Do 50GW of AI compute need $10T ARR? A DeepSeek-pricing back-of-envelope debate
teortaxesTex · x · 2026-10-21
Using DeepInfer efficiency stats, one estimate: at DeepSeek Flash 4.1 offpeak prices, 1GW yields only $38 revenue per GB200, so absorbing 50GW by end-2027 needs $2.5T ARR at DeepSeek pricing or $10T (8% of global GDP) at US-lab prices. teortaxes rejects the bearish thesis: unit prices will crash and current token-consumption modes become marginal, since LLMs are a general substrate for robots, spreadsheets, bio experiments and more.
Related event: Debate erupts over AI compute economics: what revenue does 50GW need(7 posts)→
More from Venture
- What Would a High-Margin LLM Business Look Like When Everyone Bills by Token? — WholeOrganization915 · 2026-10-21
- Confidential Inference quietly grows as onchain AI workloads demand privacy — 0xJeff · 2026-10-21
- YC-Backed Manufacturing Startup Nox Metals Hits $25M Annualized Revenue in Under a Year — ycombinator · 2026-10-21
- Nvidia's $12.9B Hugging Face acquisition is its largest ever — FinanceYF5 · 2026-10-21
- 8 US cybersecurity giants each bought an AI-native firewall startup in two years — FinanceYF5 · 2026-10-21
- Why OpenRouter is worth $7.5B: a 5.5% toll on every model API call — FinanceYF5 · 2026-10-21