Broadcom’s Hock Tan: The next AI bottleneck may be financing, not chips, power, or memory

Broadcom CEO Hock Tan is drawing attention to a different kind of constraint on artificial intelligence expansion. In remarks reported by Yahoo Finance, he suggested that AI’s next bottleneck is unlikely to be chips, power, or memory, and instead may be rooted in financing, a limit driven by how quickly customers and investors are willing to fund increasingly expensive AI buildouts.

The framing matters because it cuts against the common focus on whether semiconductor supply, electricity availability, or storage capacity will cap progress. Tan’s view, as presented in the report, is that those categories are not the core issue going forward. Instead, the question could become whether the economics of AI justify the scale of spending across the industry.

The comments also reflect a broader market tension around AI. Even as AI demand remains strong, the report indicates that investors are debating whether the large amounts being spent on AI systems will generate lasting profits. In other words, capital may become more selective, and that selectivity could affect which projects move forward and how quickly.

Broadcom’s position in this discussion is central. As a major supplier of semiconductors and infrastructure technology used across the AI stack, the company can benefit when customers build more compute and networking capacity. But if financing conditions tighten or if budgets become more constrained by returns expectations, the timing and magnitude of new orders can become harder to predict.

The reported remarks also underscore that “bottleneck” can mean different things at different phases. Early stages of AI rollouts often focus on getting hardware and supporting infrastructure in place. Later stages can shift toward deployment economics, including capex planning, financing cost, and the ability of customers to sustain operating returns as systems scale.

While the report attributes Tan’s view to AI’s next phase, it does not provide additional quantitative details, such as specific funding timelines, measures of cost of capital, or evidence of financing delays. It also does not name particular customer segments, datacenter operators, or geographies where financing is supposedly becoming the limiting factor.

For investors and industry participants, the implication is that the debate about AI may be moving from pure engineering capacity toward financial capacity. That does not mean chip supply or utility-scale power are unimportant, but it suggests that at some point, capital allocation decisions could dominate whether AI expansion continues at the same pace.

What to watch next is whether other AI infrastructure executives and major customers echo Tan’s “financing bottleneck” thesis. If the market narrative shifts in that direction, company guidance, order patterns, and commentary on customer spending discipline could take on added importance.