Micron Technology just told Wall Street that the worst of the memory shortage is still ahead. Speaking on the company’s fiscal fourth-quarter 2026 earnings call on September 30, 2026, CEO Sanjay Mehrotra said memory and storage supply-demand conditions will be “much tighter” in calendar 2027 and 2028 than they were in 2026, a year that already saw DDR5 prices climb more than 100% and squeeze PC builders, console makers and data center operators alike. That same September, Micron said in its September 30, 2026 results release that it had begun sampling the industry’s first 512GB ultra dense DDR5 RDIMM modules, capable of speeds up to 9,200 MT/s and of putting as much as 12TB of memory into a single 24-slot server, a reminder that even as the company races to add high-density capacity, it is not promising that capacity will meaningfully loosen the broader market anytime soon.
More than 75% of Micron’s 2027 production is already committed to customers, according to Mehrotra, and most conversations with buyers have already shifted to securing 2028 supply. The comments, reported October 1, 2026 by Ars Technica, The Register, Computerworld, PCMag and Tom’s Hardware, landed alongside a record quarter that underscored just how profitable the shortage has become for the memory maker.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Mehrotra’s Message to Wall Street
In prepared remarks for the earnings call, Mehrotra told investors: “We expect memory and storage supply-demand conditions to be much tighter in calendar 2027 and 2028 than they were in 2026,” according to Computerworld and PCMag. Pressed by analysts on how long the cycle might run, he went further, saying “In calendar 2027 as well as 2028, we see demand exceeding supply. In fact, we see greater tightness in the industry in 2027 and in 2028 versus 2026. Overall, supply-demand environment is only getting tighter,” as reported by The Register and Tom’s Hardware. Micron has not walked any of that back since; as of October 2026, the company is still telling investors that demand will outstrip supply in both 2027 and 2028, with no softer language in later commentary.
That framing matters because it reverses what some analysts had expected. The Register noted that earlier industry forecasts had pointed to supply and demand returning to balance by 2028. Mehrotra closed that door directly, telling analysts, “We do not have line of sight to when supply and demand will return to balance,” per Computerworld’s transcript of the call.
More Than 75% of 2027 Supply Is Already Gone
The headline figure from the call is the one that will worry IT buyers most: over 75% of Micron’s 2027 output is already locked in under long-term contracts, which Micron calls strategic customer agreements, or SCAs, according to Computerworld. These are multiyear, take-or-pay deals that guarantee volume to large customers, typically cloud providers and AI hardware makers, well before product ships. Micron had 26 strategic customer agreements as of October 1, 2026, estimated to represent more than 35% of revenue through 2030, with some agreements extending to 2031, and it has said both 2027 and 2028 supply-demand conditions will stay tighter than 2026, which is part of why buyers without an SCA already in hand are finding so little open capacity left to negotiate for.
Mehrotra told analysts that “the majority of our discussions with customers today are already around 2028,” and added that “customers want assurance now even beyond 2030 time frame,” according to PCMag’s report on the call. Any customer negotiating a new SCA now is doing so at higher prices, he said: “Any new discussions on SCAs where pricing is involved are negotiated with higher pricing based on prevailing market conditions and outlook,” per Computerworld.
For buyers without an existing long-term agreement, that leaves the spot and open market as the primary source of supply, and TrendForce said in a July report cited by Computerworld that this group is increasingly exposed to the steepest price increases for server DRAM.
HBM Pricing Climbs as AI Chipmakers Lock In Capacity
High-bandwidth memory, the stacked DRAM that sits alongside AI accelerators from Nvidia and AMD, is at the center of the shortage. Mehrotra said Micron’s HBM pricing for 2027 was negotiated with customers last year, and that “now, 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,” according to PCMag. The more than 75% figure applies to Micron’s overall 2027 output, while HBM was described separately as having only a “large part” of its 2027 volume sold out, so the two claims are not identical. As of October 2026, Micron is applying that same “much higher” framing to 2027 customer pricing more broadly, not just HBM, which is why the increases showing up in Micron’s quarterly numbers are not limited to the AI accelerator side of the business. That pattern tracks with the broader shift described in earlier reporting on HBM4 inventory levels falling below ten days of supply as AI data center buildouts accelerated through 2026.
