Memory Shortage to Tighten Through 2028 as Micron Posts Record Quarter

Micron's fiscal fourth quarter revenue rose nearly fivefold from a year earlier to $54.23bn, the company said. Net income reached $37.7bn against $3.2bn a year earlier. Gross margin hit 86.8%, up from 44.7%. Analysts polled by LSEG had expected $51.07bn for the quarter, according to CNBC.

"We do not have line of sight to when supply and demand will return to balance," chief executive Sanjay Mehrotra told investors on the earnings call.

That is the news peg, and it is a demand story dressed as a supply story. Mehrotra said demand will exceed supply in both 2027 and 2028, and that the market will be tighter in those years than in 2026. About 75% of Micron's memory output for 2027 is already spoken for, and most current sales talks are about 2028. Much of next year's high-bandwidth memory (HBM) is sold out at prices "much higher" than this year's, he said. HBM is the stacked memory that sits inside AI accelerators, and its demand is growing faster than ordinary server memory.

Micron said it is working with Nvidia on what it called the industry's first custom HBM4E product, plans new factory clean rooms in 2028, and has set about $25bn of capital spending for the first half of its new financial year. Output from those clean rooms ramps up only gradually, Mehrotra said, so they will not quickly ease supply.

The product mix tells you where the money is going. DRAM revenue rose 343% year on year to a record $39.8bn, or 73% of the total, while NAND flash revenue rose 526% to $14.1bn. The core data centre unit made $18bn in revenue, up from $1.58bn a year earlier, at a 90% gross margin. The cloud memory unit, which sells HBM, made $16.28bn at an 83% margin.

For the current quarter, Micron forecast revenue of $61.5bn, give or take $1.5bn, against analyst expectations of $57bn, again per LSEG. Tom's Hardware reported the guidance alongside a record gross margin of 86.25% and noted that Micron expects DRAM and NAND shipments to grow in the low-to-mid 20s percentage range for the next two years, still not enough to match demand.

"In calendar 2027 as well as 2028, we see demand exceeding supply," Mehrotra said on the call, according to Tom's Hardware. "Overall, supply-demand environment is only getting tighter."

There is a labour angle buried in this, and it is not the one usually told about chip shortages. Mehrotra said Micron raised pay for every employee in fiscal 2026, a detail CNBC picked up in the context of hundreds of staff in Taiwan who had threatened to strike over pay. The memory buildout is a capital and headcount race at the same time: clean rooms need process engineers, equipment technicians and yield specialists, and those people are scarce in every geography that is now trying to build fabs.

The consumer side is already paying. Raspberry Pi has raised the price of its 2GB Pi 4 and Pi 5 by $12.50, citing memory costs; the 2GB Pi 4, once a $35 computer, now costs $67.50, according to The Next Web. Samsung has added $100 to most of its Galaxy S26 range, and Samsung executive Kim Taewoo said this week that HBM will take almost 30% of the industry's DRAM wafer capacity in 2027, up from 20% this year, as Reuters reported. That shift is the mechanism by which an AI data centre boom becomes a shortage of ordinary memory for phones, laptops and single-board computers.

While Micron was reporting, it was also suing. The Boise, Idaho company filed a lawsuit this month against China's YMTC, accusing it of poaching its leading engineers, obtaining NAND know-how from them and then using that information to build its own non-volatile memory and patent portfolio, Tom's Hardware reported. Micron further alleges that YMTC uses patents granted to its former employees to accuse Micron of patent infringement in various jurisdictions.

According to that report, Micron claims the former employees cited a broad range of personal or family reasons for resigning and did not update their LinkedIn profiles to reflect their new roles at YMTC. The employees had confidentiality obligations that continued after termination, Micron says, and the company argues it "had no reason to suspect YMTC itself" of planning to use its confidential information because YMTC did not have any material market share in the industry. The claims are untested in court and YMTC's response is not in the dossier.

Engineers, in other words, are the asset in the middle of the fight, not a background input. The same week produced two other reminders that engineering labour is being reorganised around AI rather than simply replaced by it. InfoQ reported on 1 October that Chelsea Troy, in a Craft conference talk, argued that skill development requires slowing down: activities that promote learning are "by design, inefficient", and tools used as outcome accelerants often do not support it. She suggested spreading learning across several bouts of quality sleep and using outcome prediction, asking "what do I think the next step should be and why" before checking the answer.

