Broadcom's chief executive, Hock Tan, pushed back on Monday against market concerns that a slowdown in frontier AI model development could hurt the chipmaker's business, reiterating that the company's long-term revenue targets remain unchanged. His comments came in response to a weekend essay by Anthropic CEO Dario Amodei advocating for a more measured pace of model development—remarks that rattled investors across the sector.

Amodei's position, which drew support from OpenAI CEO Sam Altman and Elon Musk, prompted investors in AI infrastructure suppliers to reassess their compute demand expectations during Monday's trading session. The fallout was immediate: Broadcom, which counts Anthropic among its most important custom chip customers, saw its shares fall 4.8% on the day, while the iShares Semiconductor ETF dropped 5.6%. Other data center component suppliers also took a significant hit.

During Monday's interview, the program host asked Tan whether the debate over slowing AI development would lead him to reconsider Broadcom's performance forecasts for its AI chip business in fiscal 2027 and fiscal 2028. His response was unequivocal: "No, not at all." He added, "We see sustained, and I believe highly durable, demand for compute infrastructure—both for AI research and frontier model training, as well as for the inference compute needed to roll out products globally."

At Broadcom's fiscal third-quarter 2026 earnings call on September 2, Tan projected the company's AI chip business would generate $115 billion in revenue in fiscal 2027, with that figure expected to double again to $230 billion in fiscal 2028—a target that exceeded market expectations and stood out as a highlight of the earnings report. Broadcom's AI revenue stream encompasses custom AI accelerators and networking chips used in AI systems.

Tan also noted that Anthropic is expected to become Broadcom's largest custom chip customer in 2027 and retain that position through 2028, which explains why investors in the company are so sensitive to Amodei's remarks. Google, meanwhile, has long been regarded as Broadcom's biggest custom silicon client, with the two companies having jointly designed Google's tensor processing units.

In the Monday interview, Tan expressed particular optimism about "inference"—the everyday usage scenario for AI models once training is complete. "I can't speak to the training side," Tan said, "but once you move toward productizing inference capabilities, I believe demand will remain very, very strong."

Amodei's weekend essay intensified an ongoing debate over whether the AI industry is moving too fast in developing high-performance models. He proposed a three-step approach to decelerate development without "sacrificing commercial advantage or America's leadership position in AI." Tan said he agrees with Amodei that some guardrails on AI are necessary, but he is far less concerned about where the technology is headed.

"Like all tools, it's important to establish governance rules and safety measures around how they are used," Tan said. However, he added, "It's not a living being that's going to run amok on its own." Instead, he emphasized AI's potential to boost productivity. "AI, generative AI, and the emergence of these frontier models... will create enormous value," Tan said, likening the moment to the Industrial Revolution that began in 18th-century Britain. "At the end of the day, it's just a tool that can move our society and humanity toward a higher standard of living," he concluded.