Micron’s AI Memory Boom Demands $25 Billion Capex as HBM Sells Out and Technical Signals Flash Caution

Micron's pre-sold HBM capacity through 2026 and surging revenue highlight a disconnect as insider selling and technical overextension weigh on the stock.

A sold-out product pipeline stretching two years into the future sounds like an investor’s dream. Yet Micron Technology’s stock ended last week with an 8.11% thud, closing in Frankfurt at €624 — roughly 9% below the 52-week peak hit on May 13. The contradiction encapsulates the tension between exceptional fundamentals and a market that may have already priced them in.

The core of the rally remains high-bandwidth memory. Micron has pre-sold its HBM capacity for current and next-generation stacks through the end of 2026. New customers face waiting lists of roughly 18 months. The chipmaker is now producing in volume a 36-gigabyte, 12-high HBM stack co-developed for Nvidia’s upcoming Vera-Rubin platform. Compared with HBM3E, bandwidth jumps 2.3-fold while power consumption drops more than 20% — a tangible edge for data-center operators where energy costs directly hit the bottom line.

The operational numbers underscore the boom’s magnitude. In the second fiscal quarter of 2026, Micron posted revenue of $23.86 billion, a 196% surge from a year earlier. Earnings per share hit $12.20, and gross margin expanded to 74.9%. For the current third quarter, management targets revenue of roughly $33.5 billion — a 260% year-on-year leap. DRAM contract prices are expected to rise as much as 63% quarter over quarter, and Gartner projects a 125% jump for the full year. Roughly 70% of all memory chips produced in 2026 are expected to feed AI data centers.

To keep up, Micron has lifted its full-year capital expenditure plan from $20 billion to more than $25 billion. New fabrication capacity typically takes two years to come online, meaning the supply deficit could persist well into 2027. For 2027, the market already anticipates further spending increases. The bet: as long as HBM remains scarce, heavy investment acts as a lever on future growth. If NAND prices soften or the cycle turns, free cash flow will come under sharper scrutiny.

Inside the company, however, a notable pattern emerged. Over the past three months, Micron insiders have been exclusively selling shares. Chief executive Sanjay Mehrotra disposed of 40,000 shares in early May under a pre-arranged trading plan. Board member Steven J. Gomo sold 2,000 shares in mid-May at prices between $786 and $788. After a 130%-plus run, such moves often reflect personal wealth management rather than a bearish signal — but they add to the cautionary backdrop.

The analyst community is split in an unusual way. Twenty-seven of 30 analysts rate the stock a buy. Deutsche Bank’s Melissa Weathers has the highest Wall Street price target at $1,000. Yet the consensus average sits at roughly $592, implying downside from current levels. That divergence captures the uncertainty: bullish on the story, nervous on the entry price.

Technically, the stock looks overextended. The relative strength index stands at 77, a level that typically triggers profit-taking. Even after Friday’s retreat, the shares are up 131.97% year to date and a staggering 630.85% over twelve months. The pullback did not come out of nowhere; the extreme run had made a correction almost inevitable.

Risks are not limited to valuation. Micron ranks third in HBM behind SK Hynix and Samsung Electronics. SK Hynix already supplies Nvidia with critical components. If either rival expands capacity faster than expected, Micron’s pricing power could erode. Moreover, a potential strike at Samsung starting in late May could rattle an already tight market — but any disruption at a competitor might also tighten supply further, a double-edged sword.

Investors get fresh clarity this week. On May 20, Micron’s management speaks at the J.P. Morgan Global Technology Conference, where commentary on pricing, supply chains, and investment plans could move the stock. Third-quarter results follow on July 1. The company’s own target of $33.5 billion in revenue will be the clearest test yet of whether the HBM backlog can sustain a valuation that has already priced in so much.