The HBM3E-to-HBM4 Trade Ratio Problem
Part of the reason HBM is squeezing the rest of the memory market is structural. Mehrotra explained that as production shifts from HBM3E to a larger mix of HBM4 and HBM4E, the manufacturing “trade ratio” works against total bit output: “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,” he said, a quote carried by Ars Technica. In practical terms, each HBM die consumes more wafer capacity than a conventional DRAM die, so every additional gigabyte of HBM shipped pulls more fabrication capacity away from standard DRAM and NAND production.
What Micron’s HBM4 and SOCAMM2 Output Looks Like
Micron’s own pages describe where that HBM capacity is going. Micron is in high-volume production of HBM4 36GB 12H, designed for NVIDIA Vera Rubin. The company lists pin speeds over 11 Gb/s, enabling bandwidth greater than 2.8 TB/s, which it describes as a 2.3 times bandwidth increase over HBM3E. It also claims a greater than 20% power efficiency improvement over HBM3E, and both comparisons use the same capacity and stack height (36GB 12H).
The same pages list a 192GB SOCAMM2 module in high-volume production, part of a broad portfolio of SOCAMM2 products spanning 48GB to 256GB capacities. Micron says SOCAMM2 is designed for NVIDIA Vera Rubin NVL72 systems and standalone NVIDIA Vera CPU platforms, enabling up to 2TB of memory and 1.2 TB/s of bandwidth per CPU. For denser stacks, Micron states that its 48GB cube provides a 33% increase in capacity per HBM placement when comparing HBM4 36GB 12H to HBM4 48GB 16H.
On the supply chain side, Micron announced an investment with Indium Corporation in New York critical-minerals processing on October 20, 2026, which it calls the latest investment under its up to $3 billion U.S. supply-chain plan. For readers weighing what the sell-out means, these specifications show why AI platform makers want HBM4 allocation secured early, and why the large share of 2027 HBM volume already sold is a separate claim from the more than 75% figure for Micron’s total 2027 output.
DRAM and NAND Prices by the Numbers
Micron’s own numbers from the September quarter show how quickly prices have already moved. CFO Mark Murphy told analysts that Micron’s “inventory levels and supply remain extremely tight,” and disclosed that DRAM prices rose by a percentage in the high teens in the fiscal fourth quarter alone, while NAND prices climbed about 30%, according to Computerworld’s account of the call.
Independent market researchers are forecasting more of the same. Taiwan-based TrendForce said in a September 30 report, cited by Computerworld, that it expects conventional DRAM contract prices to rise another 10% to 15% in the fourth quarter compared with the third, with NAND flash prices climbing 15% to 20% over the same period. TrendForce noted that the pace of increases is slowing but that the market remains undersupplied.
| Metric | Figure | Source |
|---|---|---|
| Micron DRAM price change, fiscal Q4 2026 | Up in the high teens (%) | Computerworld (Mark Murphy, CFO) |
| Micron NAND price change, fiscal Q4 2026 | Up about 30% | Computerworld (Mark Murphy, CFO) |
| TrendForce DRAM contract price forecast, Q4 2026 vs Q3 | Up 10% to 15% | TrendForce via Computerworld |
| TrendForce NAND flash price forecast, Q4 2026 vs Q3 | Up 15% to 20% | TrendForce via Computerworld |
| IDC average PC selling price forecast, 2026 | Up 17% | IDC via Computerworld |
| Server memory module trend since H1 2026 | Shift from 96GB/128GB to 32GB/64GB modules | TrendForce via Computerworld |
| Micron 2027 output already committed | More than 75% | PCMag, Ars Technica |
Server Buyers Are Already Downsizing Memory
One of the more telling data points from TrendForce’s July report, relayed by Computerworld, is that cloud providers and server OEMs have been quietly moving some server configurations from 96GB and 128GB memory modules down to 32GB and 64GB modules since the first half of 2026. That is a direct response to cost and availability pressure rather than a technical decision, and it echoes a trend already visible on the storage side, where NAND price jumps have pushed buyers toward smaller, pricier-per-gigabyte configurations.
