Goldman: US to Outspend China $806B to $110B on AI Infra in 2026, China Fights Back on Efficiency
FinanceYF5 · x · 2026-09-23
A Goldman Sachs analysis frames US-China AI competition as scale vs efficiency:
- Capex gap: US major cloud providers are projected to spend $806B on AI infrastructure in 2026 vs $110B for China — a 7x+ gap.
- Efficiency as strategy: DeepSeek V4.1 Flash cut per-token KV cache requirements by 437x vs V1, showing how aggressively Chinese labs optimize inference economics.
- Adoption gains: Chinese models keep gaining OpenRouter token share and repeatedly top usage rankings, driven by lower costs.
- Higher-value workloads: usage is shifting from chat to agents and coding, where reliability and business value are higher.
- Price as a weapon: cheap Chinese models are pushing global inference prices down; at parity of capability, lowest unit cost wins usage.
- Next battleground: Goldman expects digital workers to expand from coding into finance, law and accounting.
More from Venture
- AI infrastructure may repay fine: deployed capex already at 121% revenue coverage — FinanceYF5 · 2026-09-23
- Chuhaiqu Releases Free 304K-Word Playbook Covering Full Overseas Product Launch — SimplyAnnisa · 2026-09-23
- NaiveAI, founded by Tsinghua's Dai Jifeng, raises $400M in 7 months betting on post-training — 创业邦 · 2026-09-23
- AI companies pay huge salaries before profit: Zhipu execs average 3.39M yuan/month, topping Tencent — 创业邦 · 2026-09-23
- Wispr Flow scaled to millions of users via Dub at ~4x ROAS, paying out $300K+ in commissions — JosephJacks_ · 2026-09-23
- Ex-Huawei model leads raise $100M+ seed funding for native physics foundation model startup — 智东西 · 2026-09-23