Micron and Samsung executives have delivered a blunt message to the memory market: the shortage that has already raised device prices and constrained PC configurations is not close to over. The two companies said this week that memory supply will remain tight for at least the next couple of years, with Micron CEO Sanjay Mehrotra telling investors that demand for the company's memory will exceed available supply over that period.

The remarks come as artificial intelligence infrastructure absorbs an expanding share of the world's DRAM and high-bandwidth memory output. Micron no longer sells consumer RAM, and Mehrotra's comments refer to business-to-business sales of HBM for AI and DRAM for servers. But the distinction does not isolate consumers from the consequences. Manufacturing capacity is being prioritized for AI and server products, which limits the supply of memory manufactured for consumer devices.

The central claims

• Micron CEO Sanjay Mehrotra expects memory demand to exceed supply for at least the next couple of years, with tightness continuing through 2028.
• Micron no longer sells consumer RAM; its comments refer to HBM for AI and server DRAM, but capacity priorities still affect consumer memory.
• Micron plans new clean rooms in 2028 but says ramps will be gradual and that HBM mix and node transitions will limit supply growth.
• Seventy-five percent of Micron's 2027 memory output is already accounted for, and most current sales discussions are about 2028.
• Samsung EVP Kim Taewoo says HBM will account for almost 30 percent of DRAM wafer capacity in 2027, compared with 20 percent this year.
• Consumers face higher prebuilt PC costs, lower RAM configurations at higher prices, and higher prices for smartphones, streaming sticks, and gaming consoles.

Why the shortage is lasting

Memory has long been a cyclical industry, prone to sharp swings between shortage and oversupply. But the current cycle is being shaped by a different force: AI infrastructure. High-bandwidth memory is a critical component in AI accelerators, and it is more complex to manufacture than standard DRAM. HBM uses stacked DRAM dies and advanced packaging, which consumes more wafer capacity per bit than conventional memory. Server DRAM is also in heavy demand as cloud providers and enterprises expand data center capacity for AI training and inference.

Mehrotra described the market in stark terms. "Overall, supply-demand environment is only getting tighter," he said. That assessment reflects both sides of the equation. Supply growth is constrained by manufacturing complexity, while demand is being pulled forward by larger AI models, growing context windows, more concurrent users, and the spread of agents across enterprise and consumer applications.

New clean rooms will not deliver immediate relief

Micron plans to open new clean rooms for memory manufacturing in 2028, but Mehrotra expects supply to remain limited even after that capacity comes online. "Even after they are built, even after first wafer output, production ramps up only gradually in the clean rooms," he said. "That's just the nature of what it takes to bring up production, and with HBM going from 3E to a greater mix of 4 and 4E, and with the trade ratio that exists, that ... creates headwinds with respect to supply growth. Node transitions of the future give less productivity gain per wafer as well."

That passage matters because it challenges the assumption that building fabs will solve the shortage on the announced timeline. Clean room construction, tool installation, process qualification, and yield learning all take time. Production ramps are gradual, not instantaneous. The shift from HBM3E to a greater mix of HBM4 and HBM4E may improve performance, but it can also reduce effective bit output per wafer because of stacking, packaging, and trade-ratio constraints. Future node transitions are also expected to deliver less productivity gain per wafer, making supply growth harder to achieve than in past cycles.

2027 is already sold out

Seventy-five percent of Micron's memory output for 2027 is already accounted for, and most of the company's current memory sales discussions are about 2028, according to Mehrotra. Demand for HBM is surpassing demand for Micron's DRAM. "For 2027, a large part of the volume is already sold out for 2027 for HBM, and the prices are much higher than 2026 prices," he said.

Micron does not know when supply will catch up with demand because the demand trends are still expanding. Mehrotra pointed to larger models, growing context, more concurrency, greater agents across enterprise, and consumer demand as forces that continue to drive greater need for memory, greater need for memory content, and greater need for higher-performance memory.

This means memory buyers are competing for allocation years ahead. For OEMs, procurement teams, and device makers, planning windows are extending beyond the usual quarterly negotiations. Contracts for 2027 and 2028 are being discussed while 2026 supply remains tight.

