Micron hits $1 trillion market cap on AI memory demand

For four decades, the rule on memory was simple: it is a commodity, and commodity producers are valued on the next downturn, not the next quarter. DRAM glut cycles bankrupted competitors and held Micron's multiple at a fraction of its logic-chip peers. That assumption is what UBS's Timothy Arcuri just torched. Arcuri's new framework argues that the locked-in long-term agreements Micron has signed with hyperscalers — covering pricing and volume through 2029 — are functionally an infrastructure franchise, not a spot market. His direct words: 'We believe the market will start to put a more "normal" multiple on the stock and MU will continue to re-rate higher as more details emerge about the structural changes AI has driven to the entire memory complex' ^[1].

The Street is starting to agree. D.A. Davidson's Gil Luria put it bluntly — 'Memory companies are becoming less cyclical' ^[2]— and the price action ratified the view. Micron closed up 19.29% at $895.88, its best session since November 2011, on the way to the 11th-largest market cap among U.S. public companies ^[3]. Twelve months earlier, the same stock traded under $100.

What is being re-rated is not Micron's manufacturing. It is the predictability of its cash flows. If the contracts hold, a memory company starts to look like a utility with optionality. If they don't, the cycle takes back what it gave.

To understand why Wall Street is willing to extend Micron a multi-year multiple, look at the physical constraint underneath the trade. High-bandwidth memory now consumes roughly 23% of total DRAM wafer output, and all three suppliers — Micron, SK Hynix, and Samsung — have sold out their 2026 HBM capacity. New fab capacity does not reach meaningful volume until 2027 or later ^[4]. That is not a marketing line; that is a 12- to 18-month physics-and-CapEx wall.

Micron CEO Sanjay Mehrotra confirmed the squeeze on his own end of the table, admitting the company can currently meet only 50% to 66% of customer demand for HBM and framing memory as 'a strategic asset for our customers' rather than a commodity input ^[5]. The contractual response — 3-to-5 year long-term agreements with AWS, Microsoft Azure, Google Cloud, Meta, Oracle, and Nvidia, with partially fixed pricing — converts that shortage into visibility ^[6]. TrendForce expects another roughly 20% HBM price increase in early 2026 on top of an already elevated base ^[7].

The oligopoly is also self-reinforcing. SK Hynix had already sold out 2026 by October 2025 ^[7], which pushed marginal demand to Micron — the only U.S.-domiciled supplier in the three-firm club — and gave it pricing leverage it has historically never enjoyed. The single most important number in the entire bull case is not a revenue forecast; it is the number of new HBM fabs coming online before 2027. That number is roughly zero.

The financial scoreboard explains why this stopped being a chip story and started being a megacap story. Since the March 30 market low, Micron has appreciated roughly 180% and added about $650 billion in market value — a dollar gain that nearly matched Amazon's over the same window despite starting from a base less than a tenth the size ^[8]. Year-to-date the stock is up 213%; trailing 12 months it is up roughly 780% ^[9].

The operating side is moving in lockstep. Fiscal Q1 2026 revenue came in at $13.64 billion (+57% year-over-year), with Q2 FY2026 reported at $23.86 billion ^[5]. Micron has hiked FY2026 CapEx to $20 billion to chase the supply gap, with HBM4 ramping in Q2 of that fiscal year ^[10]. UBS now models forward EPS of $155 in 2027, $167 in 2028, and $117 in 2029 — the 2029 dip reflects an assumed pricing normalization, not a demand collapse — and applies a 15x multiple to that 2029 number to back into its $1,625 target, implying roughly $1.8 trillion in market cap ^[6]. The HBM market itself is projected to reach about $100 billion by 2028 ^[4], growing alongside an AI-server architecture that requires 6-to-8x more DRAM than conventional enterprise servers ^[4].

The contrarian case is not that AI demand is fake — it is that the second derivative is fragile. Bearish analysts note memory remains structurally cyclical and that an AI capex slowdown, a geopolitical shock, or a Samsung HBM4 redesign success could reverse the pricing power that the entire $1,625 thesis rests on ^[11]. Up to 40% of advanced wafer capacity is being redirected toward HBM, which is already rationing commodity PC and smartphone DRAM and is forecast to lift broader server prices in 2026 ^[7]— a tax that eventually shows up in someone's margin compression.

Community discussion is split along the same fault line. One popular framing on stock forums points out that Micron is 'pulling 1/4th of Nvidia's quarterly revenue at 1/10th their valuation,' which is the bull thesis distilled. Sitting opposite is the older Wall Street muscle memory — 'memory is a commodity, cure for high prices is high prices, China is coming' — and concrete shortage anecdotes circulating in builder communities, where server DDR5 ECC RAM has reportedly jumped from $125 to roughly $1,000 per stick, both a sign of real undersupply and a leading indicator of demand destruction in adjacent markets.

The through-line of the skeptical read is execution risk, not demand risk. Mehrotra's pricing power and Arcuri's multiple now matter more than the headline number. If either cracks before HBM5 ramps — a Samsung qualification win, a hyperscaler pause on the next training-cluster build-out, or even Chinese supply coming online faster than expected — the $1T print becomes the high-water mark instead of the floor.