Broadcom's 221% AI Revenue Surge, XPU vs GPU & TSMC Texas | Ep 1

Jamie Cole (0:00)

A Hardware & Chips Daily News.

I'm Jamie Cole, thanks for joining me. Today Broadcom's XBU Gamble, six customers, $115 billion revenue target.

Broadcom just posted 221% year over year growth in AI semiconductor revenue, and that number deserves more attention than it's getting.

The company's AI chip revenue hit $16.7 billion in the most recent quarter. Management is projecting $115 billion for fiscal 2027, and $230 billion for fiscal 2028

Those are extraordinary targets. The signal here is that Broadcom isn't positioning itself as an Nvidia alternative, it's positioning itself as the infrastructure layer for hyperscalers who've decided they don't want Nvidia at all. Here's what's driving that thesis. Anthropic, OpenAI, Meta, and Google are all deploying custom accelerators, known as XPUS, that their teams claim match GDP performance at roughly half the cost. That cost arbitrage is reshaping how the biggest AI spenders allocate their capex. When you're operating at the scale these companies are, halving your chip costs per workload isn't a margin footnote, it's a strategic reorientation. Broadcom sits at the center of that shift. It designs and supplies custom silicon for several of these players, earning revenue from co-developed chips that wouldn't exist without close hyper-salar partnerships.

The important distinction is that Broadcom isn't building a general purpose chip and selling it into an open market. It's building to spec for specific customers. That creates a fundamentally different competitive position than Nvidia South. Nvidia isn't standing still, but it is constrained. The company disclosed it can only fulfill around 70% of current demand. Memory costs are pushing fourth quarter gross margin down to 71-72%.

For a company trading at 46x forward earnings, those are real pressure points. Broadcom, meanwhile, trades at 19x forward earnings, despite faster AI revenue growth than Nvidia. That valuation gap is something analysts are actively flagging, with price targets around $426, implying roughly 20% upside. The market hasn't fully reconciled the growth rate with the multiple. That matters because it suggests the custom silicon story is still being priced as speculative rather than structural. Here's the thing. The risk in Broadcom's story is concentrated in specific. Its entire XPU thesis depends on six hyperscaler customers executing their deployment plans on schedule. Only Anthropic is publicly confirmed as the largest 2027 customer. If power delays or financing pressures push Anthropic or OpenAI's data center buildouts into 2028, the $115 billion guidance compresses fast. That's not a remote scenario.

Debt concerns at AI infrastructure firms surfaced just recently and hit both Nvidia and Broadcom stocks. The residual value guarantees embedded in AI-lad financing are tail risk that the market is still working out how to price. Separate from the Broadcom story, TSMC sent evaluation teams to Texas recently and confirmed discussions with Elon Musk's Terafab project. TSMC's position is clear. Any facility must be operated by TSMC itself. Musk's model assumes the opposite. That governance dispute is unresolved, and it's a real obstacle. TSMC's broader expansion context makes the stakes clear. The company raised four-year capital expenditure guidance to $60-64 billion, up from $52-56 billion. It just reported record quarterly net profit of $24 billion, up 69% year-over-year. It is the dominant foundry for every chip in this conversation, and its geographic decisions carry outsized consequences. Consider this. The near-term signals to track are narrow but clear. Broadcom's next earnings report lands in December. Meta's NTIA shipment disclosures will test where the custom silicon deployment timelines are holding.

And the TSMC Terafab standoff needs a resolution before it becomes a serious delay to US Alpha Advanced Fab capacity. The broader question is whether custom silicon's cost advantage is structural or cyclical. Right now, the economics favor XP use. But that only holds if hyperscaler deployment timelines stay intact and Nvidia's supply constraints don't resolve faster than expected. Thanks for listening. This podcast was built using AI technology. A YesWee production.