Navigating a Deepening AI Chip Crunch

Download this article in PDF format.

Semiconductor manufacturers recognized early that the transition toward artificial intelligence-centric computing was going to be a major movement, but they may not have understood just how quickly demand for AI chips would grow. According to a new Sourceability report, NVIDIA’s H100 and AMD’s MI300 series GPUs saw near-immediate sellouts across all production cycles in 2024.  

“A growing segment of AI startups and enterprise AI labs joined major cloud providers in clearing the shelves,” the company reports in “AI chip shortages deepen amid tariff risks.” “As this period of soaring demand shows no signs of slowing, it is reshaping the memory hierarchy.” It’s also redefining performance benchmarks across industries, which are using foundational model training, experimenting with high-powered large language models (LLMs) and launching large-scale AI deployments.

These and other developments are now straining global computing infrastructure, and three key components are under the biggest strain: high-bandwidth memory (HBM), high-performance graphics processing units (GPUs) and advanced solid-state drives (SSDs). “Once confined to specialized clusters,” Sourceability says, “these parts are now essential building blocks for every hyperscale data center racing to meet the next wave of AI acceleration.”

Five Trends to Watch

Here are five more signs currently foreshadowing a deeper AI chip shortage in the near future: 

1. Uncertainty over tariffs continues. Sourceability says the convergence of growing AI demand, current semiconductor capacity and ongoing geopolitical headwinds is placing “unprecedented stress” on the semiconductor supply chain. “With memory requirements per chip at all-time highs and AI workloads forecasted to grow 25-35% year-over-year through 2027,” it says, “the start of protracted shortages of these sought-after components is actively unfolding.

2. Core minerals are in short supply. In “4 Issues Exacerbating the Looming AI Chip Shortage,” Zac Amos writes about how AI chips need rare and hard-to-source materials (i.e., gallium, germanium, cobalt and tungsten), for their semiconductors. “These minerals are in short supply, and many come from politically volatile regions or countries with tight control over them,” he explains, noting that China has a 60% share of worldwide rare earth elements output. These resources have become “essential to the production of chips,” he writes, “and demand will only keep growing.”

3. Memory components are experiencing 6-12 month bottlenecks. According to Searchability, HBM3 (the memory standard used to power AI accelerators) is at the center of a new supply bottleneck. SK hynix, Samsung and Micron—the three dominant producers of HBM chips—have been operating near full capacity. These companies are currently reporting 6-12 month lead times amid a backlog of orders.  

4. The pressure extends beyond front-end chip fabrication. The industry is also dealing with the complex constraints of packaging and integration needed to bring high-end AI-related components to the market. “Many extra wafers produced thanks to upscaled fab capacity are now in line for the advanced packaging steps required to make them usable,” the company says.

5. Chip makers and distributors are soldiering through the uncertainty. According to AI News, Taiwan Semiconductor Manufacturing Company (TSMC) is one company that’s facing “unprecedented AI chip demand that it cannot fully satisfy.” At TSMC’s annual shareholders meeting, the company reported that its business continues to benefit from “explosive growth” in artificial AI applications. It also emphasized that AI chip demand remains “very strong” and consistently outpaces the company’s ability to supply. The company’s customer roster includes tech giants Apple and Nvidia, both of which have been major drivers of AI-related semiconductor demand. 

More Disruption on the Near Horizon

AI News says the mismatch between AI chip demand and available supply has become a defining challenge for companies like TSMC, which is now “actively working to increase production capacity to satisfy [its] customers,” though the scale of demand continues to strain even the world’s most advanced semiconductor manufacturing capabilities. 

“This capacity constraint reflects broader industry dynamics where AI applications—from data center processors to consumer devices,” AI News reports, “require increasingly sophisticated and powerful chips that only a handful of manufacturers can produce at scale.”