Broadcom's CEO Sees Years of AI Demand But Warns of a Bigger Threat to the Boom

Key Points

• Broadcom CEO Hock Tan expects AI computing demand to remain strong for several years, with Intel CEO Lip-Bu Tan expressing similar optimism.
• Hock Tan identified capital availability as a potential constraint on AI growth, even as semiconductor supply shortages begin to ease.
• Broadcom's ambitious AI revenue outlook depends partly on customers such as Anthropic continuing to expand their computing capacity.

Many AI stocks have come under significant pressure over the past few weeks, including those in the semiconductor industry. So far in October, the iShares Semiconductor ETF NASDAQ: SOXX has fallen approximately 10%.

One key contributor to this decline was reports about Anthropic’s third-quarter revenue update. Although its sales increased by a blistering 45% quarter-over-quarter, a significant slowdown from the previous quarter's pace raised questions about the sustainability of overall AI growth.

In this context, recent comments made by Broadcom's NASDAQ: AVGO CEO Hock Tan are noteworthy, providing both reassurance and reason for caution among investors. Tan expects demand for AI compute to remain strong for several years and sees physical constraints eventually easing. However, he also highlighted concerns around capital needs and the importance of returns on AI investment.

Broadcom and Intel CEOs Point to Strong Compute Demand Ahead

In Singapore, Tan spoke at DBS Bank’s Asia Future Forward Conference, joined by Intel NASDAQ: INTC CEO Lip-Bu Tan. Notably, Hock Tan said that demand for AI compute could continue to exceed supply through 2029, or even 2030, depending on the pace of AI adoption. Lip-Bu Tan made similar statements, noting that Intel does not see a slowdown in demand over the next two to three years. This would be a strong setup for Broadcom, Intel, and other companies that supply compute.

Additionally, Tan expressed confidence in the industry’s ability to work around physical supply constraints. Tan noted that the industry has faced many bottlenecks over the past two years but has eventually overcome them. Looking ahead, he said that the memory shortage seen today “might not be that way a year from now” and that wafer shortages could ease by 2028 or 2029.

This is another positive sign for Broadcom, given that the company says supply constraints are limiting its growth. Despite expecting its AI chip sales to double to $115 billion in fiscal 2027, and double again to $230 billion in 2028, Broadcom said in its last earnings call that demand exceeds these figures. Tan notes that Broadcom has secured enough supply to meet these forecasts, suggesting potential upside if constraints ease further and customers have sufficient deployment capacity.

Hock Tan Highlights Capital as a Key AI Constraint

On the other hand, Tan emphasized a non-physical limitation: capital, saying “What I think might be the biggest constraint is capital." Here, Tan is referring to the ability of those purchasing AI compute to muster sufficient capital to sustain their buildouts. Tan estimated that across the U.S. and China, companies are spending around $200 billion a year on token generation. Comparatively, Tan estimated that companies are generating around $150 billion a year in revenue from token generation, concluding, “So AI is underwater, right?" This connects back to concerns around the slowdown in Anthropic’s revenue growth as the industry looks to flip this equation.

If companies like Anthropic cannot boost their profits relatively quickly, they will need to turn toward greater external financing to continue building. This could become increasingly difficult, given that Anthropic is seeking a massive IPO valuation of approximately $2 trillion, which some investors may believe is too lofty. Additionally, bond yields have risen significantly over recent months, increasing the cost of debt-based financing.

Tan also noted that today, the companies making much of the money in AI are the “picks and shovels” providers. This includes chip suppliers like Broadcom. However, he says that if the situation remains this way, the AI investment trend will not be sustainable. Ultimately, the underlying source of picks and shovels demand is AI model developers who need to achieve a return on investment to continue spending long term.

Tan hinted at one pathway through which model developers could generate a higher return on investment: challenging established software providers. Tan said that frontier model developers “will start to disintermediate all the software guys, one by one." In doing so, value could move up the stack to frontier model developers. While Broadcom is winning as a picks and shovels provider today, this scenario playing out would also benefit the chipmaker by helping its customers sustain purchases over the longer term.

Hyperscaler Earnings and IPOs Remain Top of Mind for AI Investors

Overall, while Tan highlighted real concerns regarding the sustainability of the AI buildout, his outlook for compute demand over the next few years remained encouraging. Still, the issues raised around capital and AI return on investment are important for investors to monitor. This puts a spotlight on not only the upcoming earnings calls of publicly traded AI hyperscalers but also the financial disclosures surrounding potential Anthropic and OpenAI IPOs.

Related to this, although its growth seems to have slowed, it is positive to see that Anthropic also reportedly saw its adjusted operating margin exceed 11% in Q3. This compares to its previously projected Q2 adjusted operating margin of 5.1%, suggesting improving profitability, although the comparison is between a reported result and an earlier projection. Continued gains on this front will be key to Anthropic sustaining chip purchases longer term. This is particularly relevant to Broadcom, as it expects Anthropic to become its largest XPU customer in 2027.