Micron Technology spent most of 2026 telling investors that 2027 would be tight. On its fiscal fourth-quarter 2026 earnings call on September 30, CEO Sanjay Mehrotra went further, warning that 2028 will be tighter still. According to a transcript cited by CIO, Mehrotra told analysts the company expects memory and storage supply-demand conditions to be much tighter in calendar 2027 and 2028 than they were in 2026. That is a meaningfully different message from “next year is sold out.” It says the shortage has no obvious end date.
The distinction matters for anyone buying a laptop, a graphics card, or a phone in the next two years. Micron’s earlier disclosure that more than 75% of its 2027 output was already committed read like a snapshot of one bad year. Mehrotra’s newer comments, paired with corroborating remarks from Samsung, describe a structural shift in how memory gets allocated, one that pushes commodity DRAM and NAND further down the priority list behind AI accelerators and servers. This piece looks at what changed, who gets squeezed hardest, and how the 2026-2028 cycle stacks up against memory shortages the industry has lived through before.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Sanjay Mehrotra Actually Told Investors
Micron’s prepared remarks for the September 30 call, as reported by Computerworld, stated plainly that industry demand had strengthened since the prior earnings call and that memory and storage supply-demand conditions would be much tighter in fiscal 2027 and 2028 than in 2026. Mehrotra reinforced that framing during the question-and-answer session, saying the company expects demand to exceed supply in both calendar 2027 and 2028, with the industry growing tighter rather than easing. Micron’s fiscal year does not line up exactly with the calendar year, so some coverage uses “fiscal 2027/2028” and some uses “calendar 2027/2028.” Both versions point to the same conclusion: two more years of a seller’s market, not one.
Mehrotra also addressed why Micron cannot simply build its way out of the problem. The company disclosed that it is planning approximately $25 billion in capital expenditure for new factories and added capacity during the first half of its new fiscal year, but he cautioned that new clean rooms planned for 2028 will take time to contribute meaningfully, since production ramps gradually even after the first wafers come off the line, and that spending is not expected to bring prices down quickly once those factories do come online. Micron’s ID1 fab is scheduled for wafer output in mid-2027, and its ID2 fab is scheduled to begin output in late 2028. He pointed to the transition from third-generation HBM to a mix of fourth-generation and enhanced fourth-generation stacks as a factor that works against raw supply growth, because the trade ratios involved mean each wafer yields comparatively less usable output during the switchover. Node transitions further down the roadmap, he added, are delivering smaller productivity gains per wafer than earlier transitions did.
The Numbers Behind the Warning: Bit Growth vs. Bit Demand
Micron is not forecasting a production decline. The company expects industry-wide DRAM bit shipments to grow in the low-20% range annually through 2027 and 2028, with NAND bit shipments growing in the mid-20% range over the same period. HBM bit shipments are expected to grow even faster than conventional DRAM through 2028. On paper, that looks like healthy expansion. The problem, according to Micron’s own framing, is that AI-driven demand is growing faster than even those expansion rates, which is why the company still describes both DRAM and NAND as supply-constrained in 2027 and 2028 despite double-digit output growth.
Micron backed the outlook with results that showed how far the current cycle has already run. The company reported fiscal fourth-quarter 2026 revenue of $54.23 billion, with DRAM alone contributing $39.8B, or 73% of revenue, and guided for fiscal first-quarter 2027 revenue of $61.5 billion, plus or minus $1.5 billion. Micron said DRAM prices rose by a high-teens percentage during the quarter. Sales of data-center solid-state drives climbed sharply year over year during the quarter, and in September 2026 Micron also began sampling the industry’s first 512GB ultra dense DDR5 RDIMM modules, capable of speeds up to 9,200 MT/s, with volume production targeted for the second half of 2027, a sign of how squarely its product roadmap is now aimed at data-center and AI customers. That trend lines up with the broader shift toward AI infrastructure spending that has also lifted chipmakers like Broadcom, whose AI chip revenue has climbed sharply this year as hyperscalers keep building out data centers.
