The global memory shortage is expected to worsen, with component scarcity already delaying projects and causing prices to rise by 5–7 times. This was stated by Intel CEO Lip-Bu Tan at the AI Infra Summit in Santa Clara, according to Seoul Economic Daily.
“When I mentioned early last year that memory could become a significant bottleneck, many, I believe, did not realize it. It has indeed happened, and the situation will get worse,” Tan emphasized.
According to him, available production capacities remain severely limited. Due to the memory shortage, companies are postponing projects, and manufacturers of low-cost and mid-range smartphones and laptops are struggling to procure components.
Speaking about price increases, Tan did not specify which types of chips were involved, the period of comparison, or the purchasing conditions considered.
According to TrendForce, in the second quarter, DRAM manufacturers’ revenue grew by 59.5% compared to the previous three months. Analysts attributed this trend to rising contract prices and demand from AI servers.
In the third quarter, TrendForce expects the growth of contract prices for standard DRAM to slow to 13–18% quarter-on-quarter. For comparison, in the first quarter, the figure was 93–98%, writes Reuters.
The shortage persists. Supplier inventories are near historical lows, and additional capacities are primarily directed to the server segment.
AI Increases Memory Demand
The expansion of AI infrastructure has increased demand for several types of memory. Accelerators require high-bandwidth HBM, while servers need large volumes of DDR5 and high-capacity RDIMM.
Manufacturers are reallocating capacities in favor of these products. As a result, memory supply for personal computers and smartphones is growing more slowly.
The situation is already affecting consumer electronics. According to Reuters, smaller smartphone and laptop manufacturers are changing device designs, pre-booking components, and in some cases using memory from old equipment. For some companies, the main constraint has become not the price, but the physical availability of chips.
Previously, SK Hynix CEO Kwak No Jung called 2027 potentially the worst year in the industry’s history in terms of supply. He estimates that demand could exceed production capacity even after 2030.
Energy and Cooling Become Next Constraints
Tan identified memory as just one of the obstacles to further scaling AI infrastructure. He also considers access to electricity and cooling systems as challenges.
“Another area is cooling. Besides air cooling, there is liquid and microfluidic cooling, and we are investing in several such directions,” noted the Intel head.
According to him, the industry is gradually shifting from selling individual processors to comprehensive systems. Nvidia and AMD are integrating GPU, central processors, networking equipment, and software into ready-to-use server racks.
Tan believes that with this approach, manufacturers need to consider the entire infrastructure, including power consumption, component connections, and heat dissipation.
He also confirmed Intel’s intention to continue collaborating with Nvidia. The companies are simultaneously competing in the AI equipment market and developing joint products for data centers and personal computers.
Earlier, on September 10, it was reported that Huawei, Cambricon, MetaX, and Iluvatar CoreX raised prices for AI accelerators. One of the main reasons cited by Reuters sources was the global shortage of HBM.
