepMind's unpublished research as a market failure and notes Anthropic prioritized coding as P0 from day one. The post does not disclose a timeline for AMP's grid.

Why it matters: Anjney Midha puts a specific number on xAI's training MFU (under 10%), turning vague complaints about compute waste into a quantifiable discussion. Score capped at 78 because it's a podcast opinion, not a product launch or paper — no reproducible verification path.

Read the original ↗Export Markdown

Take

This episode is worth clicking because Anjney Midha drops a concrete number: xAI's training MFU is under 10%. For context, GPT-3 hit 21%, PaLM reached 46%, and today's best teams get 60–70%. Google once treated 95% utilization as an outage. The bottleneck isn't GPU supply anymore—it's systems engineering: scheduling, networking, parallelism, cluster reliability. If any of those slip, your theoretical FLOPs never become real training progress.

Midha backed Anthropic, Mistral, and Black Forest Labs before starting AMP, which aims to build a 1.2 GW base-load compute grid with 6 GW spike capacity. He also flags DeepMind's unpublished research as a market failure and notes Anthropic prioritized coding as P0 from day one—that's why Claude got good at it early.

But the post doesn't give a timeline for AMP's grid, so the 1.2 GW vision is still on paper. The MFU figure comes from a SemiAnalysis tweet and Midha's own claim, not an official xAI disclosure—I'd discount it a bit until we see more.

Source: Latent Space · latent.space
Published: Jun 19, 2026, 01:30 UTC+8 · Jun 19, 2026