| Metric | 2026 | 2027 outlook | 2028 outlook |
| Industry DRAM bit shipment growth | baseline year | ~low-20% range | ~low-20% range |
| Industry NAND bit shipment growth | baseline year | ~mid-20% range | ~mid-20% range |
| HBM bit shipment growth vs. conventional DRAM | baseline year | faster than DRAM | faster than DRAM |
| Supply-demand balance (Micron’s framing) | tight | demand exceeds supply | demand exceeds supply, tighter than 2027 |
| Share of Micron’s 2027 output already committed | n/a | more than 75% | most current sales talks now center on 2028 |
| Micron fiscal-quarter revenue | $54.2B (Q4 FY2026, reported) | $61.5B ± $1.5B (Q1 FY2027, guided) | not yet guided |
HBM Is the Real Bottleneck, Not Commodity DRAM
High-bandwidth memory sits at the center of Micron’s warning. Mehrotra said a large part of Micron’s 2027 HBM volume was already sold and that the negotiated prices for that volume are much higher than 2026 prices, a detail reported by Investing.com’s transcript of the call. HBM goes into the memory modules stacked next to AI accelerators from Nvidia, AMD, and a growing list of custom silicon providers, where bandwidth matters more than raw capacity per dollar. Every HBM wafer that goes toward an AI accelerator is a wafer that is not making commodity DRAM for a laptop or a phone, which is the mechanical link between the AI boom and rising prices on ordinary consumer electronics.
That dynamic is not unique to Micron. GPU makers have already felt memory supply constraints ripple into their own roadmaps. Micron’s decision to discontinue 2GB GDDR7 modules squeezed supply for Nvidia’s RTX 50 series earlier this year, and Nvidia’s RTX 5080 Super has already seen its launch pushed back partly over GDDR7 cost pressure. If HBM keeps absorbing a growing share of wafer capacity through 2028, graphics card makers and the companies building AI accelerators both end up competing for the same shrinking pool of leading-edge memory capacity.
Where Micron’s HBM4 Output and New Fabs Stand
Micron’s fiscal fourth-quarter 2026 results, reported on Sept. 30, 2026, state that the company is in high-volume production of HBM4 36GB 12H. Micron lists pin speeds over 11 Gb/s, bandwidth greater than 2.8 TB/s, a 2.3 times bandwidth increase over HBM3E and a greater than 20% power efficiency improvement over HBM3E. On October 6, 2026, Micron was described as working with NVIDIA on what was called the industry’s first custom HBM4E implementation, named NV-HBM, for next-generation GPUs and NVLink Fusion platforms. SK hynix is moving the same way, having demonstrated HBM4 in a Vera Rubin system at OCP Global Summit 2026 on October 16, 2026.
The capacity picture is also moving, though slowly. On October 20, 2026, Micron said its expected U.S. fab and technology investment through 2035 exceeds $250 billion, and announced an investment with Indium Corporation to expand critical-minerals processing and high-purity-materials manufacturing in Central New York. Its ID1 fab is scheduled for wafer output in mid-2027 and its ID2 fab is scheduled to begin output in late 2028. Micron has said it has no line of sight to when supply and demand will return to balance, despite planned industry cleanroom expansion.
For buyers, the useful reading is that factory milestones mark when output begins, not when prices fall. Because HBM production uses manufacturing capacity that could otherwise support conventional DRAM, every step up in HBM4 output keeps pressure on server and general-purpose memory supply. Anyone budgeting for servers, workstations or GPU-heavy builds should plan around that constraint rather than around the opening dates of new fabs, and should expect the HBM transition to compete with conventional DRAM for capacity throughout the period these fabs are being built.
Samsung Backs Up Micron’s Timeline
Micron is not alone in pointing to the shift toward HBM. SK hynix demonstrated HBM4 in a Vera Rubin system at OCP Global Summit 2026, and Samsung publicly identified both HBM4 supply and HBM4E at its own OCP event. Because HBM uses manufacturing capacity that could otherwise support conventional DRAM, that shift keeps adding pressure on conventional DRAM supply, and it suggests the move toward HBM is accelerating, not plateauing.
Ars Technica’s reporting also noted that Micron and Samsung executives both expect the shortage to continue for at least the next couple of years, giving the market two independent, named sources pointing at the same 2027-2028 window rather than one company’s isolated forecast. For an industry that has a long history of suppliers downplaying each other’s gloomier predictions, having Samsung’s numbers line up with Micron’s language is a meaningfully stronger signal than either company’s comments alone.
The End of Consumer RAM at Micron’s Crucial Brand
One of the clearest signs of where Micron’s priorities now sit is what the company stopped doing. Its Crucial brand, long a familiar name on retail RAM kits and SSDs, exited consumer RAM sales entirely. Micron announced the change in December 2025, with chief business officer Sumit Sadana pointing to AI-driven demand for data-center memory and the company’s desire to improve supply and support for its larger, faster-growing strategic customers, according to Ars Technica’s reporting. In plain terms, Micron decided retail RAM buyers were a worse use of constrained wafer capacity than hyperscalers and AI accelerator vendors.
