Broadcom's AI Chip Revenue Triples to $16.7 Billion, But Cautious Guidance Sends Shares Lower

Broadcom Inc. reported an 86 percent jump in quarterly revenue on record demand for the custom artificial-intelligence chips it builds for the largest technology companies, then watched its stock slide after warning that near-term growth, while still explosive, would not accelerate quite as fast as investors had priced in.

The Palo Alto, California-based semiconductor and infrastructure-software maker said fiscal third-quarter revenue reached $29.6 billion, topping Wall Street estimates near $29.4 billion, according to results filed with securities regulators after markets closed on Wednesday. Adjusted profit came in at $3.32 a share, also ahead of forecasts. Yet shares fell as much as 6 percent in after-hours and early trading, as investors focused less on the beat than on guidance for the current quarter that fell just short of the loftiest expectations built up over a year of surging AI infrastructure spending.

The numbers

Revenue from Broadcom's AI semiconductor business, which includes custom accelerator chips and networking gear sold to a handful of large cloud and AI companies, more than tripled from a year earlier to $16.7 billion, up 221 percent and 54 percent higher than the prior quarter. That business now makes up more than half of the company's $20.8 billion semiconductor solutions segment; the remaining $8.8 billion of revenue came from infrastructure software, up 29 percent, largely reflecting the integration of VMware. Non-GAAP operating income rose 92 percent to roughly $20.1 billion, and free cash flow hit a record $13.7 billion, equal to 46 percent of revenue. GAAP net income totaled $13.1 billion.

For the fiscal fourth quarter, Broadcom guided to consolidated revenue of about $34.8 billion, a 93 percent year-over-year increase but modestly below the roughly $35 billion analysts had penciled in, and AI semiconductor revenue of $21.7 billion, up 236 percent from a year earlier. The board also declared a quarterly cash dividend of 65 cents a share, payable September 30 to shareholders of record as of September 21, part of the same disclosure filed with the Securities and Exchange Commission.

How Broadcom got here

The results extend a run that has made Broadcom, alongside Nvidia, one of the two dominant beneficiaries of the buildout of AI data centers. Where Nvidia sells general-purpose graphics processors, Broadcom has built its AI business chiefly around custom accelerator chips, known as XPUs, designed jointly with individual cloud customers for their own specific workloads, plus the networking silicon that links thousands of chips together inside a data center. Chief Executive Hock Tan has argued that a chip co-designed for a particular large language model can outperform an off-the-shelf GPU on power and cost, a pitch that has drawn a widening roster of hyperscale customers over the past two years. The chip business now sits alongside an enterprise software arm built around VMware, the virtualization company Broadcom acquired in 2023, which gives the combined company a steadier stream of software revenue to cushion swings in chip demand.

On a call with analysts, Tan said Broadcom now counts six customers for its custom accelerators and detailed multi-year commitments behind the growth: Google, its longest-standing XPU partner, is planning to deploy tens of billions of dollars of custom AI chips annually, while Meta has committed to roughly 3 gigawatts of Broadcom-designed chips across three generations through 2028. Tan said Anthropic, which deployed about 1 gigawatt of its custom accelerator this year, is expected to scale that to 5 gigawatts in 2027, and that OpenAI's first custom chip, internally code-named Jalapeño, remains on track for deployment in 2027. Tan told analysts Broadcom has "secured the supply" to roughly double AI semiconductor revenue again in fiscal 2027, to about $115 billion, and pointed to a trajectory that could reach $230 billion by fiscal 2028 if current customer commitments hold.

The acceleration has been rapid even by the standards of the current AI buildout. Broadcom's AI semiconductor revenue was $8.4 billion in the fiscal first quarter, then $10.8 billion in the second quarter, before jumping to $16.7 billion in the third — a trajectory the company has repeatedly said is constrained more by its own manufacturing capacity and that of its foundry partners than by customer demand. Broadcom itself is a product of decades of consolidation in the chip industry: the company traces its roots to Hewlett-Packard's original semiconductor division, was rebuilt as Avago Technologies before taking the Broadcom name in a 2016 acquisition, and has since folded in networking, security and enterprise software businesses, most recently VMware.

Who is exposed

The results matter well beyond Broadcom's own shareholders. The company's AI chip revenue is now concentrated among a small number of the world's best-funded technology firms, which means Broadcom's growth is effectively a proxy for how much Google, Meta, OpenAI and Anthropic are willing to keep spending on data-center buildouts, and any slowdown among them would show up quickly in Broadcom's numbers. That concentration cuts both ways: Broadcom has a long-term agreement to supply Google with custom AI processors through 2031, but Google has also signed a competing custom-chip arrangement with rival chipmaker Marvell Technology, a reminder that even Broadcom's largest customers are diversifying their suppliers rather than committing to one vendor indefinitely. Broadcom employees and suppliers across its chip-design and manufacturing partners, including Taiwan Semiconductor Manufacturing Co., are also tied to the pace of that spending, as are investors in index funds that now carry outsized exposure to a handful of AI infrastructure names. The concentration has become a recurring worry among strategists who note that Broadcom's fortunes now hinge on continued heavy capital spending by roughly half a dozen companies; if any one of them were to pause or renegotiate a multi-year commitment, the effect on Broadcom's reported growth would be immediate and large, a risk the company itself has flagged in the quarterly filing package it lodged with regulators even as it keeps signing new multi-year supply agreements.

Reaction and what's next

The stock reaction reflected a familiar pattern this earnings season: a solid beat overshadowed by guidance that, while still describing triple-digit growth, undershot the more bullish end of what some investors had modeled. Ahead of the report, Morgan Stanley had told clients the central question was whether Broadcom's fiscal 2027 AI outlook would approach $150 billion, well above the bank's own $120 billion estimate, and warned that any shortfall against those elevated hopes could drive volatility regardless of the quarter's actual results.

"Demand for our custom AI accelerators and networking continues to be very strong," Hock Tan said in the earnings statement, noting that AI semiconductor revenue grew 221 percent from a year earlier.

Not every analyst read the post-earnings drop as a verdict on the business. Morningstar analysts described the sell-off as an overreaction to what they characterized as deliberately conservative guidance, arguing Broadcom's stated 2027 and 2028 AI targets likely understate what the company's existing customer commitments can deliver. Shares of Broadcom remain well below the record high reached earlier this year even after roughly tripling over the past two years, a run chronicled as one of the AI boom's biggest winners even as its valuation has drawn scrutiny.

Attention now turns to whether Broadcom's hyperscale customers reaffirm their capital-spending plans when they report their own quarterly results in the coming weeks, and to Broadcom's fiscal fourth-quarter report expected in early December, which will show whether the $34.8 billion guidance proves conservative, as Morningstar argued, or whether AI infrastructure spending is beginning to moderate from its recent pace. The steady dividend and record free cash flow reflect a company betting that the multi-year customer commitments already on its books will keep converting into revenue regardless of near-term stock swings.