Micron Q4 Revenue Nearly Quadruples as AI Memory Demand Outstrips Supply

Micron reported better-than-expected quarterly results on Wednesday, with revenue almost quadrupling from $11.32 billion a year earlier, according to the company's statement as carried by CNBC. The Boise, Idaho-based chipmaker is the only U.S.-based maker of high-bandwidth memory, or HBM, the stacked DRAM that AI accelerators from Nvidia and AMD need in ever-larger quantities.

Net income in the quarter climbed to $37.7 billion, or $32.87 per share, from $3.2 billion, or $2.83 per share, a year ago. DRAM revenue rose 343% year over year to $39.8 billion and accounted for 73% of total sales.

The guidance was the bigger surprise. For the fiscal first quarter, Micron said it expects revenue of about $61.5 billion and adjusted earnings per share of $38.15. Analysts polled by LSEG had expected $35.40 in adjusted earnings per share on $57 billion in revenue, CNBC reported.

"At this point, I have not heard any negative data points pointing to the memory cycle reversing toward a decline anytime soon for the foreseeable future," Hendi Susanto, a portfolio manager at Gabelli Funds, said in an email after the release, according to CNBC.

Micron's stock has soared more than 500% in the past year and its market capitalisation has topped $1.2 trillion, even though it holds the smallest HBM market share of the three main suppliers, behind South Korea's SK Hynix and Samsung. The share move reflects a supply crunch: HBM leaders cannot make enough of the chips, and the shortage has spilled into general DRAM, pushing up prices for consumer devices including Apple's iPads and MacBooks.

CEO Sanjay Mehrotra said on the earnings call that Micron has a "strong roadmap for future HBM products" and is working with Nvidia on what he called the industry's "first custom HBM implementation." Micron is investing $250 billion to build two new campuses for HBM manufacturing. The largest broke ground in Clay, New York, in January, and its first new fab in Boise is scheduled to come online next year.

Capacity is the constraint, not orders

The earnings season that Micron opened is being read across Asia as a signal on how long the memory boom runs. Korea JoongAng Daily reported on Thursday that Micron's record results point to upside for Samsung and SK Hynix, while The Korea Herald said the same day that the memory shortage is deepening and extending the boom for Korean chipmakers. Japanese chipmakers rallied on Micron's numbers, Investing.com reported on Thursday.

Other outlets put the quarter's headline number slightly differently. qz.com reported $54.2 billion in revenue on AI memory demand, and Seoul Economic Daily also carried the $54.2 billion figure. The company's own statement gives the year-ago comparison of $11.32 billion; some coverage described the jump as 379%, while CNBC's phrasing was that revenue "almost quadrupled." Both describe the same direction, and the company has not disputed either.

TradingView reported on Thursday that Micron's gross margin hit 86.8% on AI-driven demand. That figure, alongside the capex forecast, is part of why the stock reaction was muted: TradingView also reported that Micron shares dropped after hours as a heightened capex forecast offset the earnings beat and the above-consensus guidance. CNBC noted the stock rose only slightly in extended trading.

Analysts largely stayed constructive. Yahoo Finance Canada reported on Thursday that analysts remain bullish after another beat and raise, with the memory demand outlook extending into 2027. The Business Times reported that Micron's AI-fuelled revenue forecast blew past estimates and that its backlog swelled.

Mehrotra's week was not only about earnings. He participated in a summit on AI regulation hosted by President Donald Trump on Tuesday, days after attending a White House dinner with Chinese President Xi Jinping on his first visit to the U.S. in more than a decade, CNBC reported.

On the call, Mehrotra said Micron increased compensation for every employee in fiscal 2026. That detail lands against a backdrop of labour tension: CNBC reported that hundreds of Micron staffers threatened to strike at its factories in Taiwan over labour negotiations and pay, and that similar strikes at SK Hynix and Samsung resulted in bonuses upwards of $500,000.

Elsewhere in the supply chain

Not every chip story this week was about memory. EE Times reported on Wednesday that Emergence AI is moving its neuroformal AI technology into active deployments with fabless semiconductor companies, while integrated device manufacturers among its customers are asking it to extend the work into the fab. The startup targets manufacturing data to flag early signs of degradation before a problem affects a large number of wafers.

Satya Nitta, co-founder and executive chairman of Emergence AI, told EE Times at SEMICON India 2026: "You cannot make chips any faster, but what you can do is definitely get more chips per wafer yielding with AI." Nitta said demand for chips exceeds supply and every fab is running at capacity, so the way to ease the shortage is to raise yield per wafer rather than push more wafers through.

Nitta became executive chairman and chief scientist about a month ago, when Emergence AI appointed Ian Eslick as CEO, according to a LinkedIn post by Nitta cited by EE Times. Eslick founded Silicon Spice, an MIT spinout that Broadcom acquired for $1.2 billion, and later led technology initiatives at U.S. Bank and SoFi, according to the same post.

Nitta described the approach as pairing large language models with symbolic AI: "Algorithms propose, and symbolic AI verifies," he said. He gave the example of an agent finding the same pattern across 30% of 1,500 products, with failures in a ring oscillator block, and suggesting a redesign of that block.

On the manufacturing side of the data centre buildout, GlobalFoundries sees strong demand for Chinese optical modules. Shankaran Janardhanan, senior vice-president of the firm's photonics and radio frequency division, told the South China Morning Post in an interview published on 27 September that demand in the optical networking module segment was "strong," and that it had partly driven the company's decision to bet heavily on photonics. He spoke at the company's annual tech summit in Shanghai.

Two other data points frame the demand picture. On the battery side, Transport & Environment published analysis on 30 September finding that by 2030 the EU will have sufficient Made-in-EU cells for the scope of the Industrial Accelerator Act, provided all projects materialise, including medium-confidence ones. The group said local content requirements for cells and cathode active materials could mean an additional 34% in EU battery demand compared with the current baseline in 2027, and that China controls up to 90% of global pCAM and CAM production capacity depending on the chemistry, and 95% for LFP.

And in machine learning research, a paper submitted to arXiv on 28 September measured how 17 open-weight language models redistribute attention-head activity as tasks demand more serial steps. The authors, Johnny Jingze Li, Abdulla Kuleib, Kalyan Basu and Gabriel A. Silva, found that in most models layers just before mid-depth concentrate activity while later layers spread it, and that models differ in where and how strongly this happens. They reported that Qwen2.5 base models from 0.5B to 7B spread less than the average model and concentrate activity in parts of their second half, where Llama models from 1B to 8B and OLMo-2 spread.

None of that changes the near-term read from Micron. The company's guidance implies the memory shortage it is selling into persists at least through the current quarter, and its $250 billion buildout is aimed at a supply gap its rivals are also racing to fill.