Cheaper Inference Could Expand Demand for AI Chips and Data Centers
Advisors Capital Management’s JoAnne Feeney argues that cheaper AI inference should expand usage and sustain demand for data centers and chip suppliers such as Nvidia and Broadcom, even as intensifying model competition erodes providers’ ability to defend margins. She sees the recent AI-stock volatility as a reassessment of where durable advantages lie, not evidence that infrastructure demand has broken. Feeney is more cautious on memory stocks including Micron, arguing that today’s strong pricing will eventually draw new capacity and revive the sector’s familiar cycle.
Lower inference costs can expand, not erase, chip demand
JoAnne Feeney sees the recent volatility in AI-linked equities as a repricing of where durable advantages sit, rather than a verdict that the infrastructure buildout has ended. Investors, she says, are beginning to distinguish between companies supplying the physical and operational foundation for AI and companies selling models into an increasingly competitive market.
That distinction matters because implementation costs are falling. New high-powered models from China and lower token prices have raised the prospect that model providers will struggle to sustain differentiation—and, in turn, their margins. Feeney’s conclusion is not that cheaper AI means less hardware demand. Model competition can pressure the margins of companies selling AI models, while chip suppliers and data-center operators have capabilities beyond simply running models. Feeney pointed to security, stability, reliability, and data handling as sources of a more durable moat that could help sustain margins.
Nvidia and Broadcom, both long-held client investments for Advisors Capital Management, sit among the suppliers to data-center construction and AI enablement that Feeney considers more defensible. As model capabilities converge, she expects the gap between competing models to shrink over time.
Feeney’s core demand argument is that lower inference costs raise demand, which in turn spurs demand for more data centers, Nvidia chips, and Broadcom-designed chips.
Ed Ludlow framed that case through Kimi K3, which he described as a 2.8 trillion-parameter model priced at $3 per million input tokens and $15 per million output tokens. His interpretation was that Chinese AI companies may be seeing demand at those price levels and that a shift toward inference could require more Nvidia- or GPU-based infrastructure. Feeney broadly agreed, while stressing that competition among models will take time to play out.
The investment question, she said, is therefore not simply whether AI demand exists. It is where profit margins can remain intact as demand grows. In Feeney’s view, lower-cost inference can increase usage while making it harder for model makers to maintain a premium based on differentiation.
Memory still follows the capacity cycle
JoAnne Feeney is more concerned about memory, including Micron. A low earnings multiple is not automatically a bargain: Micron’s valuation—discussed at roughly six to 6.5 times forward 12-month earnings—reflects a risk that the unusually favorable pricing environment will not persist.
The sector’s sales and profits have been lifted by price increases running into the hundreds of percent, Feeney said. But memory suppliers will add capacity, and the eventual increase in supply should pressure prices, including in high-bandwidth memory. There are three players at the leading edge, alongside suppliers at the trailing edge, and all are building more capacity.
Feeney rejects the claim that memory is no longer cyclical. The industry’s incentive structure has not changed, she argues: each producer has reason to build more capacity than its current market share alone would justify because it wants to gain share. Collectively, that behavior has historically produced too much capacity.
The timing problem is that Micron may continue to benefit from strong pricing and profit growth before the next turn arrives. Investors do not know when the market will receive concrete news about capacity additions from the major suppliers, or how sharply prices will fall once that supply comes online. Feeney said stocks tend to decline ahead of the reported deterioration in pricing and earnings.
Micron was up more than 10% intraday during the discussion, and Ludlow noted that it was the company’s biggest gain in about three weeks. Feeney’s response was that such moves do not resolve the larger question. If memory prices collapse, she said, earnings growth can turn negative—as it has in Micron’s past. That possibility helps explain both the low multiple and why she considers the stock a riskier place to invest.
Volatility can reflect diversification, not a broken demand case
JoAnne Feeney does not treat a day, week, or month of price moves as a reliable explanation of what companies are worth. The relevant questions are how earnings support fundamental valuation and what cash flow a company can generate over time. Big run-ups and reversals are noise unless they change those longer-term economics.
That frame also applies to the pullback in AI stocks. Feeney does not consider it inherently problematic: investors may have become overly concentrated in particular names, and diversification can be a sensible response even when it produces sharp volatility. Her priority for clients is avoiding an overweight position in any single company rather than trying to interpret every abrupt rally or decline.