Micron Capex Is Going Up, Not Down — The Only Question Is Whether It's Margin-Accretive HBM or a Classic Memory Overbuild

The question I am getting most this week is whether Micron is about to telegraph a capex cut on Wednesday. The answer, on the public evidence, is no. Capex is going up, not down. FY26 was raised to above $25B. FY27 is expected to step up meaningfully on top of that. The 10-Q says the same thing in formal language. There is no signal yet that they are pulling back.

That doesn't make this print bearish. The better question is what kind of capex this is — margin-accretive HBM and advanced-DRAM spend, or the early shape of a classic memory-cycle overbuild. The trade is not 'capex up = sell.' The trade is: capex up + gross margin guide up = bullish. Capex up + gross margin guide flat or down = sell-the-news. That's the whole framework. Everything below is how I get there.

Near term, capex up is not automatically bad for gross margins. For this earnings print specifically, higher capex is more likely to be interpreted as proof that demand is strong — especially if management ties it to HBM, DRAM tightness, strategic customer agreements, and constrained supply. That's the bull narrative the tape has been pricing for nine months.

Medium term, capex up becomes the bear case. If Micron keeps accelerating equipment and cleanroom spend into 2027 and 2028, and pricing starts flattening, the market will start discounting margin compression long before the actual gross margin turns. That is how every memory peak in history has been called — not on the GM print, but on the capex-versus-pricing divergence.

So the simple rule for Wednesday: capex up paired with a Q4 gross margin guide that pushes through 81% is the bull case the stock is already paying for. Capex up paired with a Q4 guide stuck at or below 81% is the first sign the margin curve is flattening. Trade the combination, not the capex number in isolation.

Micron's own materials are not subtle on this. In the latest Q2 deck, the company said fiscal 2026 capex is now expected to be above $25B, and that the majority of the increase from the prior estimate is tied to cleanroom facility-related capex — Tongluo as the largest factor, followed by higher U.S. fab construction spending. They also flagged that fiscal 2027 capex should 'step up meaningfully,' with construction-related capex rising by over $10B year over year and equipment spending also higher.

The 10-Q says the same thing in more formal language: Micron estimates 2026 property, plant, and equipment capex, net of government incentives, will be above $25B, with actual amounts varying depending on market conditions and incentive timing. The only 'capex could go down' language anywhere in the filings is the standard flexibility disclosure — they can adjust if the market changes. There is no concrete evidence yet that they are backing off.

That is not a capex-cut setup. That is an expansion setup.

The important detail is what kind of capex this is. A lot of the FY26 increase is cleanroom and construction capex, not immediate wafer-output capex. Construction does not instantly flood the market with bits. It becomes a margin issue later — once equipment is installed, production ramps, depreciation starts hitting cost of goods, and the supply hits the market. That is a 2027 and 2028 problem, not a Q3 problem.

Micron's Q2 deck spells this out: 2026 DRAM and NAND bit demand should remain constrained by supply, and management specifically blames cleanroom constraints, long construction lead times, higher HBM trade ratios, and slower bits-per-wafer growth from node migrations. That supports the near-term margin bull case directly out of the company's own slides.

TrendForce echoes it: 2026 memory capex growth should have limited impact on bit supply growth because spending is focused more on process upgrades, hybrid bonding, HBM, and high-value products rather than raw capacity expansion. TrendForce expects DRAM industry capex to rise from $53.7B in 2025 to $61.3B in 2026, but says cleanroom limits keep bit-supply growth constrained. That is why margins can keep going higher even while capex is going higher.

Micron guided Q3 revenue to $33.5B plus or minus $750M, and gross margin to roughly 81%. At that revenue base, one point of gross margin is about $335M of gross profit. That is the unit of measure for everything that follows.

If capex, ramp, and depreciation eventually add $500M of quarterly manufacturing burden, that is roughly 150 basis points of gross-margin drag at $33.5B in revenue. That is manageable if pricing is still ripping higher. It is not manageable if ASPs flatten. That sensitivity is the entire reason the call's pricing language matters more than the capex number itself.

Right now, Micron's Q3 margin expansion is being driven by higher price, lower cost, and favorable mix, according to the company's own Q2 presentation. If they say the same thing again for Q4 and guide gross margin above 81%, the stock can still squeeze. But if they tell you capex is stepping up while gross margin is stuck around 81%, that is the first crack in the bull thesis.

This is the part most people miss. TrendForce says HBM wafer revenue was overtaken by DDR5 64GB RDIMM in Q1 2026, and that HBM profitability has fallen below DDR5 64GB RDIMM since Q1 2026. That means 'more HBM' is not automatically 'more margin' unless HBM pricing resets higher. That is why the 2027 HBM pricing commentary on this call matters so much — more than the FY27 capex number itself.

TrendForce also says HBM wafer input among the top three suppliers should rise from about 18% of DRAM wafer input at end-2025, to 22% in 2026, and 30% in 2027. But HBM bit supply is expected to be only 8%, 9%, and 13% of total DRAM bit supply over the same period.

Translation: HBM eats a lot of wafer capacity for relatively fewer bits. That is bullish if customers pay up for it. It is a margin problem if HBM pricing does not rise enough to compensate for the wafer tradeoff against high-end DDR5 RDIMM that, right now, is the more profitable wafer.

