SemiAnalysis cuts 2028 AI capex forecast 28% to $758B, NVIDIA hit hardest
SumitGup · x · 2026-11-17
SemiAnalysis sharply revised down global AI accelerator demand, shipments and capex forecasts, with the biggest cuts from 2028 (2028 AI capex lowered 28% to $758B).
- NVIDIA hit hardest: 2027–2030 shipments cut 22–66% by compute revenue; Feynman GPU pricing slashed (Feynman $110K→$78K, Ultra $175K→$98K, Next $191K→$118K) as NVIDIA must accept lower margins; cuts extend to Groq LPUs and standalone CPUs.
- Others fare better: Broadcom's ASIC revenue shielded by financing support; MediaTek faces larger TPU cuts; AMD milder thanks to hyperscaler ties; Amazon Trainium largely unchanged near-term.
- Customer exposure: NVIDIA GPUs are the most marginal of Anthropic's four compute platforms; OpenAI favors AMD (warrant deal) and in-house ASICs; NVIDIA has the highest exposure to Neocloud long-tail customers, some renting to Chinese end customers — a risk if the RASA act passes export-control review.
More from Venture
- Courses are becoming SaaS: packaging your method as agent skills buyers can run — VibeMarketer_ · 2026-11-17
- SEC filings: Jensen Huang moved $10.5B of Nvidia stock to family trust before resigning — Polymarket · 2026-11-17
- Goldman trading desk: AI sentiment at rare lows, chip stocks priced for negative growth — GaryMarcus · 2026-11-17
- SemiAnalysis Slashes Feynman ASPs to $78K: NVIDIA May Sacrifice Gross Margins to Defend Share Against OpenAI Chips and Google TPUs — SumitGup · 2026-11-17
- Japanese Local AI Startup Kotoba Joins ElevenLabs — lukeharries · 2026-11-17
- Polymarket puts 70% odds on OpenAI's GPT Astra launching this week as Anthropic IPO looms — Polymarket · 2026-11-17