Broadcom CEO Dismisses AI Slowdown Fears, Keeps Revenue Targets Intact

Hock Tan, the chief executive of Broadcom, has firmly rejected the idea that a temporary lull in the development of advanced AI models will hurt his company’s bottom line. Speaking on Monday, Tan told CNBC that the chipmaker is sticking to its ambitious long-term revenue projections. This stance comes as a sharp contrast to the nervousness seen in the stock market, where investors reacted with caution to recent calls for a slower pace in artificial intelligence advancement.

The tension in the market was triggered by a weekend essay from Anthropic’s CEO, Dario Amodei, who argued for a more measured approach to developing frontier AI models. Amodei’s position, which was echoed by other industry leaders like Sam Altman and Elon Musk, suggested that the current sprint might be too fast. This sparked immediate concern among shareholders of AI infrastructure providers, who worry that a deliberate slowdown could reduce the massive demand for the specialized hardware these companies sell.

Market reaction to AI pace debate

The financial impact of this debate was visible immediately on Monday. Broadcom shares dropped by 4.8%, while a broader index of semiconductor stocks fell by 5.6%. The decline was particularly significant for Broadcom because Anthropic is one of its most critical customers for custom chips. The sensitivity of the market reflects the deep link between the rapid creation of new AI models and the physical infrastructure required to run them.

According to reporting by GN technics/ai (en-US), the concern stems from the possibility that if leading AI labs slow down their training cycles, they may not need to buy as many high-end processors. This fear has led investors to re-evaluate their expectations for future compute demand, causing a ripple effect across the sector that sells data-center components and networking hardware.

Broadcom maintains strong revenue forecasts

Despite the market jitters, Tan remains confident in his company’s trajectory. He recently forecast that Broadcom’s AI semiconductor revenue will reach $115 billion in fiscal 2027, with plans to double that figure to $230 billion by 2028. Tan clarified that this revenue stream includes both the custom accelerators used to train models and the networking chips that allow data to move efficiently within AI systems. He described the demand for this infrastructure as both strong and durable, suggesting that the underlying need for computing power is not likely to vanish.

A key part of Broadcom’s strategy is its growing relationship with Anthropic. Tan indicated that Anthropic is on track to become the company’s largest customer for custom chips in 2027, a position it is expected to hold in 2028 as well. This shift in customer hierarchy highlights why any commentary from Anthropic leadership has such a direct impact on Broadcom’s stock price. The trade-off for investors is that while the long-term outlook is bright, short-term sentiment is heavily influenced by the public rhetoric of the companies driving AI development.

Distinguishing training from daily usage

Tan offered a nuanced view on where the real demand lies. He distinguished between the intensive process of training new models and the day-to-day usage of those models, known as inference. While he acknowledged uncertainty about the future of training, he expressed strong optimism about inference. He argued that as AI products become more integrated into daily life, the need for efficient, large-scale inference will continue to grow rapidly.

Regarding the broader debate on AI safety and pace, Tan agreed that governance and safeguards are necessary. However, he rejected the idea that AI is an uncontrollable force. He likened the technology to a tool, comparable to the Industrial Revolution, arguing that it is designed to boost productivity and improve living standards. For Broadcom, the catch is that this optimism relies on the assumption that the industry will continue to invest heavily in the hardware needed to run these tools, regardless of how slowly the models themselves evolve.