The AI Sell-Off Is About Pricing Power, Not Demand

Why markets are repricing AI infrastructure, from Broadcom to data centers, and what really matters for returns

Over the past few weeks, markets have clearly tightened. AI-exposed infrastructure stocks are down, neocloud names are under pressure, and even companies that delivered strong earnings have not been spared. Broadcom sold off. Data-center and power-linked equities followed.

This is not a collapse in AI demand. It is a reassessment of returns.

The market has moved past the question of whether AI will continue to grow. That question is settled. The new and more difficult question is who can turn massive AI capital expenditures into sustainable pricing power and acceptable returns on invested capital.

That shift in focus explains why even strong performers are being repriced. In Broadcom’s case, the concern is not execution or demand. AI-related revenue continues to grow rapidly. What the market is reacting to is product mix. As AI systems and rack-level solutions become a larger share of sales, margins can temporarily compress because these offerings include more third-party components and lower-margin hardware. This is not a structural problem for Broadcom. It is the market stress-testing whether AI growth can scale without diluting returns.

The same logic is now being applied across AI infrastructure. For companies like IREN, Cipher Mining, and other power-and-compute providers, the key uncertainty is no longer whether capacity will be needed. It is at what price, for how long, and under what contractual terms that capacity will be monetized. Growth in megawatts alone is no longer enough. Investors want confidence that utilization can be locked in at pricing levels that justify the capital deployed.

This shift is also visible at the institutional level. Recent research from Goldman Sachs on AI data-center capacity does not question the durability of AI demand. Instead, it focuses on utilization, power constraints, and capital efficiency. The takeaway is not oversupply. It is selectivity. Future returns will increasingly depend on who can secure long-term demand at attractive pricing rather than who can simply build the most capacity the fastest.

A useful contrast is Microsoft. To support its AI investment cycle, Microsoft effectively raised Office prices by roughly 30 percent, embedding AI costs into a product with exceptional pricing power. That move matters. It demonstrates how AI infrastructure ultimately gets paid for, not through raw usage alone, but through products and services where pricing can be passed through without destroying demand.

This is the dividing line the market is now drawing. Not between those who believe in AI and those who do not, but between companies that can monetize AI capex and those whose pricing power remains uncertain.

AI demand is not going away. But the era of unconditional capital allocation is ending. The next phase of this cycle will reward companies that can price their offerings, manage capital intensity, and earn returns above their cost of capital, not those that simply build the largest footprint.

This publication is for educational and informational purposes only and does not constitute financial, investment, or trading advice. Readers are solely responsible for their own investment decisions. The author may hold positions in the securities mentioned.