Broadcom (AVGO) Q3 FY26 earnings call: guidance and Q&A

Regarding the doubling of AI business next year and demand potentially exceeding supply, what are the supply bottlenecks, and what variables could increase the number if resolved?

Hock Tan stated that the company is careful and conservative with its outlook, acknowledging that demand could be significantly higher. He mentioned that the company has secured supply for the $115 billion FY27 and $230 billion FY28 outlooks, but deployment timing (getting chips deployed on time) is a key consideration. He also noted that the Singapore fab will start deploying for substrates in FY27 to address bottlenecks.

“But certainly, our customers want us to ship more. But we have secured supply, and we think it's the right number to put it to $115 billion.”

asked by Joseph Moore · answered by Hock Tan

Has any XPU decided to use Broadcom's Singapore capacity for substrates, and how does that fit into the supply picture? Also, clarification on whether the 10 gigawatts for Anthropic is separate from Google's numbers.

Hock Tan confirmed that the Singapore fab will start deploying for substrates in fiscal '27 to address key supply bottlenecks. He also clarified that the 10 gigawatts mentioned is specifically for Anthropic and not combined with Google's figures.

“Well, we are going to start deploying our Singapore fab for substrates. By the way, starting fiscal '27. And that would, I guess, address a key part of our supply bottlenecks.”

asked by Blayne Curtis · answered by Hock Tan

Verification of gigawatt counts for '27 and '28 for Anthropic and OpenAI, and inquiry about the implied revenue per gigawatt ($11B-$12B) compared to competitors' suggestions ($40B), asking if this is the right content level and how it trends with future XPU generations.

Hock Tan confirmed the gigawatt counts for the outlined customers, noting that the total demand across all 6 customers could be 30 gigawatts over the next two years, but the company's outlook is conservative based on actual deployment readiness. He explained that XPUs offer superior performance at less than half the cost of GPUs, making them highly cost-effective. He also clarified that dollars per gigawatt remain relatively stable ($20B-$30B) despite increasing chip performance, as higher performance chips mean fewer are needed per gigawatt.

“But the dollars per gigawatt will remain in the $20 billion to $30 billion level. of our content, but we have seen and we expect to see an acceleration in the number of gigawatts just to address the exponential compute demand required by our customers.”

asked by Stacy Rasgon · answered by Hock Tan

Clarification on gross margin impact in '27 and '28 due to XPU mix and memory costs. Also, why Google isn't taking more chips if they are supply constrained, and how much choice frontier labs have in silicon given their dependence on CSPs/funding partners.

Amie O'Toole stated that gross margin reflects revenue mix and increasing memory content in XPUs, and the company guides one quarter at a time. Hock Tan emphasized focusing on operating margin due to strong operating leverage. Regarding customer choice, he explained that the XPV platform helps financially insecure labs (Anthropic, OpenAI) fund their compute capacity, viewing it as a strategic investment. He believes these labs will eventually become hyperscalers, running their own data centers and choosing optimized silicon, rather than relying on CSPs.

“Don't forget, the rate -- with the rate of growth these startups. And we're talking about -- you know we have 6 customers, and we are creating this financial vehicles, as Amie described them for just 2 of them. Not all.”

asked by Vivek Arya · answered by Hock Tan

Beyond land, power, and shell, what other supply chain constraints impact the outlook, such as memory, substrates, or other components, and how are they being ameliorated?

Hock Tan confirmed that land, power, and shell (LPS) are significant, dictating deployment timing. Other constraints include leading-edge silicon availability, substrates (addressed by building Singapore fab for FY27), HBM memory, and system memory for AI servers. He described it as a multidimensional problem with various bottlenecks emerging at different times, which the company manages closely with its 6 customers.

“Then even deeper than that, we talk about substrates, which is specific issues, which is leading us to build our own substrate capacity at scale in our factory in Singapore. Jointly, we won our partners. And of course, we all know about memory, HBM memory.”

asked by Jim Schneider · answered by Hock Tan