Steeve Morin: Why Google Will Win the AI Arms Race & OpenAI Will Not | E1262
Watch on YouTube ↗ Summary based on the YouTube transcript and episode description.
Steeve Morin (ZML) argues Google will win AI because it alone owns products, data, and compute — OpenAI owns none of the three.
- Google is the only player with all three assets — products, data, compute — making it the sleeping giant; OpenAI owns none of its compute, Microsoft does.
- In five years Morin expects 95% of AI workloads to be inference, only 5% training.
- AMD delivers roughly 4x inference efficiency vs. Nvidia H100 on 70B models at 30% lower cost; Microsoft used this gap to make ChatGPT inference profitable.
- H100 is priced at 5x the A100 but delivers only ~2x inference speed — Morin calls this a bubble with multi-year amortization plans on depreciating collateral.
- Agents and reasoning shift demand from throughput to single-stream latency, a constraint GPUs handle poorly and which may let challengers displace Nvidia in inference.
- Blackwell has a chip-bending heat-dissipation problem; orders are being cancelled and Morin says he has seen worrying operational numbers from early operators.
- Etched and Visor are the two companies Morin bets will deliver affordable high-speed inference chips; Rain and Fractile are the frontier bets on compute-in-memory.
- ZML’s core thesis: zero switching cost across hardware vendors (Nvidia, AMD, TPU, Trainium) is the unlock — being 7x better on specs is not enough to make enterprises switch on its own.
2025-02-24 · Watch on YouTube