A Record Quarter Fuels the Warning
The supply warning arrived alongside some of the strongest financial results in Micron’s history: the September 30, 2026 release reported fourth-quarter revenue of $54.23 billion, versus $41.46 billion for the prior quarter and $11.32 billion for the same period last year, and full-year revenue of $133.19 billion versus $37.38 billion for the prior year. The Register reported that Micron’s cloud memory business unit, which is centered on HBM, posted a fiscal Q4 gross margin of 83%, up sharply from 59% a year earlier. The core data center business unit, which sells more conventional DRAM, recorded a Q4 gross margin of 90%, up from 41% in the prior-year quarter. Sales of solid-state drives for data centers reached almost $10 billion for the quarter, up more than 1,000% year over year, according to The Register’s reading of Micron’s earnings disclosure.
Looking ahead, Micron’s guidance for its current fiscal first quarter of 2027 calls for record revenue of $61.5 billion and a gross margin of 86.25%, implying gross profit above $53 billion, according to Tom’s Hardware. After accounting for roughly $2.06 billion in operating expenses and a tax rate near 15.5%, Micron guided to earnings per share of $38.15, plus or minus $1.00, on approximately 1.15 billion shares outstanding, per the same report. The Register separately noted Micron’s plan to spend $25 billion in capital expenditure in the first half of its new fiscal year, largely aimed at bringing new factories online in 2028.
| Micron metric | Value | Source |
|---|---|---|
| Cloud memory (HBM) unit gross margin, Q4 FY2026 | 83% (vs. 59% a year ago) | The Register |
| Core data center (DRAM) unit gross margin, Q4 FY2026 | 90% (vs. 41% a year ago) | The Register |
| Data center SSD revenue, Q4 FY2026 | Nearly $10 billion, up over 1,000% YoY | The Register |
| FQ1 2027 revenue guidance | $61.5 billion (record) | Tom’s Hardware |
| FQ1 2027 gross margin guidance | 86.25% | Tom’s Hardware |
| FQ1 2027 implied gross profit | More than $53 billion | Tom’s Hardware |
| FQ1 2027 EPS guidance | $38.15 ± $1.00 | Tom’s Hardware |
| Planned capex, first half of new fiscal year | $25 billion | The Register |
Mehrotra was blunt about why new factory spending won’t bring quick relief. “Production from new DRAM and NAND fabrication facilities takes time to ramp and gradually becomes more meaningful starting a few quarters after initial output,” he said, according to Computerworld. He expanded on the point for PCMag: “Even with any new clean room space coming up in 2028, we see continuing tight supply conditions because, first of all, clean rooms take a long time to build. Even after they’re built, even after the first wafer output, production only ramps up gradually in the clean room; that’s just the nature of what it takes to bring up production.”
Samsung and SK Hynix Ride the Same Wave
Micron is not alone in the memory market, and the structural tightness Mehrotra described applies to the industry as a whole, not just one supplier. Samsung Electronics and SK Hynix compete directly with Micron in DRAM, NAND and HBM, and both have been expanding HBM4 capacity to meet the same AI accelerator demand driving Micron’s order book. Because all three suppliers draw on a similar pool of leading-edge wafer capacity, tightness at one supplier tends to show up as tightness across the industry, which is part of why Micron’s commentary moved sentiment around memory stocks broadly rather than Micron’s shares alone.
SK Hynix Floats a 2030 Timeline
PCMag reported that an executive at SK Hynix has suggested the memory shortage could persist into 2030, a timeline that goes further than Micron’s own 2027-2028 guidance. Neither company has published a specific date for when supply and demand might rebalance, and Mehrotra himself told analysts Micron does not have visibility into when that balance returns, so any 2030 marker should be read as a directional signal from the industry rather than a confirmed forecast.
The Fallout Is Already Hitting Shelves
The price pressure described on Micron’s call is not abstract for buyers. PCMag reported that computer board maker Raspberry Pi announced another price increase the same week, raising the price of the 2GB Raspberry Pi 4 and Raspberry Pi 5 by $12.50 each, to $67.50 and $77.50 respectively, with the company citing the cost of production. It is a small example of a pattern that has played out across PCs, consoles and next-generation game consoles throughout 2026, and Mehrotra’s comments suggest vendors should not expect relief soon.
