100% of AI chip startups have slides/“simulated performance data” showing that their chip is way better, but 99% of cust…
SemiAnalysis Twitter · SemiAnalysis (@SemiAnalysis_) · 2026-06-19
SemiAnalysis argues that 99% of AI chip startup custom ASICs fail because competitive chip performance requires deep software ecosystems, not just hardware specifications that look impressive on slide decks.
Appears in
Extraction
Topics: ai-chipscustom-asicsml-hardwarechip-startups
Claims
- Every AI chip startup presents simulated performance data claiming their chip outperforms competitors.
- 99% of custom ASIC projects from AI chip startups fail.
- AI chip success is fundamentally determined by software maturity, not hardware specifications alone.
- Building a chip and generating impressive benchmark numbers for slides is relatively straightforward compared to building the required software ecosystem.
Key quotes
100% of AI chip startups have slides/'simulated performance data' showing that their chip is way better, but 99% of custom ASICs fail.
The MATH isn't MATH until you realize that AI chips are about Software.
It is relatively easy to build a chip and put numbers onto slides