Broadcom ASICs podcast: GPU supply means little when substrates and MLCCs bottleneck output
BenBajarin · x · 2026-09-09
Ben Bajarin shares takeaways from The Circuit podcast episode "Agents Gone Wild" on Broadcom's custom ASICs and the AI frontier.
- Bottleneck economics: adding GPUs or memory doesn't raise total system output if substrates, MLCCs, or wafer testing are short — shipments are set by the single most constrained component.
- Capacity timing: substrate, power/analog, foundry, and memory suppliers only make greenfield investments once long-term demand is certain. Broadcom, Nvidia and others signed sufficiently long LTAs with supply chains around late 2025–early 2026, so meaningful new capacity arrives years later, pointing to peak constraints around 2027.
- The episode also covers modeling custom silicon, 2028 capacity and pricing debates, Nvidia's acquisition of Hugging Face, AI security, and enterprise governance of AI agents.
More from Infra
- vLLM x AgentX: Full-Stack Optimizations for Real-World Agentic Serving — jfiance · 2026-09-09
- Desert Ant Labs introduces on-device intelligence for every product — Arcuru · 2026-09-09
- Speculative Decoding With Qwen3-30B-A3B Yields 1.5x Local Speedup, Up to 5x — Arindam_1729 · 2026-09-09
- Explainer: Speculative Decoding Speeds Up LLM Inference by ~100% — blaizedsouza · 2026-09-09
- Cerebras CTO's chip architecture deep dives—WSE-3, Hot Chips 34, Cornell lectures—barely get any views — blaizedsouza · 2026-09-09
- Cosmos3 (64B) INT4 Quants Bring Local Image and Video Gen to Mac and CUDA — Formal-Swordfish-228 · 2026-09-09