Everyone talks about Nvidia when AI chips come up. But there’s another player quietly locking up the biggest hyperscalers in the world — and billionaire Ken Fisher has $4.79 billion riding on it.
Broadcom (NASDAQ: AVGO) has spent years building something Nvidia cannot easily replicate: custom, workload-specific silicon. While Nvidia’s GPUs dominate general-purpose AI training, Broadcom designs application-specific ASICs and XPUs that the biggest cloud companies prefer for large-scale inference tasks. The reason is simple math — Broadcom’s chips can cut total cost of ownership by roughly 40–60% compared to GPU clusters when running steady, predictable production workloads. Google, Anthropic, and TikTok/ByteDance have all signed multi-year contracts.
Here’s the macro tailwind that makes this even more compelling: the custom AI chip market is projected to exceed $50 billion in 2026, according to Deloitte. The shift driving that number is a move from AI model training (Nvidia’s sweet spot) toward inference — the act of running those models in production. Inference is expected to account for roughly two-thirds of all AI compute in 2026. Broadcom’s chips are built for exactly that environment.
The moat here is real. There are only a handful of companies capable of designing high-performance custom chips for hyperscale AI workloads. Broadcom is one of them, and that exclusivity has real pricing power. Analysts also estimate that deploying Broadcom’s silicon at scale can save companies billions in upfront capex and cut electricity costs by about half — a compelling pitch when data center energy bills are making headlines.
Ken Fisher’s firm holds $4.79 billion in AVGO, ranking it 8th in his AI-focused portfolio. While AVGO isn’t the sexiest AI name in the room, it might be the most defensible. Every hyperscaler racing to cut inference costs is essentially a potential Broadcom customer. That’s not a niche market — that’s the entire AI infrastructure buildout.
