Wolfe Research estimates Broadcom could see $200B in AI revenue by 2028

Wolfe Research estimates Broadcom could generate up to $200 billion in AI revenue by 2028, driven by its involvement in XPV, an AI infrastructure financing vehicle with Apollo and Blackstone supporting over 20 gigawatts of compute capacity. The projection, dated August 13, 2026, assumes roughly 14 GW tied to OpenAI and Anthropic at $10-15 billion per gigawatt, potentially accounting for over 80% of Broadcom's projected 2028 revenue of $245 billion. However, Broadcom's $30 billion guarantee on a $36 billion tranche for Anthropic infrastructure adds risk, though Wolfe sees limited oversupply risk through 2028 due to foundry constraints.

Via broadcomfoundation.org Wolfe Research estimates Broadcom could see $200B in AI revenue by 2028 The chipmaker's bet on AI infrastructure financing could account for most of its projected revenue, but a $30 billion guarantee adds real risk to the equation Broadcom has quietly positioned itself at the center of a massive AI infrastructure financing machine. Wolfe Research now thinks that bet could be worth up to $200 billion in revenue by 2028. The estimate stems from Broadcom’s involvement in XPV, a newly established AI infrastructure financing vehicle created in partnership with Apollo and Blackstone. The platform is designed to support over 20 gigawatts of compute capacity by 2028, with an initial focus on powering Anthropic’s AI ambitions. The math behind the mega-forecast Wolfe’s analysis, dated August 13, 2026, breaks the projection down with satisfying specificity. Of the 20-plus GW of planned capacity, roughly 14 GW could be tied to OpenAI and Anthropic. At a projected cost of $10 billion to $15 billion per gigawatt, that translates to $140 billion to $200 billion in potential revenue for Broadcom. To put that in perspective, Broadcom’s total consensus revenue for 2028 sits at around $245 billion. If the upper end of Wolfe’s AI estimate materializes, it would mean AI alone accounts for more than 80% of the company’s entire top line. This isn’t coming out of nowhere. Wolfe’s earlier analysis from January 30, 2026 projected Broadcom’s AI ASIC revenue at roughly $44 billion in 2026 and $78.4 billion in 2027, showing a steep growth trajectory that makes the 2028 figures look like a natural extension rather than a fantasy. Broadcom CEO Hock Tan has added fuel to the fire by publicly stating a line of sight to over $100 billion in AI semiconductor revenue for fiscal 2027 alone. Google’s TPU program as a growth engine Part of what gives Wolfe confidence is the scaling of Google’s Tensor Processing Unit program. The firm’s January estimates indicated Google’s TPU shipments reaching approximately 3.3 million units in 2026, climbing to 5.1 million in 2027, and potentially hitting 7 million annually by 2028. Broadcom designs the custom silicon behind Google’s TPUs, making this scaling trajectory a direct revenue driver. The $30 billion catch Not everything in the Wolfe report reads like a growth stock pitch deck. Broadcom has committed to guaranteeing $30 billion in residual value on a $36 billion initial tranche dedicated to 1 GW of Anthropic infrastructure, targeted for delivery by mid-2026. Wolfe doesn’t seem panicked about this risk, at least not yet. The firm points to physical constraints at foundries like TSMC that should prevent a chip oversupply scenario through 2028. Still, the report acknowledges that industry-wide backstops, where chipmakers guarantee infrastructure value, could create compounding risk exposure if multiple companies face writedowns simultaneously. Wolfe values Broadcom at approximately 20 times consensus earnings per share for 2027 during its analysis of these financing risks. What investors should watch For the bull case to play out, two things need to hold. First, AI compute demand from OpenAI, Anthropic, and Google needs to keep accelerating at roughly the current pace. Second, the infrastructure Broadcom is helping finance needs to retain its value over multi-year time horizons. The bear case is simpler: AI training costs could decline faster than expected as efficiency improves, leaving expensive infrastructure stranded. If next-generation chips deliver the same performance at lower power requirements, 20 GW of capacity could prove excessive. And that $30 billion guarantee would shift from theoretical risk to very real liability. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy https://cryptobriefing.com/editorial-policy/ .