Broadcom expects its AI-related chip business to generate $230 billion in fiscal 2028, up from $2.6 billion in fiscal 2022 — a two-order-of-magnitude increase spread across custom XPU accelerators, AI networking, and supporting silicon for AI clusters. CEO Hock Tan laid out the forecast while presenting third-quarter fiscal 2026 results to Wall Street, telling analysts to expect no quarterly updates but making clear that major revenue is coming over the next several years.

What did Q3 F2026 deliver?

The quarter, which closed in early August, brought in $29.59 billion in total revenue, up 85.5 percent year on year and 33.4 percent sequentially. Operating income reached $15.96 billion, 2.71 times the year-ago figure. Net income rose 3.17 times to $13.09 billion — 44.2 percent of revenue.

The Semiconductor Solutions group posted $20.84 billion in sales, up 127.4 percent, with operating income of $12.71 billion, up 143.3 percent. The Infrastructure Software group, which includes VMware, recorded $8.75 billion in sales, up 29 percent, and operating income of $7.35 billion — 84 percent of revenue — up 40.7 percent.

AI-related semiconductor revenue reached $16.74 billion in the quarter, up 3.22 times year on year and 54 percent sequentially. The split: $12.22 billion for AI compute and $4.52 billion for AI networking. Broadcom ended the quarter with $23.98 billion in cash against $59.42 billion in debt from the VMware acquisition.

Who are the six XPU customers?

Broadcom's custom AI accelerator roster includes Google, Anthropic, OpenAI, and Meta Platforms, with ByteDance or Apple likely filling the remaining two slots. Amazon Web Services uses Broadcom for packaging and integration on Trainium 4. Across the six customers, Tan cited roughly 30 gigawatts of total compute capacity.

That footprint implies XPU system installations at around $12 billion per gigawatt, excluding CPUs, memory, and racks — closer to $15 billion per gigawatt with those components included. The source contrasts this with approximately $18 billion per gigawatt for Nvidia Grace-Hopper systems and $40 billion per gigawatt for Vera-Rubin systems.

What did Tan say about Google's TPUs?

On the call, Tan said Broadcom will ship "multi-tens of billions of dollars of TPUs annually over the next several years" to Google. "We have consistently delivered the fastest time to market for TPUs from product definition to production without the need for resubmits. We believe these are very deep moats for any competitor to cross."

Broadcom has worked with Google on TPU design and TSMC foundry and packaging for roughly seven years before the GenAI boom began.

Anthropic is deploying 1 gigawatt of Ironwood TPU 7 accelerators this year, 5 gigawatts of TPU 8i systems in 2027, and wants 10 gigawatts of the follow-on TPU 9 in 2028. Tan said Anthropic is on track to become Broadcom's largest XPU customer in 2027 and to hold that position in 2028.

OpenAI has chosen Broadcom to shepherd its "Jalapeno" XPU and is on track to buy 1.3 gigawatts for Jalapeno clusters in 2027. In 2028, Broadcom expects OpenAI to deploy another 5 gigawatts of Jalapeno ASICs and its follow-on, which would make OpenAI the second-largest XPU customer and push Google to third.

The follow-on OpenAI chip is approaching tapeout, and the two companies have begun work on a third-generation XPU design.

With Meta Platforms, Broadcom will deliver three generations of MTIA accelerators between now and the end of fiscal 2027, with line of sight on 3 gigawatts of capacity through the end of fiscal 2028.

What about networking?

Broadcom says its AI boom is dominated by networking, not just XPUs. The 100 Gb/sec and 200 Gb/sec SerDes variants of the Tomahawk 6 Ethernet switch ASIC are selling strongly, and every hyperscaler using Broadcom for custom XPU shepherding also uses Tomahawk 6 for scale-out networking on AI clusters. Tomahawk Ultra, which cuts latency in the Ethernet stack for scale-up coherent memory interconnects within rackscale machines, is seeing surprising adoption now, with the ramp progressing through fiscal 2027.

The 204.8 Tb/sec Tomahawk 7 — not yet announced, built on 400 Gb/sec SerDes, delivering 3.2 Tb/sec per port across eight lanes — has just taped out and could appear in switches in early calendar 2028, possibly late 2027.

Can Broadcom meet demand?

Broadcom has secured HBM memory, chip wafers, and packaging to fulfill the forecast AI compute and networking capacity. Tan said the company can add more capacity if demand runs higher.

The source also speculates that Broadcom may expand into CPU shepherding, potentially focusing on packaging as it does with Fujitsu's forthcoming "Monaka" Arm server CPU. Fujitsu and the six XPU customers all use Broadcom's 3.5D Extreme Dimension System in Package (XDSiP) chiplet stacking, which places stacked DRAM and stacked CPU chiplets into a single package. The CPU total addressable market is projected to grow from $26 billion in 2025 to $220 billion by 2030.