Samsung's warning on wafer capacity

Mehrotra's statements are supported by comments from Kim Taewoo, an executive vice president at Samsung. He said HBM will account for almost 30 percent of DRAM manufacturers' wafer capacity in 2027, compared with 20 percent this year. That would limit the amount of DRAM made for consumer devices.

The shift from 20 percent to nearly 30 percent is significant. HBM is not a drop-in replacement for commodity DRAM. It requires known-good dies, stacking, through-silicon vias, advanced packaging, and testing. Every wafer dedicated to HBM is a wafer not making standard DDR5 or LPDDR for PCs, phones, and other consumer products. When HBM's share of wafer capacity rises, the remaining capacity for conventional DRAM must serve servers, PCs, smartphones, and embedded devices. That competition is what turns an AI data center boom into a consumer memory crunch.

Consumer fallout

Consumers have already felt the memory crunch. The shortage has raised costs for prebuilt PCs and pushed OEMs to offer lower RAM configurations at higher prices. Prices for other types of gadgets, including smartphones, streaming sticks, and gaming consoles, are also up.

The effects show up in product design and retail pricing. PC makers can respond by reducing standard memory, using slower modules, cutting upgrade options, or raising prices. Smartphone makers may adjust memory tiers. Console makers and streaming device vendors face bill-of-materials pressure. For buyers, the old assumption that memory capacity per dollar will improve steadily is being challenged. In some segments, the opposite is happening: less memory at higher prices.

Micron's Crucial brand stopped selling RAM to consumers. The change was announced in December 2025. Sumit Sadana, EVP and chief business officer, pointed to AI driving demand for data center memory and Micron's interest in improving "supply and support for our larger, strategic customers in faster-growing segments." The exit underscores a broader reallocation: consumer memory is a smaller priority when data center customers are willing to commit to large volumes and higher prices.

What it means for pricing and product planning

For device makers, memory is no longer just a commodity to buy on spot markets. It is a strategic input. Companies that locked in supply earlier may have an advantage. Those that did not may face allocation, higher prices, or both. The shortage also changes product road maps. OEMs may delay memory-heavy configurations, prioritize premium models, or segment markets more sharply. Enterprise buyers may see server prices rise as HBM and DDR5 demand remains strong. Consumers may see fewer entry-level devices with generous RAM.

Memory makers, meanwhile, face pressure to invest while avoiding overbuilding. The industry's history is full of cycles in which high prices encouraged capacity additions, followed by oversupply and price collapses. This time, executives argue that HBM complexity and node transition economics make supply growth harder. If they are right, the current tightness could persist longer than past upcycles. If demand cools unexpectedly, the same capacity could become a burden. But near term, the public guidance points to continued scarcity.

Why 2028 remains a question mark

Micron's new clean rooms are scheduled for 2028, but that does not mean relief arrives on January 1. Production ramps gradually. Equipment must be installed and qualified. Yields must mature. HBM packaging capacity must keep pace. Customers must validate products. Each step adds lag. Mehrotra's comments suggest that even with new capacity, supply may not catch demand in 2028. The company's sales discussions for 2028 indicate customers are already trying to secure future output.

The demand side is not standing still. AI models are growing larger. Context windows are expanding. More users are accessing models concurrently. Enterprises are deploying agents that can perform tasks across software systems, increasing memory and compute requirements. Consumers are also using more AI features on devices, which can increase memory content. These trends raise the memory needed per system, not just the number of systems sold. That combination supports Mehrotra's view that demand will continue to outpace supply.

The broader market signal

The memory shortage is therefore not a temporary inventory problem. It reflects a shift in capacity allocation toward AI and servers, where memory is a critical performance component. HBM's rise from 20 percent to nearly 30 percent of DRAM wafer capacity in 2027 illustrates the scale of that shift. For consumer electronics, the result is less available DRAM, higher prices, and pressure on product configurations. For data center operators, the result is intense competition for HBM and server DRAM. For memory manufacturers, the result is a market with strong pricing but difficult execution and long lead times for new capacity.

Until supply growth accelerates or demand cools, buyers should expect tight allocation. The executives' comments point to 2028 as the earliest window for meaningful new capacity, but not necessarily the moment when the shortage ends. With 2027 output largely committed and HBM demand exceeding standard DRAM demand, the market is planning further ahead than usual. That planning horizon is now a central fact of the memory business.

Source: Ars Technica News