Micron at $895: The Market Has Already Booked 2027’s HBM Earnings

The market’s story on Micron is tidy enough to fit on an index card. HBM supply is sold out through 2026, hyperscaler capex is structural rather than cyclical, gross margins are guided to roughly 81 percent next quarter, and the company has effectively transformed from a commodity DRAM cyclical into something closer to a contracted-revenue compounder. The stock has done what stocks do when that story takes hold. It went from $412 to $895 in three months, with most of that ramp packed into the four weeks after the March print, when revenue beat by nearly four billion dollars and management guided Q3 to $33.5 billion at the midpoint. Analyst price targets stretched from $249 to $1,100. Passive flows showed up after the S&P 100 inclusion. The CHIPS Act money for the New York and Idaho fabs lowered the perceived domestic capex burden. Every input to the bull case has, in some form, already happened.

And that is precisely the problem.

At $895, the equity is priced as though the next two years of HBM bit growth, pricing, and margin will land exactly where the guide implies, with no air pocket between contracted supply and realized revenue. The forward multiple near ten times looks cheap only if you treat forward earnings as a planning assumption rather than a peak. Operating income for the February quarter came in at $16.14 billion. The same line was $1.30 billion in FY2024. That is not margin recovery. That is the top of a pricing cycle being annualized into perpetuity by a market that has historically been bad at telling the difference.

The evidence consensus is ignoring is in plain view. National semiconductor capacity utilization was 71 percent as of January, well below the roughly 80 percent historical norm. Micron’s ending inventory was $8.3 billion at 123 days, with DRAM under 120. Capex is guided above $25 billion for FY2026 against $13.8 billion the prior year, a near-doubling of capital intensity into a node transition where the 1γ DRAM ramp and G9 NAND are both targeted to cross majority bit share by mid-calendar 2026. HBM is sold out. The rest of the bit supply is not. Memory has always settled on the marginal wafer, not the contracted one, and in this cycle that wafer is being financed at a capital intensity Micron has rarely sustained without subsequent free-cash-flow compression.

Options markets are not subtle about the positioning. Over $1.4 billion in profit has piled into the $400 strike calls. That is not hedging. That is a directional bet that has already paid, looking for someone to take the other side at $895.

The macro read is selective. The CHIPS Act grant of roughly $6.1 billion does reduce the net cash cost of the domestic fab buildout, and the export-control posture toward China gives Micron a defensible structural position against rivals that cannot tape-out at the leading edge. Both are real. Both have also been in the price for months. The 2-year Treasury at 3.80 percent against a 3.64 percent Fed Funds rate is the bond market pricing tighter financial conditions ahead, the wrong backdrop for a stock whose forward earnings depend on a friendly discount rate while $25 billion a year goes into the ground.

The strongest bull case is not the HBM sellout. It is the claim that memory has structurally consolidated to three players who have learned capex discipline, and that AI training and inference create a sustained second leg of bit demand that breaks the historical four-year cycle. In that frame, $895 is mid-cycle, not peak, and the normalized earnings power that justifies it sits somewhere north of $40 in EPS once HBM4 ramps in calendar 2027. The market made similar assumptions about memory pricing power during prior data-center buildouts, and each time the argument was backed by genuine demand inflections that did, in fact, happen. The demand was real. The duration estimate was what broke.

The trouble with that case is what the industry data say if you read them in order. Utilization drives pricing. Pricing drives gross margin. Capex discipline either preserves the cycle or destroys it. Right now utilization is below norm at the industry level, pricing is at a cyclical peak, gross margin is being guided to eighty-one percent, and capex is being raised by roughly $11 billion year over year. Three of those four indicators point the same direction memory always points before bit supply catches up. The fourth, capex, decides how violent the catch-up is.

Inventory days at customers is the variable I will watch, not backlog. Hyperscalers stockpiling HBM ahead of GPU deliveries looks identical to genuine end demand until the GPU ship schedule slips by a quarter. Then it looks like a glut.

If Micron sustains gross margin above 70 percent for four consecutive quarters past the September 2026 print while inventory days at hyperscalers stay flat, my concern about cycle reversion is wrong and the price has been pricing reality, not a story. Until then, $895 is the market paying full freight for an earnings stream it has not yet seen, on a balance sheet about to absorb a capex doubling, in an industry where the marginal wafer always finds a way to ruin the contracted one.

Markets love stacking assumptions like that right before they rediscover gravity. Which assumption breaks first?

© Mathstock