That decision already shows up in retail pricing. Shortages have pushed up the cost of prebuilt PCs, encouraged OEMs to ship smaller RAM configurations at higher prices, and nudged up prices on smartphones, streaming sticks, and game consoles, Ars Technica reported. NAND pricing has moved in a similar direction, with Samsung’s P9 SSD line arriving alongside a sharp jump in NAND contract prices this year, and Micron itself reported that its NAND prices increased about 30% sequentially in fiscal fourth quarter 2026, underscoring that the squeeze is not confined to DRAM.
How This Differs From the “75% Sold Out” Headline
It is worth separating two numbers that have circulated in coverage of Micron’s call. The 75% figure refers to how much of Micron’s 2027 memory output is already committed, a snapshot of booking status that this site covered in detail when Micron’s 2027 memory supply was reported as three-quarters sold out. By October 2026, Micron said most of its customer supply conversations had already moved past 2027 entirely and were focused on securing 2028 output instead, which shows just how far forward buyers are now being forced to plan. The newer, broader statement from Mehrotra is a forecast about the entire 2027-2028 window, independent of any single booking percentage: demand will exceed supply in both years, and 2028 will be tighter than 2027. One is a measurement of what has already been sold. The other is a prediction about where the whole market is headed, regardless of how contracts get booked along the way.
That distinction matters because a sold-out percentage can theoretically improve if new capacity comes online faster than expected. A multi-year demand-exceeds-supply forecast is a different, more structural claim, and it is the one that should worry buyers planning purchases two or three years out rather than next quarter.
Market Impact: PC and Laptop Makers
PC makers are caught in the middle of this cycle. Component costs for DRAM and NAND feed directly into bill-of-materials calculations, and when memory suppliers prioritize data-center and AI customers, PC OEMs compete for whatever capacity is left. The practical result, already visible in 2026, is laptops and desktops shipping with smaller default memory configurations while sticker prices hold steady or rise. Buyers who want 32GB or 64GB configurations are increasingly paying a premium that did not exist a year or two ago, and some vendors have quietly shifted baseline configurations down a tier to protect margins without raising headline prices too visibly.
If Micron and Samsung are both right about 2028 being tighter than 2027, PC makers have little reason to expect relief soon. Procurement teams at major OEMs are reportedly locking in multi-quarter supply agreements earlier than usual, a defensive move that mirrors what happened during past memory crunches, but one that also tends to lock in higher average costs for longer.
Market Impact: GPU and AI Accelerator Supply Chains
Graphics card and AI accelerator makers sit closer to the center of the squeeze than PC OEMs do, because HBM is both the fastest-growing and the most constrained category Micron described. Nvidia, AMD, and the expanding field of custom AI silicon vendors all compete for HBM allocation, and the broader GPU market beyond Nvidia and AMD remains a small fraction of total shipments, meaning there is little slack capacity elsewhere in the ecosystem to absorb a supply shortfall. When HBM allocation tightens, it tends to show up first in AI accelerator lead times and later in consumer GPU pricing, since memory makers have clear financial incentive to fill AI orders before gaming-card orders.
The GDDR7 situation already previewed what this looks like in practice. When Micron pulled back from making certain GDDR7 densities, Nvidia’s own roadmap absorbed the hit, and card makers had to adjust specifications and pricing in response. A multi-year HBM squeeze raises the odds of similar disruptions recurring on both the AI-accelerator and gaming-GPU sides of the business through 2028.
Market Impact: Phones, Consoles, and Everyday Gadgets
Consumer electronics outside the PC and GPU categories are feeling the squeeze too. Smartphone makers draw on the same DRAM and NAND pools as PC OEMs, and higher component costs typically show up as smaller base-storage tiers, slower year-over-year RAM upgrades, or quietly higher prices on mid-range models where margins are thinner. Game consoles and streaming devices, which rely on comparatively small amounts of memory per unit but ship in enormous volumes, are similarly exposed, since manufacturers negotiating supply contracts now are doing so against a backdrop where Micron itself says the environment keeps getting tighter rather than easing.
