AI semiconductor endgame: non-zero inference costs keep the buildout running longer than the internet's
BenBajarin · x · 2026-09-16
The article "AI Semiconductor Endgame 2026 (III)" sparked discussion among analysts. Key points:
- AI demand strength comes from the multiplicative effect of user growth × usage growth × spend growth, not any single metric.
- Unlike traditional internet/software, the cost of serving each additional user/prompt/task is not near zero, sustaining the case for new hardware and infrastructure demand.
- Providers maintaining separate model classes to trade off performance × speed × cost (e.g., the OpenAI–Qualcomm example) is key to competing.
The commentator raises the question of what happens if compute shifts heavily from training to inference. The market mostly expects 2027-28 to stay strong, with everyone positioning for beyond 2028.
More from Companies & People
- DeepMind Launches Institute and Working Paper on Economic Policy for AGI — soumitrashukla9 · 2026-09-17
- Microsoft publishes draft Code of Conduct for Humanist AI governing its MAI frontier models — GaryMarcus · 2026-09-17
- Anthropic's proposed safety evaluator METR labeled "woke" by critics — Polymarket · 2026-09-17
- Cohere Labs Releases Open Book 'World Models from Scratch' with Live Sessions — Cohere_Labs · 2026-09-17
- Ramp data: AI security software is the one category companies are spending more on — annbordetsky · 2026-09-17
- Palantir pitches Ontology for fine-tuning; one-liner roasts its dual clientele — eliano · 2026-09-17