AI Infrastructure Demand Remains Strong: Wolfe Says Memory and Foundry Overcapacity Unlikely Before 2028

Recent semiconductor stocks have corrected sharply due to market concerns over a potential slowdown in AI-related capital expenditures. However, Wolfe Research senior analyst Chris Caso stated that the current AI chip market is constrained not by oversupply, but by insufficient production capacity. Memory suppliers and foundry manufacturers still cannot produce enough AI chips to meet demand, and overcapacity is not expected until at least 2028—possibly even later, according to some semiconductor equipment suppliers.

Caso made these remarks in a CNBC interview following SK Hynix ADR (SKHY-US)'s second-quarter earnings release and the recent sharp pullback in semiconductor stocks. He noted that memory manufacturers continue to face severe supply constraints, and TSMC's (TSM-US) advanced process capacity is nearly fully booked.

Caso emphasized that the semiconductor industry's biggest challenge is not weak demand, but the lack of physical space and production capacity to manufacture more AI chips. He stated, "There simply isn't enough physical space to produce more semiconductors."

He further explained that current capacity bottlenecks make market oversupply nearly impossible before 2028. "Even if overcapacity does occur, it won't happen before 2028," he said, adding that some equipment suppliers believe this timeline could be further delayed.

He stressed that memory suppliers are under severe supply constraints: "They can't produce more products, and this is precisely why we remain bullish on the memory industry."

Beyond manufacturing capacity, Caso also pointed out that the rapid expansion of AI infrastructure faces another emerging challenge: financing capability is becoming a new bottleneck for industry growth.

He noted that NVIDIA (NVDA-US) continues to invest in suppliers and AI companies, helping expand overall capacity and strengthen the AI ecosystem. However, he distinguished between equity investments and providing financing guarantees to customers.

While Caso acknowledged NVIDIA has the strongest balance sheet in the semiconductor industry, he expressed caution, noting that excessive financing guarantees could become future liabilities.

Semiconductor stocks faced significant selling pressure this month, reflecting market skepticism about the return on AI capital spending rather than weakening demand. The iShares PHLX SOX Semiconductor Sector Index Fund (SOXX-US) fell 27.43% in July but remains up 48.24% year-to-date and has gained 87.54% over the past year.

Another ETF, the VanEck Semiconductor ETF (SMH-US), which holds major chip stocks including NVIDIA (NVDA-US), Micron (MU-US), AMD (AMD-US), Broadcom (AVGO-US), and Intel (INTC-US), declined 23.12% in July but is up 24.97% year-to-date and 35.07% over the past year.

SOXX closed down 5.38% on Wednesday at $465 and rebounded 1.3% in Thursday's pre-market trading.

According to Benzinga Edge data, while SOXX shows weakening short- and medium-term momentum, its Momentum score remains in the 92nd percentile, and the long-term price trend remains positive. Caso concluded that with continued expansion in demand for AI data centers, HBM, and advanced processes, the real market constraint remains production capacity, not demand—thus, the long-term fundamentals of the memory industry and AI supply chain remain solid.