None of this means consumer gadgets stop shipping. It means the easy, several-years-long trend of more memory for the same price is pausing, and in some product categories reversing, as manufacturers absorb higher input costs or pass them along.
Historical Context: How This Compares to Past Memory Cycles
Memory markets have run through boom-and-bust cycles for decades, and the 2017-2018 DRAM shortage is the most commonly cited reference point. That cycle was driven by strong demand across smartphones, servers, and PCs colliding with limited supplier capacity, and it pushed DRAM prices sharply higher before oversupply eventually brought prices back down. The 2021 cycle told a different story: pandemic-driven electronics demand and supply-chain disruption drove shortages and price spikes, followed by a sharp downturn once inventories built back up and demand cooled.
The current cycle has a structural feature those two did not share at the same scale: sustained AI infrastructure buildout. Past shortages were largely demand spikes layered on top of ordinary capacity cycles, and they resolved once suppliers caught up or demand normalized. Micron’s own language, demand exceeding supply through 2028 with conditions getting tighter rather than looser, describes something closer to a durable reallocation of capacity toward AI and servers than a temporary spike. Whether that holds for a third straight year, let alone beyond 2028, is the open question the whole industry is now pricing into contracts years in advance.
Competitive Landscape: Micron vs. Samsung vs. SK Hynix
Micron, Samsung, and SK Hynix remain the three suppliers whose decisions move this market. Micron’s public commentary has been the most detailed of the three on 2027-2028 specifics, covering bit-shipment growth rates, committed output percentages, and HBM pricing. Samsung’s contribution, through Kim Taewoo’s comments on HBM’s rising share of wafer capacity, independently supports Micron’s framing without directly echoing Micron’s numbers, which is exactly the kind of corroboration that should carry weight with buyers trying to decide whether to believe a single company’s forecast.
SK Hynix has not been quoted with comparable specificity in the coverage reviewed for this piece, and attributing a matching forecast to the company without a direct, sourced statement would be speculation. What can be said is that all three suppliers operate in the same constrained wafer environment Micron described, and that Micron no longer sells memory directly to consumers the way it once did through Crucial, leaving Samsung and SK Hynix as the more direct consumer-facing suppliers most affected by any further prioritization of AI customers across the industry.
What Industry Voices Are Saying
Micron’s own executives have been the most direct public voices on this story, and several of their statements are worth quoting in full. Sanjay Mehrotra, Micron’s CEO, said on the fiscal Q4 2026 earnings call: “We expect memory and storage supply-demand conditions to be much tighter in calendar 2027 and 2028 than they were in 2026,” according to a transcript covered by Computerworld.
On the NAND side, a Micron executive on the same call said: “For calendar 2027 and 2028, we expect industry NAND bit shipments to grow approximately in the mid-20s percentage range and the industry to remain supply constrained in both years,” per the Investing.com transcript. The same transcript recorded the company’s HBM expectation: “We expect industry HBM bit shipments to grow faster than conventional DRAM through calendar 2028.”
Mehrotra put the demand side even more bluntly on an earlier call, saying: “If we look at this multiyear time horizon, even going beyond 2027, into 2028, et cetera, we are able to get very high confidence demand from our customers that is far in excess of our ability to support using our supply,” according to a Yahoo Finance earnings-call transcript. On HBM specifically, he said: “Absolutely, our demand for HBM, not just in 2027, but even 2028, is well above our ability to supply across all the different HBM flavors,” per a separate Yahoo Finance transcript of the call.
Taken together, these statements describe a company that is not hedging its language. Micron’s executives used the same tighter-not-looser framing across at least two separate earnings calls, which suggests the message is a deliberate, consistent signal to customers and investors rather than an offhand remark that got amplified by headlines.
Timeline: How the Story Developed
| Date | Development | Reported by |
| December 2025 | Micron’s Crucial brand exits consumer RAM sales, citing AI-driven data-center demand | Ars Technica |
| September 30, 2026 | Micron’s fiscal Q4 2026 earnings call; Mehrotra says 2027-2028 will be tighter than 2026 | CIO, Computerworld |
| September 30, 2026 | Micron reports $54.2B in Q4 FY2026 revenue; guides $61.5B ± $1.5B for Q1 FY2027 | CIO, Computerworld |
| September 30, 2026 | Micron discloses 2027 HBM volume largely sold out at prices well above 2026 levels | Investing.com |
| Week of September 29, 2026 | Samsung EVP Kim Taewoo says HBM will reach ~30% of DRAM makers’ wafer capacity in 2027, up from ~20% this year | Reuters, via Ars Technica |
| October 1, 2026 | Analysis coverage of the multi-year outlook and consumer impact published | Ars Technica, Computerworld |
Predictions: What to Watch Through 2028
- Expect Samsung and SK Hynix to issue similar multi-year guidance on their next earnings calls, now that Samsung’s wafer-capacity comments have already moved in the same direction as Micron’s forecast.
