Broadcom just made $16.7 billion in one quarter building chips that are not Nvidia's. Here is why that matters.

Broadcom's AI semiconductor business hit $16.7 billion in fiscal Q3 2026, up 221% year over year and 54% from the prior quarter. Total company revenue for the quarter was $29.6 billion. Management is now guiding for $21.7 billion in AI chip revenue next quarter, and roughly $115 billion for all of fiscal 2027, with a line of sight to $230 billion by fiscal 2028.

What is actually driving it

Custom AI accelerators, XPUs in Broadcom's terminology, chips designed for one company's specific workload instead of general purpose GPUs anyone can buy. Broadcom shipped Google's Ironwood TPU v7 in volume this quarter and started production on TPU v8I. OpenAI's first custom inference chip, internally called Jalapeño, also shipped, with more generations already in development.

That last detail is the one worth sitting with. OpenAI, the company that built its entire business on top of Nvidia GPUs, now has its own custom chip shipping through Broadcom. Google has been doing this for years with its TPUs. This is not a one company story anymore, it is becoming the default playbook for anyone big enough to afford it.

My actual take

This is the quiet, unglamorous half of the AI story, and it is honestly more important than most of the model release headlines. When Nvidia was effectively the only real option, every company building serious AI infrastructure was locked into one supplier's roadmap, pricing, and availability. That is a fragile position for the entire industry to be in, and everyone building on top of it, including the frontier labs, clearly noticed. Custom silicon is what a company builds when it decides being at the mercy of one vendor is too big a risk to keep taking.

I do not think this kills Nvidia's business, general purpose GPUs are not going anywhere and most companies will never have the scale or capital to justify designing their own chip. But it does mean the biggest players, the ones who actually set the direction of the industry, are actively working to reduce how much any single company controls their roadmap. That is a healthier dynamic for the industry long term even if it is a genuine threat to Nvidia's growth curve specifically.

Why an indie app developer cares about chip supply chains

Because everything downstream eventually touches this. Faster, cheaper, more available AI infrastructure is part of why on device AI features on an iPhone A series chip have gotten genuinely capable in the last two years, without needing a server call at all. The chip diversification happening at the hyperscaler level is not something I compete with directly, but it is part of the same current that has made on device processing a realistic choice for a one person studio instead of a compromise. Cheaper, more available compute at the top of the stack eventually shows up as more capable chips in your pocket.

I will be watching Broadcom's Q4 numbers when they land, and whether that $115 billion 2027 guidance holds, because if it does, that is a real signal this shift is structural, not a one quarter fluke.

Figures here reflect Broadcom's fiscal Q3 2026 earnings results and guidance (Source). This is a recap of a public earnings report, not investment advice. Verified September 6, 2026. If you are curious what a small, one person studio builds with the on device AI this same infrastructure boom trickles down into, the full lineup is at jcmobileappstudio.com/apps.