Intel Warns Global Memory Chip Shortage Could Last Until 2028 Amid AI Boom

In early February 2026, at an AI summit hosted by Cisco, a statement by Intel CEO Lip-Bu Tan poured a dose of cold water on the semiconductor industry amid the AI computing boom. The leader of the semiconductor giant issued a clear industry warning: the global memory chip supply shortage shows no signs of easing in the short term, and this situation may continue at least until 2028.

This assessment is far more pessimistic than previous market expectations. Prior to this, the industry generally believed that memory supply tightness would gradually ease between 2025 and 2026, with some institutions even optimistically predicting a supply-demand balance in the second half of 2026. However, Tan’s warning is not unfounded. It is based on in-depth discussions with two core memory manufacturers and reflects a consensus within the industry. He stated plainly: “As far as I know, there are currently no mitigation measures. The industry generally agrees that there will be no material relief before 2028.”

The root cause of this longer-than-expected memory shortage is not a recovery in traditional demand, but a structural imbalance between supply and demand triggered by the explosive growth of AI infrastructure. With the large-scale deployment of data-intensive applications such as generative AI, large model training and inference, and autonomous driving, the global AI server market has entered a phase of rapid growth. Demand for high-performance memory in such systems is rising at an “exponential” rate, with high-bandwidth memory (HBM) becoming the absolute core.

As a “supporting cornerstone” of AI chips, HBM, enabled by its 3D stacked architecture, delivers ultra-high bandwidth that perfectly matches the compute output of AI accelerators such as GPUs. Taking NVIDIA’s H100 GPU as an example, when paired with HBM3, its memory bandwidth reaches eight times that of traditional DDR5 memory, making it an essential component for training large models with hundreds of billions of parameters. However, the production barrier for HBM is extremely high. It not only involves advanced DRAM process technology and complex 3D stacking and packaging, but more critically, producing the same capacity of HBM requires roughly three times the wafer area of conventional DRAM, resulting in enormous capacity consumption.

Faced with “panic-buying” demand from the AI market, the world’s three major memory giants—Samsung, SK hynix, and Micron—have been forced to strategically tilt their capacity allocation. These three companies currently account for more than 94% of the global DRAM market and form an effective monopoly in the HBM segment. They are prioritizing their limited wafer capacity for HBM production to capture the excess profits of the AI track. Data shows that by 2026, AI servers are expected to consume 53% of global DRAM capacity, while the entire HBM capacity of giants such as SK hynix for 2026 has already been fully sold out, with some customers even locking in capacity for 2027.

This concentration of capacity toward AI directly exacerbates memory supply constraints in traditional end markets. Although demand for conventional DDR memory and NAND flash in PCs, smartphones, and consumer electronics has not seen explosive growth, upstream capacity diversion has significantly tightened supply, driving continuous price increases. According to industry statistics, DRAM prices have risen by more than 60%, while NAND flash prices have climbed by nearly 40%. This wave of price increases—dubbed a “super cycle” by the industry—is now transmitting across the entire electronics value chain.

More critically, the memory industry is both technology- and capital-intensive, with capacity expansion cycles lasting 18–24 months. From upfront capital investment and fab construction to yield ramp-up and mass production, at least two years are required. Although companies such as Samsung and Micron have announced plans to expand memory capacity, most new capacity will be released in 2027 or later. In 2026, the industry will still face a mismatch of “rapid demand growth and delayed supply release,” further validating Tan’s prediction that the shortage will last until 2028.

For the semiconductor and electronic components industry as a whole, this prolonged memory shortage is both a challenge and an opportunity for value chain reshaping. Downstream terminal manufacturers, such as PC and smartphone makers, are facing significant cost pressures. Some entry-level devices have already seen price increases, and manufacturers are being forced to respond by optimizing product configurations and raising average selling prices. This may further suppress the recovery of consumer demand—Counterpoint Research predicts that normalization of the smartphone market may not occur until the second half of 2027 or even early 2028.

For memory manufacturers and upstream players, however, the shortage represents an unprecedented market opportunity. Beyond the continued boom in HBM, supporting segments such as memory interface chips and advanced packaging are also set to benefit from demand tailwinds, allowing related companies to scale rapidly. At the same time, the oligopolistic structure of the memory industry may see subtle shifts, as some second-tier manufacturers with strong technical reserves could seize the window of capacity expansion to capture additional market share.

Tan also noted that NVIDIA’s upcoming Rubin platform and subsequent products will further intensify memory consumption, indicating that AI-driven memory demand will continue to rise. In addition, he pointed out that beyond memory shortages, AI infrastructure faces multiple bottlenecks, including thermal management, interconnects, and advanced packaging, and that the healthy development of the industry requires coordinated efforts across the entire value chain.

Looking back at the history of the memory industry, it has always exhibited cyclical fluctuations between “shortage and oversupply.” However, the AI-driven demand surge is widely regarded as structural and long-lasting, reshaping the industry’s growth logic. For semiconductor practitioners, Intel’s warning serves more like a “wake-up call”: the memory shortage lasting until 2028 is not a short-term market fluctuation, but an inevitable manifestation of industrial transformation in the AI era.

Over the next two years, how to cope with the cost pressures caused by memory shortages and how to seize AI-driven opportunities in the memory sector will become core challenges for terminal manufacturers, chip designers, and memory suppliers alike. For the entire value chain, only by accelerating technological innovation, Optimizing production capacity layout, and strengthening collaboration can the industry achieve win-win outcomes amid this structural transformation and promote the sustained, healthy development of the semiconductor industry.