- PC and laptop OEMs will likely keep trimming default RAM and storage configurations through 2027 rather than raise headline prices outright, since quiet downgrades are easier to absorb than visible price hikes.
- HBM4 and HBM4E transition costs will probably keep AI accelerator memory the tightest single link in the chain, ahead of commodity DRAM and NAND, through at least 2028.
- Consumer GPU pricing and specifications will likely keep reacting to memory supply news, following the pattern already set by GDDR7 availability affecting Nvidia’s RTX 50 series roadmap.
- Smaller or regional memory module makers may try to fill part of the gap left by Micron’s exit from consumer RAM, but without comparable leading-edge wafer capacity, they are unlikely to meaningfully ease pricing pressure across the broader market.
These are forward-looking assessments based on the trajectory described by Micron and Samsung executives, not guaranteed outcomes, and actual market conditions could shift if new capacity comes online faster than currently planned or if AI infrastructure spending slows.
What Buyers and IT Teams Should Do Now
For anyone planning hardware purchases on a multi-quarter or multi-year horizon, the practical takeaway from Mehrotra’s comments is to stop assuming memory prices will normalize soon. Enterprise IT teams refreshing fleets of laptops or servers may want to lock in configurations and pricing earlier than usual, following the same instinct that is reportedly driving OEM procurement teams to sign longer agreements now. Consumers shopping for a new PC, phone, or console should expect base memory and storage configurations to keep trending smaller relative to price, a pattern that is unlikely to reverse while Micron and Samsung are both describing the next two years as tighter, not looser, than the year just finished.
Frequently Asked Questions
Why does Micron expect memory to be tighter in 2027 and 2028 than in 2026?
CEO Sanjay Mehrotra told investors on the fiscal Q4 2026 earnings call that AI-driven demand for memory and storage is growing faster than the industry can add supply, even though Micron expects double-digit bit-shipment growth for DRAM and NAND through 2028 and disclosed approximately $25 billion in planned capital expenditure during the first half of its new fiscal year on new factories and capacity that still won’t ease prices quickly once those factories open.
Is this the same as the “75% sold out” story about Micron’s 2027 memory?
No. The 75% figure describes how much of Micron’s 2027 output is already committed under contract. The newer comments are a broader forecast that demand will exceed supply in both 2027 and 2028, independent of that specific booking percentage.
Does Micron still sell RAM to consumers?
No. Micron’s Crucial brand exited consumer RAM sales, a change the company announced in December 2025, citing AI-driven data-center demand and a desire to prioritize larger, strategic customers.
Will graphics card prices keep rising because of this?
Memory supply tightness has already contributed to GDDR7 availability issues affecting Nvidia’s RTX 50 series, and continued HBM demand growth through 2028 raises the risk of similar pressure on consumer GPU pricing and specifications.
Are Samsung and SK Hynix facing the same supply constraints as Micron?
Samsung executive Kim Taewoo has said HBM will account for almost 30% of DRAM makers’ wafer capacity in 2027, up from about 20% this year, which independently supports Micron’s view that the industry as a whole is shifting capacity toward HBM. SK Hynix has not been quoted with comparable specifics in the coverage reviewed for this article.
How does this compare to the 2017-2018 and 2021 memory shortages?
Those earlier cycles were largely driven by temporary demand spikes against ordinary capacity cycles and eventually resolved once supply caught up or demand cooled. The current cycle is tied to sustained AI infrastructure buildout, which Micron’s own language suggests is a more durable shift in how capacity gets allocated rather than a short-term spike.
Will phone and laptop prices go up because of this?
Reporting already links the memory squeeze to higher prebuilt PC costs, smaller default RAM configurations at similar price points, and rising prices on smartphones, streaming devices, and game consoles.
When might memory supply loosen up again?
Micron has not given a timeline for supply and demand returning to balance. Mehrotra said the company lacks visibility into when that will happen, partly because new clean-room capacity planned for 2028 will take time to ramp even after construction finishes.
