Intel Lip-Bu Tan Warns AI-fueled Memory Crunch to Deepen Next Year as Memory Prices Soar

AI Boom Drives Up Memory Demand, Supply Chain Pressure Continues to Escalate

Intel CEO, Lip-Bu Tan, recently issued a warning said at the AI Infrastructure Summit in Santa Clara, California, saying that as AI demand continues to climb, the memory shortage will worsen further next year. He noted that prices for some memory products have already risen five to seven times, and many companies are slowing the pace of business expansion because they cannot secure enough memory supply. As demand for high-bandwidth memory (HBM) surges in the AI sector, supply pressure on general-purpose DRAM is also increasing.

Tan also mentioned that when making entry-level smartphones or laptops, you can't spend 70% to 80% of the product cost on memory. If the rise in memory prices continues, it could affect production of entry-level smartphones and PCs, which are relatively less profitable.

Advanced Packaging Becomes Key, Intel Concentrates on EMIB Technology

Tan emphasized the importance of advanced packaging that more tightly links memory and compute chips in the AI era. He said CPU, AI accelerators, HBM, and I/O chips should be connected within a single package to shorten data travel distance and improve processing speed and power efficiency.

Intel is targeting the AI Semiconductor market with its own advanced packaging technology, EMIB, at the forefront. It is a technology that connects only the necessary parts with small silicon bridges instead of a large silicon interposer, and is Intel's key packaging technology competing with TSMC's CoWoS.

Regarding package substrates for AI chips, Tan explained that there are only four key substrate suppliers, two in Japan and two in Taiwan. It is necessary to secure capacity and stabilize packaging production yield and performance.

AI Architecture Moving Toward Heterogeneous Computing, Intel Partners with Nvidia to Expand the Market

Tan predicted that AI infrastructure centered on GPUs will shift to a heterogeneous computing architecture in which CPUs, GPUs, NPUs, and application-specific accelerators share roles. Citing SambaNova and Cerebras Systems, in which Intel has invested, he emphasized the need for diverse AI accelerator architectures that can improve power efficiency.

He also said that Intel will also continue cooperation with Nvidia while Intel will combine Intel CPUs, Nvidia GPUs, and NVLink to target new markets.