Broadcom said it would stick to its long-term AI semiconductor revenue target despite Anthropic’s call to slow the pace of AI development.

CNBC reported on Sept. 14 that Broadcom CEO Hock Tan (혹 탄) drew a line under suggestions that a potential slowdown in AI model development would shake the company’s earnings outlook.

The comments came after Anthropic CEO Dario Amodei (다리오 아모데이) posted a piece over the weekend calling for a slowdown in the development of high-performance AI models. Sam Altman and Elon Musk backed the post, and the market moved to recheck projections for computing demand at AI infrastructure companies. Broadcom shares fell 4.8%, and the broader semiconductor sector also weakened. The iShares Semiconductor ETF dropped 5.6%, and shares of data centre component makers also slid.

Broadcom, however, maintained its existing guidance. When Jim Cramer asked whether the debate over slowing AI had made him rethink AI semiconductor prospects for fiscal 2027 and fiscal 2028, Tan replied, "Not at all." He said demand for AI development infrastructure, along with inference demand as finished models are used in real products and services, remained strong and highly durable.

In its fiscal 2026 third-quarter earnings on Sept. 2, Broadcom projected fiscal 2027 AI semiconductor revenue of $115 billion and forecast that it would double to $230 billion in fiscal 2028. That figure was seen as a key metric that exceeded market expectations. Broadcom’s AI revenue includes custom AI accelerators and network chips for AI systems.

There was also a clear reason for the market sensitivity. Broadcom has Anthropic as a key custom chip customer, and Tan said Anthropic would become Broadcom’s largest custom chip customer in 2027 and keep that position in 2028. Google has been cited as Broadcom’s biggest custom customer to date. Google has worked with Broadcom on designing its tensor processing units.

Tan put particular weight on inference demand. "I do not know about training, but demand to productise inference will continue very strongly," he said. He was highlighting that even if the pace of model training is adjusted in part, demand in the stage of attaching AI to real services may not weaken.

Broadcom did not deny the need for AI regulation itself. While Tan agreed with Amodei’s view that some restrictions were needed for AI, he drew a line under the idea that the technology could become uncontrollable. "With any tool, governance and safeguards matter in how it is used," he said, adding, "It is not a living animal that runs wild on its own."

Broadcom executives also stressed AI’s economic utility. Tan said generative AI and frontier model development would create enormous value and ultimately become a tool to raise living standards for society and humanity.

The market’s next focus is expected to be whether worries about a slowdown in AI training investment lead to actual cuts in capital spending, or whether inference-led demand highlighted by Broadcom offsets it. Broadcom made clear it would not change its long-term revenue target for now.