PCMag also noted that Micron stopped selling memory directly to consumers roughly a year before this earnings call, as the company prioritized higher-margin AI and data center customers over the retail channel. That shift has left PC builders more reliant on distributors and partners such as GPU makers who already had to adjust their own memory sourcing after Micron scaled back certain consumer-facing memory lines earlier in 2026.
Lenovo offered one of the more detailed outside perspectives on the call cycle. Computerworld reported that Lenovo chairman and CEO Yuanqing Yang, speaking on the company’s August 13 earnings call, said he expects memory demand to keep rising and supply to stay constrained at least through the end of 2027. He added that Lenovo can pass through rising component costs: “When the material costs rise, we can adjust the pricing at the front end in a timely manner.” Luca Rossi, president of Lenovo’s Intelligent Devices Group, told Computerworld he expects the PC market to shrink about 15% in units over the six months to March, with business demand holding up better than consumer demand. Ashley Gorakhpurwalla, president of Lenovo’s Infrastructure Solutions Group, said a “strong server refresh cycle is underway” on the enterprise side. Computerworld noted that Dell, HPE, HP, Cisco and Supermicro did not respond to requests for comment on the shortage by publication time.
What Analysts Are Telling CIOs to Do
Neil Shah, vice president of research at Counterpoint Research, told Computerworld that the outlook means IT leaders “will have to be prudent about which equipment to upgrade and which to stretch to maintain cost efficiencies.” His recommendations, as relayed by Computerworld, center on locking in pricing early and getting more out of existing hardware rather than waiting for prices to fall.
“Companies should secure multiyear pricing for the computing capacity they know they will need,” Shah said, warning that shifting workloads to the cloud will not dodge the cost increases “because providers will pass them on.” On which refreshes to delay, he drew a line between back-office equipment and core infrastructure: “For general back-office PCs and routine file servers, stretching lifecycles from three years to five is harmless. But for core infrastructure and engineering seats, delaying refreshes can backfire,” he told Computerworld, citing lost productivity, software support and reliability problems on aging gear.
Shah also cautioned against a common cost-cutting instinct, buying older DDR4 memory instead of current DDR5 or HBM parts. “If you buy legacy platforms today to shave 10% off upfront server costs, you’re buying into systems which won’t have serviceable parts two years from now,” he said, since memory makers have been converting production lines toward DDR5 and HBM rather than legacy DDR4. His final suggestion was to look inward before buying more hardware: “Enterprises often waste 30% to 50% of memory by provisioning for peaks that rarely occur,” he said, pointing to right-sizing virtual machines, quantizing AI models and batching workloads as ways to cut memory use without new purchases.
How This Memory Cycle Differs From Past Shortages
Memory has always been a cyclical business, with suppliers alternating between periods of oversupply and shortage as they time capacity expansions against demand. Earlier shortages, including the one that pushed DRAM prices sharply higher earlier in 2026, were driven primarily by a mix of cyclical underinvestment and surging PC and smartphone demand. What distinguishes the current cycle, based on Mehrotra’s comments, is that AI accelerator demand for HBM is now large enough to pull capacity away from conventional DRAM and NAND production even as total industry bit shipments keep growing. In past cycles, suppliers could usually count on new capacity to restore balance within a year or two of coming online. This time, Micron’s own leadership says it does not have a timeline for when that balance returns, and a competitor has floated a horizon stretching to 2030.
Market Impact: Winners, Losers and Stock Moves
For memory makers, the news is straightforwardly positive: higher prices and locked-in multiyear contracts translate into the kind of margin expansion Micron just reported, with HBM-focused gross margin jumping from 59% to 83% year over year. That dynamic should extend to Samsung and SK Hynix given their similar exposure to AI-driven HBM demand, though neither company’s specific 2027 figures have been confirmed in the reporting around this earnings call.
For buyers of memory, the picture is the inverse. PC makers, server OEMs and console manufacturers are absorbing higher input costs and, per Lenovo’s and IDC’s comments, passing much of that along to customers. IDC’s June forecast of a 17% rise in average PC selling prices for 2026, cited by Computerworld, suggests that pattern was already underway before Micron’s latest guidance, and the new 2027-2028 outlook gives buyers little reason to expect a reprieve. The GPU market has already absorbed a round of DRAM-driven price increases in 2026, and graphics card buyers in particular may see further pressure as GDDR and HBM compete for the same constrained wafer capacity.
