AI chip demand to persist as semiconductor market tops $830bn

The recent selloff in AI chip stocks overlooks sustained demand for computing infrastructure as the global semiconductor market tops $830 billion, according to Advisers Capital Management.

"Investors are conflating efficiency gains with demand destruction, but the two are not the same," JoAnne Feeney, partner at Advisers Capital Management, said. "Each new generation of AI models requires exponentially more compute, not less."

Feeney's comments come as the Philadelphia Semiconductor Index has fallen roughly 15% from its 2026 peak, driven by concerns that advances in model efficiency — such as DeepSeek's lower-cost training methods — could reduce the need for chips and data center capacity. The global semiconductor industry generated more than $830 billion in revenue over the past year, fueled largely by AI-related demand, according to industry data.

The debate carries significant implications for investors. Nvidia, whose data center revenue has surged past $100 billion annually, trades at roughly 30 times forward earnings, reflecting expectations that AI infrastructure spending will continue to grow. If Feeney's view proves correct, the current selloff represents a buying opportunity in names tied to AI compute; if the bears are right, the sector faces a structural repricing.

Hyperscaler Spending Provides a Backstop

The scale of planned investment by major cloud operators supports Feeney's thesis. Microsoft, Amazon, and Alphabet have collectively committed more than $200 billion in data center capital expenditures for 2026, much of it directed at Nvidia's graphics processing units and custom AI chips from companies such as Broadcom and Marvell Technology. These spending plans, set months in advance, are unlikely to be reversed quickly even if model efficiency improves, she said. Taiwan Semiconductor Manufacturing Co., which produces the majority of advanced AI chips at its 3nm and 5nm facilities, stands to benefit regardless of which chip designer wins the largest market share.

Memory Stocks Face a Different Calculus

Feeney drew a distinction between AI compute chips and memory semiconductors. While she remains bullish on infrastructure plays tied to Nvidia and Advanced Micro Devices, she expressed caution on memory makers such as Micron Technology. Memory demand is more cyclical and tied to end-market consumption rather than the secular buildout of AI data centers, she said.

Micron shares have declined roughly 30% from their 2026 high as investors weigh the impact of potential oversupply in DRAM and NAND markets. The company derives about half its revenue from memory used in data centers, but the remainder is exposed to PC and smartphone markets, which remain sluggish. Samsung Electronics and SK Hynix, the other major memory producers, face similar dynamics, though both have benefited from high-bandwidth memory sales to AI customers.

For investors, the divergence between AI compute and memory stocks creates a potential trade. Nvidia, trading at 30x forward earnings, and AMD at 25x, reflect the market's continued willingness to pay a premium for AI exposure. Micron, by contrast, trades at roughly 12x forward earnings, a discount that may persist until memory demand from AI inference workloads materializes at scale. Feeney's framework suggests that the AI infrastructure buildout remains in its early stages, with data center capital spending by hyperscalers expected to exceed $200 billion in 2026.

This article is for informational purposes only and does not constitute investment advice.