Intel CEO Chen Lifu: The shortage of memory chips will last at least until 2028.

On February 3rd, local time, at the Cisco AI Summit, Intel CEO Pat Gelsinger discussed the shortage of storage chips, stating: “It will not be alleviated until 2028.” Gelsinger’s projection of “2028” far exceeds the consensus on Wall Street, implying that the duration of this “super cycle” may be underestimated.

Gelsinger revealed that after discussions with two major storage manufacturers, the conclusion is that supply tightness will persist at least until 2028.

Currently, the driving force behind storage chips is no longer the cyclical fluctuations of consumer electronics but rather the structural squeeze from AI infrastructure. Gelsinger explicitly pointed out, “Artificial intelligence will ‘consume’ a large amount of memory,” especially as next-generation platforms like Nvidia Rubin have an insatiable demand for HBM (High Bandwidth Memory), ruthlessly occupying traditional DRAM capacity.

Meanwhile, Intel is attempting to accelerate its entry into the GPU market through “blood transfusions.” Gelsinger stated that the company has appointed a new chief architect to fully promote GPU development, aiming to seize the opportunities brought by the rapid growth of AI data center demands.

It is reported that Intel has hired former Qualcomm executive Eric Demmers as Chief GPU Architect, aiming to address Intel’s architectural shortcomings in graphics processing and directly compete with Nvidia’s data center business.

In an interview with the media, Gelsinger said that the GPU project is overseen by Kevork Kechichian, head of Intel’s data center chips, which means Intel no longer views GPUs as standalone graphics card components but as a key part of the overall data center solution.

Regarding Intel’s foundry business, Gelsinger revealed that multiple customers are engaging in in-depth discussions with Intel’s wafer manufacturing division, with interest focused on the 14A process technology. Mass production related to this is expected to accelerate later this year.

Gelsinger also stated that customers must specify product quantities and specific product types so that Intel can plan and allocate capacity accordingly.

According to previous media reports, Nvidia plans to collaborate with Intel on the new architecture Feynman chip, with Intel responsible for the advanced packaging needs of the GPU portion. The core GPU chips will still be manufactured by TSMC, while I/O chips will partly use Intel’s 18A process or the scheduled 14A process for mass production in 2028, depending on the yield and production status of 14A.

The I/O chips include memory controllers and are responsible for inter-chip connections. Although their performance requirements are not as high as those of GPU compute chips, they still require advanced process technology.

Additionally, the latest reports indicate that Apple is exploring outsourcing some of its low-end processors to companies outside of TSMC. The reports point out that, given TSMC’s current engagement with Nvidia and other AI companies, Apple is evaluating whether to outsource certain low-end processors to other manufacturers to address supply chain challenges. Although specific candidate manufacturers have not been explicitly named, previous rumors suggest that Intel may begin supplying some low-end processors to Apple around 2027 or 2028.

(Source: The Paper News)