The economics of an AI neolab: 1,000 GB300s cost $125-150M for 3 years at ~2-2.5MW
deedydas · x · 2026-09-27
Investor Deedy breaks down the economics of a "neolab" — an AI-researcher startup raising big pre-production money to fund GPU compute — and invites challenges to his assumptions:
- 1,000 GB300s (14 NVL72 racks) run $125-150M over 3 years, with 15-30% upfront
- 2-2.5MW of power
- Enough for 10^25 FLOPs per quarter, roughly reaching a GPT-4-level model
More from Venture
- Not Boring: the three arguments that AI isn't a bubble are exactly how bubbles form — Not Boring (Packy McCormick) · 2026-11-11
- US data brokers selling $500M in RL training environments to Chinese AI labs — The Decoder · 2026-11-11
- CoreWeave: 2027 GPU renewals could add ~$4B annualized revenue, analyst skeptical — zephyr_z9 · 2026-11-11
- What do you check first when your agent's API bill doubles? Reddit debates cost tracing — rcleffke · 2026-11-11
- Polymarket gives 48% odds Anthropic's IPO closes above $1.5T market cap — Polymarket · 2026-11-11
- Anthropic's IPO reportedly oversubscribed multiple times, per Polymarket — Polymarket · 2026-11-11