Intel's chief executive, Lip-Bu Tan, has stated that the ongoing memory shortage is poised to intensify further. He also highlighted that securing adequate power supply and advancing cooling technology are becoming critical challenges for the semiconductor industry.

Speaking at an AI infrastructure forum held in Santa Clara, California, on the 15th, Tan was asked which industry problems he would urge startup founders entering the AI infrastructure space to tackle. In response, he pointed directly to the memory deficit. "Earlier last year, I mentioned that memory could become a major bottleneck, and at the time, not many people recognized that," Tan remarked. "Now, that prediction has materialized, and the situation is set to get even worse."

He elaborated, "Memory production capacity is extremely tight. Numerous projects have been delayed due to an inability to secure sufficient memory supplies, and prices have surged by five to seven times. Even for building mid-to-low-end smartphones and laptops, obtaining memory allocations has become exceedingly difficult."

As GPUs and other AI processors face performance limits in raw compute power, the importance of High Bandwidth Memory (HBM), often dubbed 'AI memory', continues to rise. Demand for HBM has exploded, but production capacity from Samsung Electronics, SK Hynix, and Micron has failed to keep pace, exacerbating the shortage. Furthermore, as manufacturing lines pivot to HBM production, a supply gap has also emerged for general-purpose memory.

Power and cooling were also identified as formidable obstacles. "Everyone is well aware of the severity of the power issue. Electricity is paramount, similar to how nuclear engineers and scientists once struggled to secure energy sources," Tan stated. "To achieve low power consumption in certain applications, chip architecture design and interconnect technology have become vitally important."

He continued, "Another key focus area is cooling technology. Beyond traditional air cooling, there are liquid cooling and microfluidic cooling solutions, and we are investing in a variety of these technologies."

Tan advised the industry to observe the broader shift towards system-level architectures in the semiconductor market. He referenced the prevailing industry trend where companies like Nvidia and AMD integrate GPUs, CPUs, networking components, and software into complete rack-scale systems for sale. "Jensen Huang and Lisa Su are already doing this," he said, mentioning the CEOs of Nvidia and AMD respectively. "Please pay close attention to our upcoming announcements in the substrate area."

He added, "It is crucial to address the load problem from the entire rack level, develop customized solutions based on customer needs, and collaborate closely with clients."

Tan also affirmed that Intel will maintain its partnership with Nvidia, a major shareholder in Intel and a competitor in the AI chip arena. When questioned about which areas Intel chooses to collaborate on versus compete in directly, he said, "A company cannot do everything. The key is to select and focus on the segments where we truly excel, while forging partnerships with others who possess complementary strengths and working together for mutual growth."