These are already visible in the public record. One: FY26 capex was raised to above $25B. The prior number was lower; the revision was up. That is the cleanest signal. Two: FY27 capex is expected to step up again. Management specifically called out over $10B of additional construction-related capex year over year in fiscal 2027, plus higher equipment spend on top of that.

Three: Tongluo adds expansion optionality. Micron closed the Tongluo acquisition and plans to begin construction of a similar-sized second cleanroom by the end of fiscal 2026. The existing fab is expected to support meaningful product shipments beginning in fiscal 2028. Four: Idaho, New York, Singapore, Japan, and India are all active expansion nodes. The 10-Q lays out first DRAM wafer output from Idaho in mid-calendar 2027, a second Idaho fab construction expected to begin in 2026 and be operational by end-2028, first New York fab supply in 2030 and beyond, Singapore HBM packaging expansion in 2027, and Singapore NAND cleanroom output in the second half of 2028.

That is a multi-year, multi-site spend cycle. There is no version of this where capex quietly drifts lower in the next twelve months unless management explicitly chooses to slow it.

There are not many right now. The only real signs to watch would be language shifts on the call itself. A capex-down signal sounds like this: 'pacing equipment installs,' 'aligning capex with customer demand,' 'prioritizing free cash flow,' 'construction continues, but tool move-in is being adjusted,' 'customer demand visibility has changed,' 'strategic customer agreements are taking longer to finalize.'

The important distinction is the reason. Capex down because of discipline is bullish for margins — it means they are protecting supply tightness and pricing. Capex down because customers are pushing out demand is bearish for margins — it means the AI memory cycle is losing slope. For this earnings call, the most important question is not whether capex is up or down. It is why.

Bullish margin path, which is what bulls want: capex stays high, but management says it is tied to committed demand, strategic customer agreements, HBM4 and HBM4E, and supply that remains tight through 2027. HBM pricing resets higher. DDR5 and server DRAM pricing stays firm. Gross margin guide moves above 81%. The market reads high capex as growth capex, not overbuild capex.

Bearish margin path, which is what bears want: capex goes higher, equipment spending gets pulled forward, but the gross margin guide is flat or lower. Management talks about startup costs, customer mix friction, HBM yield friction, pricing moderation, or wafer-allocation tradeoffs. The market reads high capex as future depreciation and future supply pressure, and the stock starts pricing the back half of the cycle.

The single worst combination on Wednesday would be: FY27 capex step-up confirmed, plus Q4 gross margin guide flat at around 81%, plus no clear HBM pricing upside in the commentary. That would say the company is spending more, but incremental margin is no longer expanding.

In a name that has already run hard, that is the sell setup. It does not need to be a miss. It just needs to be a print where the marginal dollar of capex is no longer producing a marginal point of gross margin. Once that is on the tape, the multiple compresses fast.

Good language, the kind that keeps the bid in: 'HBM pricing is improving for 2027.' 'HBM4 yields are ahead of plan.' 'Customer commitments support our capex plans.' 'Strategic customer agreements improve visibility.' 'Supply remains constrained beyond 2026.' 'Gross margin should continue expanding.' If you hear three of those in the prepared remarks and one of them in the Q&A, the bull case is intact.

Bad language, the kind that fades the rip: 'Equipment spend is increasing faster than expected.' 'Startup costs are rising.' 'Customer mix is a headwind.' 'Gross margin remains around current levels.' 'Pricing improvement is moderating.' 'We are optimizing allocation between HBM and conventional DRAM.' That last one matters because TrendForce already showed that high-end DDR5 RDIMM economics may be better than HBM right now — and any hint that Micron is wrestling with that allocation internally is a tell.

For this print: capex up is only bearish if margins stop going up. Three scenarios, each with a clean trade.

Scenario one — 'Capex above $25B, FY27 up meaningfully, Q4 gross margin above 82%, HBM pricing stronger, demand contracted.' That is bullish. Do not fight it. Long delta into the close, hold the run.

Scenario two — 'Capex above $25B, FY27 up meaningfully, Q4 gross margin around 81%, strong demand but no margin acceleration.' That is fadeable. The setup is in the price; the print is not adding to it. Take profits, fade the after-hours pop into the next morning.

Scenario three — Micron cuts or slows capex. Bullish if they say demand is strong but they are staying disciplined. Bearish if they say customer demand or AI visibility changed. Listen to the reason, not the number.

The evidence says capex is going up, not down. Near term that can actually support higher margins, because it confirms supply is tight and demand is strong. But the minute capex acceleration stops producing gross-margin acceleration, the market will start treating it like a memory-cycle peak.

The number to watch on Wednesday is still simple. Q4 gross margin guide above 81% means capex is being rewarded. Q4 gross margin guide flat or down with a higher FY27 capex line means capex has become the bear case. Same company, same quarter, opposite trades — decided by one sentence in the guidance.

This article is a personal opinion piece by Guy Gentile. It is not investment advice, a research report, an offer to buy or sell securities, or a recommendation. Estimates and figures referenced are drawn from Micron's publicly disclosed Q2 fiscal 2026 earnings materials, the company's 10-Q filed with the SEC, and published TrendForce commentary. The author may at any time be long, short, or flat positions in MU, related semiconductor names, options, or hedges, and may change those positions without notice. Do your own work.