Broadcom CEO Keeps $230 Billion AI-Chip Target Despite Frontier Slowdown Calls

Broadcom Chief Executive Hock Tan says the company has not revised its AI semiconductor forecasts after calls to slow frontier-model development rattled investors. The company is still targeting $115 billion in fiscal 2027 AI semiconductor revenue and $230 billion in fiscal 2028, a forecast that depends on compute demand remaining durable even if the pace of training changes.

Tan made the case in a CNBC interview as investors reconsidered what a slower push toward more capable AI models might mean for the hardware build-out. Broadcom shares fell 4.8% in the session described by CNBC, while the iShares Semiconductor ETF fell 5.6% as concern about AI compute demand intensified.

Broadcom is selling both computing and the connections around it

The forecast is not solely a wager on chips used to train frontier models. Broadcom counts custom AI accelerators and networking chips in its AI semiconductor revenue. Accelerators perform AI computing, while networking chips help connect equipment inside AI systems. That puts the company in businesses that support both the creation of models and their use at scale.

Tan said demand for compute infrastructure for AI development and inference remains strong and durable. Inference is the day-to-day work of running a trained model to produce an answer or other output. Tan was especially upbeat about demand when companies turn inference into products, while distinguishing that from the uncertain outlook for training.

The two revenue streams inside the forecast

• Custom AI accelerators: chips designed for AI computing.
• Networking chips: components used to connect equipment in AI systems.

$115 billionFiscal 2027 AI semiconductor revenue forecast

Broadcom’s forecast remains $115 billion for fiscal 2027.

$230 billionFiscal 2028 AI semiconductor revenue forecast

Broadcom’s forecast remains $230 billion for fiscal 2028.

A customer relationship sharpens the exposure

Anthropic is central to why the slowdown debate lands so directly on Broadcom. Tan said Anthropic is on track to become Broadcom’s largest custom-chip customer in 2027 and keep that position in 2028. If that expectation holds, a customer advocating moderation in frontier-model development would also be Broadcom’s biggest custom-chip buyer.

That does not mean Tan rejects the safety concern. He said AI needs governance and safeguards, but argued that generative AI and frontier models will create substantial economic value. His position is that guardrails and large-scale deployment can coexist, rather than that safety measures necessarily require a lasting retreat in infrastructure demand.

The forecast now turns on deployment

The immediate disagreement is not over Broadcom’s targets: Tan says they are unchanged. It is over the engine beneath them. A meaningful slowdown in frontier development could alter demand for training infrastructure, but Tan is betting that the commercial use of already-trained models will keep requiring accelerators and networking at a scale that supports the company’s forecast. The interview offers a confident management view, not proof that those long-range targets will be reached.