The global shortage of memory is expected to intensify, causing delays in project implementations and pushing prices up by 5 to 7 times. This warning came from Intel CEO Pat Gelsinger during the AI Infra Summit in Santa Clara, as reported by Seoul Economic Daily.
“When I mentioned at the start of last year that memory could become a significant bottleneck, many seemed unaware of it. It has indeed materialized, and the situation is set to worsen,” Gelsinger emphasized.
He noted that the available manufacturing capacities remain severely constrained. As a result of the memory shortage, various companies are postponing projects, particularly those producing budget and mid-range smartphones and laptops, who are struggling with component procurement.
While discussing price increases, Gelsinger did not specify which chip types were affected, the comparative period, or the purchasing conditions involved.
According to TrendForce, in the second quarter, DRAM manufacturers saw a revenue increase of 59.5% compared to the previous three months. Analysts attributed this surge to rising contract prices and heightened demand from AI servers.
Source: TrendForce.For the third quarter, TrendForce predicts a slowdown in contract price growth for standard DRAM to between 13% and 18% quarter-on-quarter. In contrast, the first quarter saw increases ranging from 93% to 98%, as reported by Reuters.
The shortage remains critical, with supplier inventories hovering near historic lows, while any additional capacity is primarily allocated to the server segment.
AI Demand Boosts Memory Needs
The expansion of AI infrastructure has significantly increased the demand for various types of memory. High Bandwidth Memory (HBM) is needed for accelerators, while servers require large volumes of DDR5 and high-capacity RDIMM.
Manufacturers are reallocating production resources to focus on these products, resulting in slower growth for memory supplies intended for personal computers and smartphones.
This situation is already impacting consumer electronics. According to Reuters, smaller smartphone and laptop manufacturers are modifying device designs, reserving components in advance, and in some cases, utilizing memory from older equipment. For some companies, the main constraint is not the price but the physical availability of chips.
Previously, CEO of SK Hynix, Kwon Oh-joon, identified 2027 as potentially the worst year in the industry’s history regarding supply. He estimates that demand could exceed production capabilities even after 2030.
Energy and Cooling Become Next Constraints
Gelsinger identified memory as just one of the barriers to further scaling of AI infrastructure. Other significant challenges include access to power and cooling systems.
“Another area is cooling. Beyond air cooling, there are liquid and microfluidic cooling systems, and we are investing in several of these directions,” stated the Intel CEO.
He explained that the industry is gradually shifting from selling individual processors to offering integrated systems. Companies like Nvidia and AMD are combining GPUs, central processors, networking equipment, and software into complete server racks.
Gelsinger believes that with this approach, manufacturers must consider the entire infrastructure holistically, including energy consumption, inter-component connections, and heat dissipation.
He also reaffirmed Intel's commitment to continue collaborating with Nvidia. Both companies are competing in the AI hardware market while simultaneously developing joint products for data centers and personal computers.
On September 10, it was reported that Huawei, Cambricon, MetaX, and Iluvatar CoreX had raised prices for AI accelerators, largely due to the global HBM shortage, according to sources cited by Reuters.