Broadcom's $100B Chips: Look Beyond Just Growth

Broadcom forecasting over $100 billion in AI chip sales by 2027—my first reaction wasn't that the market believes it again, but that management is treating custom chips as a scalable delivery route.

What's interesting is that despite this news, the stock price was dragged down by quarterly guidance. Reports mention current quarter AI chip revenue around $16 billion, below analyst expectations. The market talks about long-term stories verbally, but looks at next quarter's cash flow with its hands.

When I led AI platform teams, what I feared most in procurement meetings was discussing only single-card performance and supply, ignoring model migration, inference throughput, O&M, and fault recovery. Chip companies provide hardware expectations; platform teams provide business availability. The engineering costs in between are often covered up by trillion-dollar narratives.

So this direction is worth investing in, but have you calculated the ROI? That's the first question. Watching Broadcom isn't about seeing if it can steal share from Nvidia, but seeing if cloud providers and large model teams are willing to migrate stable loads to custom chips. Once migration truly happens, testing, scheduling, compilers, and monitoring must follow suit, requiring a group of people in the organization specifically dedicated to getting these dirty jobs done.

Advice for friends in infrastructure: Don't rush to budget according to chip company roadmaps. First stratify your own model loads: which fit general GPUs, which can benefit from custom chip cost advantages, and which migrations would drag down iteration for half a year. Calculate this account clearly, then talk strategy.

---

📌 This article is compiled from Bloomberg Tech. Original: https://www.bloomberg.com/news/videos/2026-09-03/broadcom-forecasts-boom-in-ai-chip-sales-video

Copyright belongs to the original author. This is a compilation and independent analysis based on public reports.