← All posts / Industry

Broadcom's AI Chip Revenue Triples to $16.7 Billion — and Hock Tan Is Now Doubling Down Through 2028

Broadcom's Q3 FY26 print: AI semiconductor revenue up 221% to $16.7B, Q4 guided to $21.7B, and a staggering new roadmap targeting $115B in FY27 and $230B in FY28.

Broadcom's AI Chip Revenue Triples to $16.7 Billion — and Hock Tan Is Now Doubling Down Through 2028

Broadcom just delivered the single most important datapoint in the AI infrastructure trade — and then immediately raised the stakes for the next two years.

After the close on September 2, the company reported fiscal Q3 2026 revenue of $29.59 billion, up 86% year over year and ahead of the roughly $29.2–29.4 billion Wall Street consensus. Adjusted EPS came in at $3.32 versus approximately $3.22–$3.24 expected — the company’s ninth consecutive beat. But the headline that matters is buried in the segment tables: AI semiconductor revenue hit $16.7 billion, up 221% year over year and 54% sequentially, pushing Semiconductor Solutions to roughly 70% of total company revenue.

“Demand for our custom AI accelerators and networking continues to be very strong,” CEO Hock Tan said in the release. “Q3 AI semiconductor revenue of $16.7 billion grew 221% year-over-year, and 54% quarter-over-quarter.”

The numbers, in context

To understand why a routine earnings report is being treated as a bellwether for the entire AI buildout, trace Broadcom’s AI revenue curve through fiscal 2026:

  • Q1 FY26: $8.4 billion
  • Q2 FY26: $10.8 billion (+143% YoY)
  • Q3 FY26: $16.7 billion (+221% YoY) — beating the company’s own $16.0 billion guide
  • Q4 FY26 guidance: $21.7 billion (+236% YoY)

That trajectory puts full-year FY26 AI revenue at roughly $57.6 billion, comfortably ahead of the $56 billion management guided in June — with second-half AI revenue nearly doubling the first half. And on the earnings call, Tan went further than he ever has publicly: he told investors Broadcom is aiming to roughly double AI revenue to $115 billion in fiscal 2027, and then double again to $230 billion in fiscal 2028.

Let that sink in. A company whose total revenue was $51.6 billion in fiscal 2025 is now forecasting that a single segment — custom AI silicon and the networking that connects it — will approach a quarter of a trillion dollars in annual sales within roughly two years.

The custom silicon engine

Broadcom’s AI business runs on two pillars. The first is XPUs: custom accelerators designed with hyperscale customers. Google’s TPU program is the anchor, and the customer list has visibly widened — Anthropic, revealed in December 2025 as Broadcom’s mysterious “$10 billion customer,” and Meta are now firmly in the fold, with OpenAI’s first custom inference chip, the Jalapeño accelerator co-developed with Broadcom on TSMC’s N3P process, adding a fourth marquee relationship. The second pillar is AI networking — Tomahawk and Jericho switch silicon — which grows in lockstep with every new accelerator cluster, because custom XPUs still need to be wired together at enormous scale.

According to live coverage of the call, Tan said order visibility for AI now extends into 2028 across Google, Anthropic, and Meta — an extraordinary claim in a semiconductor industry where lead times are normally measured in quarters, not years.

The software side of the house is no longer the story, but it is quietly compounding: infrastructure software, anchored by VMware, was guided to roughly $8.9 billion for the quarter, up 31% year over year — a sharp acceleration from Q2’s 9% growth as the VCF (VMware Cloud Foundation) conversion cycle matures.

The market’s awkward reaction

Here is the strange part: despite the beat-and-raise quarter, the after-hours tape was ugly. Shares initially ticked up about 0.4% on the release, then slid as much as 5% as investors digested the guidance — continuing a pattern that has punished Broadcom all year for the crime of “not blowing out” estimates by enough.

The context matters. Broadcom closed September 2 around $368, down roughly 0.4% on the day and about 25% below its 52-week high of $495 — a peak set on June 3, the day of the Q2 report, before the stock cratered 12.6% the next session when Tan declined to raise full-year guidance. Since then, the setup has been tense: FY27 EPS estimates climbed from $18.22 to $19.53 over 90 days, and analysts’ published price targets sit well above the current price. A 1.2% revenue beat and a Q4 guide that merely matches elevated expectations was, apparently, not the “gut check” moment buyers wanted — at least not in the first minutes of trading.

Why $230 billion is both plausible and terrifying

The bull case writes itself: Tan’s guidance conservatism is well documented — he guided Q3 AI revenue to $16.0 billion and delivered $16.7 billion. Google, Anthropic, Meta, and OpenAI have collectively committed to hundreds of billions of dollars in multi-year infrastructure spend, and every dollar of custom-silicon spend flows disproportionately to Broadcom, which owns the design methodology, the SerDes and networking IP, and the deepest co-packaging relationships with TSMC.

The bear case is the one this blog has tracked all summer. Broadcom’s AI empire is increasingly built on debt-financed demand: Bank of America flagged roughly $370 billion in AI-related debt across the complex in mid-August, and Broadcom itself has been assembling a financing package — reported at $70–100 billion — with Apollo and Blackstone to fund Anthropic’s chip purchases. When your customers finance their orders with lenders who are sometimes also your shareholders, the demand signal gets noisier, not cleaner. A $230 billion FY28 forecast assumes the financing chain holds through at least two more years of rising rates scrutiny and hyperscaler depreciation debates.

There is also the competitive question. Marvell — now backed by a Google equity warrant — MediaTek’s NVLink Fusion push, and OpenAI’s own accelerating silicon program all target the same custom-ASIC throne. Broadcom’s 221% growth says the moat is holding; the stock’s 25% drawdown from June says the market is pricing in the day it doesn’t.

The bottom line

Broadcom’s Q3 FY26 report is the clearest confirmation yet that custom AI silicon — not GPUs alone — is the fastest-growing product in the history of the semiconductor industry. Hock Tan just put a $115 billion FY27 target and a $230 billion FY28 target on the record, backed by order visibility that runs to 2028. Whether the stock likes it this quarter is almost beside the point: the buildout is still accelerating, and Broadcom is still its quiet, indispensable toll collector.