Broadcom AI chip revenue triples in fiscal Q3; company targe

Broadcom grew its AI chip revenue 221% year over year to $16.7 billion in the fiscal third quarter and extended its outlook for that business to roughly $230 billion by fiscal 2028, about four times the $58 billion it now expects for the current year. Total revenue of $29.59 billion, up 86% and ahead of the approximately $29.36 billion analysts expected, came with adjusted earnings of $3.32 per share when results were published on September 2.

AI semiconductor sales ran about 4% above the company's own $16 billion forecast and rose 54% sequentially, extending what Broadcom counts as fourteen consecutive quarters of AI-led growth. The Semiconductor Solutions segment, home to the custom accelerators, Ethernet switching, and optics, delivered $20.8 billion of the quarterly total, meaning the beat over the company's internal plan was roughly $700 million. AI now accounts for about 56% of revenue, and the fourth-quarter guidance implies that share climbs toward 62%.

Guidance for the current period calls for total revenue of approximately $34.8 billion, a 93% year-over-year gain that lands slightly under the roughly $35 billion consensus. Inside that number, AI chip revenue is projected at about $21.7 billion, up 236% from a year earlier and around 30% above the quarter just reported, while adjusted earnings of $3.32 compared with consensus near $3.24. Shares slipped roughly 2% after the release despite the beats, and the split explains the muted reaction: the AI line keeps accelerating while the shortfall sits outside it, in the slower non-AI businesses that make up the rest of the company.

The two-quarter comparison makes the same point in dollars. Excluding AI, revenue moves from roughly $12.9 billion in Q3 to about $13.1 billion in the Q4 guide, nearly flat, while the AI line is set to add $5 billion in a single quarter. Broadcom's growth is now almost entirely a function of the custom-silicon and networking franchise, with the rest of the portfolio contributing little to the momentum. For a company that built its name on diversified chip lines and acquisitions, that reversal is a measure of how completely the AI cycle has redrawn its revenue base.

Reading the AI chip revenue road map

Multi-year revenue projections are rare in semiconductors, and this one is unusually specific. Management raised the fiscal 2026 AI target from roughly $56 billion to about $58 billion, growth of 186% over fiscal 2025, then laid out a path that doubles the business to approximately $115 billion in fiscal 2027 and to $230 billion in fiscal 2028. CEO Hock Tan tied the demand to custom XPU programs for OpenAI and Anthropic, the AI labs that have moved from renting most of their compute toward commissioning their own silicon, alongside hyperscaler accounts such as Google that turned custom accelerators into a mainstream procurement model.

The timing inside the forecast is as informative as the totals. A $58 billion fiscal 2026 figure with a $16.7 billion third quarter and a $21.7 billion fourth quarter implies the first half of the year contributed roughly $19.6 billion, which places the acceleration squarely in the second half. That shape fits customers ramping new XPU designs into volume rather than a demand curve that flattened after an early surge, and it is the reason management feels confident publishing numbers two years out.

Each custom program is a multi-year engagement that spans the accelerator itself, the Ethernet fabric connecting clusters, and the optical interconnects between systems, letting Broadcom collect revenue at several layers of one deployment instead of a single chip sale. Doubling the business twice will demand materially larger deployments from existing accounts or new customers reaching comparable scale, and nothing in the current release suggests the pipeline is narrowing.

The presence of OpenAI and Anthropic in the custom program book is itself a signal. Both labs spent their early years renting compute from cloud providers running merchant GPUs, and both have now committed to commissioning their own parts, a sign that frontier developers treat control over silicon design as part of controlling cost and supply. For Broadcom, the newer labs also widen the customer set beyond hyperscalers, reducing the dependence on a single anchor account that defined the early custom-silicon era. For the merchant-GPU side of the market, that is the structural pressure inside Broadcom's guidance: the organizations driving frontier demand are increasingly building their own accelerators instead of waiting on the next merchant release cycle.

The projections also change how the AI capex cycle should be tracked. Public attention tends to fix on GPU shipment tallies and data center spending, but Broadcom's disclosures describe a parallel stream of commitments that runs through design wins and custom silicon, booked years before systems ship. With roughly $58 billion of annual AI revenue visible for the current fiscal year and the curve defined through fiscal 2028, the cycle's forward visibility now comes as much from engineering engagements as from order books.

Margins and the competitive math

The growth has not cost Broadcom its profitability. Operating margin reached 67.9% in Q3 even with a mix tilted toward custom parts, and management guided fourth-quarter non-GAAP operating income to roughly 66% of projected revenue. Free cash flow of $13.7 billion equaled 46% of revenue, funding a quarterly dividend of $0.65 per share payable September 30, 2026. The profile undercuts the old assumption that custom silicon is a lower-margin business than standardized merchant chips.

The competitive stakes are what turn the road map into a market story rather than a company story. Broadcom's numbers are the clearest public signal of how AI capital spending is dividing between off-the-shelf merchant GPUs and purpose-built accelerators, and the trajectory implies the custom share is compounding. Every XPU dollar is a dollar that does not flow to a merchant part, so a sustained move toward $115 billion and then $230 billion would press directly on Nvidia's pricing power at the top of the market, where only a short list of buyers is large enough to justify bespoke designs. Custom accelerators compete with merchant GPUs for the same marginal dollar of capacity, and Broadcom's guidance suggests that competition is intensifying.

Concentration cuts both ways. A handful of customers now drive most of Broadcom's growth, and the fiscal 2028 target assumes their plans hold through two more years of near-doubling, which in turn requires the supply chain and the customers' engineering programs to scale without slipping. The muted share reaction after such a strong report hints that investors were already positioned for aggressive AI growth, moving the open question from whether demand holds to how much of the trajectory is priced into the stock.

Why this matters

Broadcom's AI chip revenue projections put dollar figures on a structural change in how the largest AI budgets are spent: purpose-built silicon is absorbing a growing share of capacity commitments, and merchant GPU shipment numbers alone no longer measure the pace of the AI build-out. For investors and enterprise buyers, the useful signal is the curve from roughly $58 billion to $230 billion over two years, because it shows how much of the AI capex cycle is already committed through design wins. The modest Q4 revenue miss, sitting outside the AI business, does not change that arithmetic.

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.

Reading the AI chip revenue road map

Margins and the competitive math

Why this matters

Related Articles