Broadcom's AI XPU Revenue Could Reach $230B by FY2028

Broadcom's fiscal third-quarter 2026 earnings call on September 2 reshaped the company's long-term AI revenue narrative. CEO Hock Tan told analysts that Broadcom now has line of sight to $230 billion of AI semiconductor revenue in fiscal 2028, roughly four times the $58 billion it expects in fiscal 2026. At the same time, the fiscal 2027 AI revenue target moved up from the prior 'more than $100 billion' reference to roughly $115 billion. Tan added that supply for both years is already secured, making these forecasts partly a statement about manufacturing capacity rather than demand alone. The numbers are striking because the $230 billion figure is not total company revenue; it is specifically AI semiconductor revenue. If delivered, fiscal 2028 AI revenue would average nearly $58 billion per quarter, almost twice the $29.6 billion total revenue Broadcom reported in the record quarter just ended. That would elevate Broadcom from a major AI component supplier into one of the largest AI infrastructure businesses in the world. The near-term trajectory makes the math at least directionally plausible. In fiscal Q3 2026, which ended August 2, total revenue rose 86% year over year to $29.6 billion and net income more than tripled to $13.1 billion. AI semiconductor revenue climbed to a record $16.7 billion, up 221% year over year, after $10.8 billion in fiscal Q2. Guidance for fiscal Q4 calls for $21.7 billion in AI semiconductor revenue, implying sequential growth of more than $5 billion for the second consecutive quarter. For the full fiscal 2026, Broadcom expects $58 billion of AI semiconductor revenue, up 186% from about $20 billion last year. The fiscal 2027 target of about $115 billion would require another doubling, and the fiscal 2028 line of sight of $230 billion would require a second doubling. Context matters because Broadcom's AI franchise is built on custom accelerators, often called XPUs, designed in close partnership with large cloud and AI platform companies. Tan disclosed that the bulk of the outlook rests on six custom accelerator customers. That is a short customer list for a forecast of this size, and it means the revenue bridge is unusually concentrated. Broadcom's value in these programs comes from co-designing high-performance ASICs and then integrating them with advanced packaging, memory, and interconnect. The 'supply secured' statement likely refers to leading-edge wafer starts, advanced packaging capacity such as CoWoS, and high-bandwidth memory allocations. If those supplier commitments are real, they reduce one category of execution risk, but they also depend on multi-year supplier performance. Tan also said demand 'actually exceeds this outlook,' which is a bullish signal but remains an unaudited forward-looking claim from the company's own earnings call. The market impact is significant for how investors model AI infrastructure spending. Broadcom's projection implies that its end customers are planning multi-year buildouts at a scale that may have been underappreciated. For context, $230 billion in AI silicon alone would support an enormous downstream buildout in data centers, power, networking, memory, and thermal management. That creates read-through for the entire AI hardware stack, not just Broadcom. It also sharpens the comparison with Nvidia's data center trajectory, although Broadcom's custom XPU model is structurally different from Nvidia's merchant GPU model. The fiscal 2028 figure would require quarterly AI revenue to roughly match the total company revenue of the record quarter just reported, a level that few suppliers have achieved in a single segment. The biggest variable is not demand in the near term but deployment. Tan can secure chip supply, but he cannot promise that data center construction, power availability, memory attach rates, packaging yields, or customer validation will keep pace. If Broadcom sustains the guided $21.7 billion in fiscal Q4 and steps toward a $25 billion to $30 billion quarterly AI run rate in fiscal 2027, the $115 billion target becomes credible. The $230 billion fiscal 2028 number still requires another step change in both customer commitments and supply-chain capacity. Forward-looking investors and AI infrastructure observers should watch quarterly sequential AI revenue, the number and identity of XPU customers, and whether Broadcom begins to report backlog or customer-specific progress rather than only 'line of sight' estimates. The next several quarters will show whether the September 2026 forecast was a capacity-secured roadmap or an aspirational peak-cycle target.