InfoQ's 2 October preview of QCon San Francisco, which runs 16 to 20 November, frames the same problem from the operations side. Airbnb's Weiping Peng will describe how the company guardrailed an AI customer support agent that serves millions of customers and can initiate account actions, covering input sanitization, classifiers, shadow testing, false-positive management and rapid-response mitigations. OpenAI's Brian Yang will discuss building OpenAI Ads with coding agents to more than $100m in annual recurring revenue in under six weeks, and where human judgement stayed.

Security researchers are trying to give those boundaries a formal shape. A paper submitted to arXiv on 30 September, by Corinn Tiffany, Wen Zhang, Eugene Bagdasarian and Lillian Tsai, presents Sapien, a policy engine that enforces stateful contextual policies on agent tool calls. The authors report that even if an agent is fully hijacked, Sapien's policies rule out 93 to 95% of attacks on AgentDojo and 62 to 85% on Toolathlon, which they say is twice as many as tool allowlists on long-horizon tasks, while staying within a few percent of an unconstrained agent's utility.

Hiring data from the commercial side points the same way. TechCrunch reported on 2 October that Clay, which coined the GTM engineer role in 2023, says roughly 100 GTM engineering job listings now appear each month, including roles at Cursor, Lovable and Webflow, and that Clay co-founder and CEO Kareem Amin will speak on the AI Stage at TechCrunch Disrupt 2026. TechCrunch also reported in February that Clay had tripled annual recurring revenue to $100m in a year.

Not every engineering shortage is a software one. The Robot Report reported on 30 September that Autonomous Solutions Inc. and SoftBank announced a joint venture to commercialise brand-agnostic autonomous heavy equipment, with SoftBank providing $225m in new funding to ASI. "Skilled equipment operators are hard to find, and coordinating mixed fleets across a big jobsite is a real bottleneck," ASI CEO Mel Torrie told the publication.

Also on 30 September, the same outlet reported that Arrowfly, The Robot Report's parent, launched AI for Engineers, a platform with a news desk, an annual conference set for 20 to 22 September 2027 in Henderson, Nevada, research and an advisory board, citing more than 30,000 engineering and technical professionals already engaging with AI topics across its portfolio. And on 1 October, Tech.eu reported that Latvian startup Trace.Space launched Trinity, a hardware engineering platform connecting requirements, testing, design parameters and product variants for robotics, aerospace, automotive and defence teams, with AI agents given access to that engineering data.

Space is the other frontier where engineering labour and launch capacity collide. NASASpaceFlight reported on 1 October that Canada Rocket Company will build the country's first large-scale rocket engine test facility at London International Airport, a $30m complex leasing 50 acres, with up to three vertical and horizontal stands able to static-fire engines producing more than 1 MN of thrust. The Jeremy Hansen Test Facility is expected to be fully operational by 2028 and to create about 40 skilled jobs in the London area within 18 months, covering engineering, fabrication and test preparation.

"Canada's path to its own access to space starts on the ground," CEO and co-founder Hugh Kolias said. "This facility is where our engineering becomes hardware, and where we put our engines to the test."

Rocket capacity is itself constrained, which matters for anyone planning a constellation. Light Reading reported on 30 September that SpaceX's Starship launch and the deployment of the first batch of next-generation Starlink satellites renewed concerns about launch shortages for rivals, after the Wall Street Journal reported that some companies could not get Falcon 9 flights after 2028. Analysys Mason's Dallas Kasaboski told the publication that more than 230 constellation projects with 1.5 million satellites are planned, most for orbital data centres, and that stripping those out leaves 133,000 satellites planned; at 2025's record rate of 5,000 satellites a year, launching them would take almost 27 years.

The through line across all of it is that capital is abundant and trained people are not. Micron can commit $25bn to clean rooms and still warn that supply will not catch demand before 2028, because fabs and the engineers who run them take years to come online. Canada Rocket Company can lease 50 acres and build stands, but it needs 40 skilled people in 18 months. Clay can say 100 GTM engineering roles are posted a month, and Airbnb can describe guardrails for an agent that touches customer accounts, and in both cases the constraint is judgement rather than tooling.

Micron's own guidance is the sharpest number in the pile: $61.5bn in revenue expected next quarter, a figure that assumes the shortage holds. If it does, the price of memory stays high, consumer devices keep absorbing the increase, and the engineering teams building around AI inherit a supply chain that will not loosen on its own schedule.