Predictions for 2027 and 2028
- Expect further DRAM and NAND contract price increases through the rest of 2026 and into 2027, consistent with TrendForce’s quarter-over-quarter forecasts of 10-15% and 15-20% respectively.
- Server buyers will keep shifting toward smaller memory module configurations, following the TrendForce-documented move from 96GB/128GB modules to 32GB/64GB modules that began in the first half of 2026.
- HBM4 and HBM4E allocation will increasingly determine which AI hardware makers can ship on schedule, since Micron says a large share of its 2027 HBM volume is already sold and priced above 2026 levels.
- PC and console vendors will continue passing costs to consumers rather than absorbing them, following the pattern Lenovo and Raspberry Pi have already set in 2026.
- Enterprises that secure multiyear supply agreements now, as Counterpoint Research’s Neil Shah recommends, will be better insulated than those relying on spot-market purchases once 2027 contract negotiations intensify.
What It Means for Gamers, Builders and Developers
For anyone building or upgrading a PC, Mehrotra’s comments are another signal that cheaper memory is not on the near-term horizon. AI chip demand continues to reshape the broader hardware market, and with Micron, Samsung and SK Hynix all funneling capacity toward HBM for data centers, conventional DRAM and NAND used in consumer graphics cards, laptops and game consoles will keep competing for a shrinking share of total wafer output. Developers running local AI workloads on consumer GPUs face the same squeeze, since both GDDR and system memory draw on overlapping fabrication capacity.
Frequently Asked Questions
What exactly did Micron’s CEO say about memory supply?
Sanjay Mehrotra said on Micron’s fiscal Q4 2026 earnings call, held September 30, 2026, that memory and storage supply-demand conditions will be much tighter in calendar 2027 and 2028 than they were in 2026, and that the company does not have visibility into when supply and demand will return to balance, according to Computerworld and The Register.
How much of Micron’s 2027 memory output is already sold?
More than 75%, according to Mehrotra’s comments reported by PCMag and Ars Technica. Most new customer conversations are now focused on securing 2028 supply instead.
Why is HBM driving the shortage instead of easing it?
HBM requires significantly more wafer capacity per bit than conventional DRAM, and the shift from HBM3E to a larger mix of HBM4 and HBM4E worsens that trade ratio, according to Mehrotra’s remarks carried by Ars Technica. That means rising HBM output for AI accelerators pulls capacity away from standard DRAM and NAND even as total industry bit shipments grow.
How much did DRAM and NAND prices rise in Micron’s latest quarter?
Micron CFO Mark Murphy said DRAM prices rose in the high teens percentage range in fiscal Q4 2026, while NAND prices climbed about 30%, according to Computerworld.
Will new factories fix the shortage?
Not quickly. Micron plans $25 billion in capital expenditure in the first half of its new fiscal year, per The Register, largely for factories due online in 2028. Mehrotra told Computerworld that new fabrication facilities take time to ramp and only become meaningful “a few quarters after initial output,” so new capacity will not immediately ease prices.
Are Samsung and SK Hynix affected the same way?
Both companies compete in the same DRAM, NAND and HBM markets and face similar AI-driven demand pressure. PCMag reported that an SK Hynix executive suggested the shortage could extend as far as 2030, though neither Samsung nor SK Hynix has published company-specific 2027 supply figures in the reporting around this call.
How is this affecting consumer prices already?
IDC forecast in June that average PC selling prices would rise 17% in 2026, according to Computerworld. Raspberry Pi raised prices on its 2GB Raspberry Pi 4 and Raspberry Pi 5 boards by $12.50 each the same week as Micron’s earnings call, per PCMag, citing production costs.
What should IT buyers and businesses do now?
Counterpoint Research’s Neil Shah told Computerworld that companies should lock in multiyear pricing where possible, delay refreshes on back-office equipment rather than core infrastructure, avoid shifting to legacy DDR4 to save money short-term, and optimize existing memory use, since enterprises often waste 30% to 50% of provisioned memory on peaks that rarely occur.
