Sector Intelligence

Micron’s Customers Are Paying Up Front. That’s the Number I’m Watching.

Oct 02, 2026

By Ieva Bružaitė | Founder, Orion Capital | 2 October 2026

Micron reported on Wednesday evening, and the number everyone quoted was 379%. That is how much revenue grew from a year ago, to $54.229 billion in a single quarter, against $11.315 billion in the same period of fiscal 2025. It is a ridiculous figure for a company that sells a commodity.

I would point you somewhere else in the release. Buried in the balance sheet and cash-flow statement is a line that deserves more attention. Micron’s noncurrent customer contract liabilities jumped from $568 million at the end of fiscal Q3 to $12.895 billion at fiscal year-end. Separately, the full-year cash-flow statement shows $12.747 billion of proceeds from customer contract liability deposits.

That is money customers handed over before receiving anything. They did it to secure supply.

Think about what has to be true for a buyer to behave that way. The buyer must believe the product will be scarce. It must believe that waiting is riskier than parting with cash early. And it must have concluded that price is a secondary concern. None of that shows up in a growth rate, and all of it tells you more about the memory market right now than the headline does.

The quarter itself

Fiscal Q4 2026	Result	Expected	A year ago
Revenue	$54.229bn	$50.8-51.1bn	$11.315bn
Non-GAAP EPS	$33.42	$31.45-31.61	$3.03
DRAM revenue	$39.8bn (73% of sales)	–	+343% YoY
Non-GAAP gross margin	87.0%	–	45.7%
Operating cash flow	$43.973bn	–	$5.73bn
FQ1 2027 guidance (non-GAAP)	Revenue $61.5bn +/- $1.5bn; gross margin ~86.25%; EPS $38.15 +/- $1.00	$56.6bn; EPS ~$35.92	–

The full year is almost harder to absorb. Revenue reached $133.188 billion against $37.4 billion the year before. Operating cash flow was $89.675 billion, which is 67% of everything the company billed. Adjusted free cash flow came to $62.308 billion after $27.367 billion of net capital spending.

DRAM alone crossed $100 billion for the year. Data centre SSDs brought in close to $10 billion in the fourth quarter, more than ten times the year-ago figure, and now account for over two-thirds of Micron’s NAND business.

Then there is the gross margin: 87.0% on a non-GAAP basis, up another 210 basis points from the previous quarter, against 45.7% in the same quarter a year earlier. Memory manufacturers historically earn 20% to 40% gross margins and lose money outright at the bottom of the cycle. Micron is currently running at a margin most enterprise software companies would be pleased with, while operating fabrication plants that cost tens of billions to build.

So why did the stock barely move?

Micron beat on revenue by roughly $3.5 billion. It beat on earnings by nearly $2 a share. It then guided fiscal first-quarter revenue to $61.5 billion give or take $1.5 billion, against a consensus near $56.6 billion, with non-GAAP earnings of $38.15 plus or minus a dollar against roughly $35.92 expected. By any normal standard this was an exceptional print followed by an exceptional outlook.

The stock went almost nowhere. Shares ticked up a fraction in extended trading on Wednesday, slipped below $1,060 at one point, and have spent Thursday moving between roughly $1,041 and $1,089 around a $1.2 trillion market capitalisation. Goldman Sachs raised its price target and expected the shares to rise modestly. Mizuho kept its buy rating. The tape has been less decisive than the analysts.

Two details in the release explain the hesitation better than sentiment does. Micron guided gross margin to roughly 86.25%, a little below the 87.0% it just delivered. And capital spending is climbing, which slows the conversion of those enormous profits into free cash flow. Neither is alarming. Both are early signs of a company spending into a boom rather than simply harvesting one.

Readers who have followed this platform through the summer will recognise the pattern. SK Hynix posted the best quarter in its history in July and fell 9.6%. Marvell beat and raised in August and fell 10%. Broadcom beat on every headline number in September and fell too. In each case the business performed and the share price did not, because expectations had already travelled further than the results could.

Micron is a mild version of the same thing, and in one respect a more encouraging one. The shares did not fall. They have risen more than 500% over the past year and the market capitalisation has passed $1.2 trillion, which means the bar was high and the company cleared it without being punished for merely clearing it. That counts as progress compared with July.

What the market is actually unsure about

The hesitation is not about demand. It is about the word cycle.

Memory has always been cyclical. Shortages produce extraordinary margins, extraordinary margins attract capital, capital builds capacity, capacity arrives late and prices collapse. Anyone who has held Micron for a decade has lived through this at least twice. An 87% gross margin is not a steady state in this industry. It is what the top of a cycle looks like.

So the market is doing something quite rational. It is awarding Micron a low multiple on very large earnings, because it does not believe those earnings persist. The argument is not that the company is doing badly. The argument is that it cannot keep doing this well.

Micron’s answer is that the structure has changed. Management points to Strategic Customer Agreements, to supply committed years ahead, and to those prepayments. Sanjay Mehrotra told analysts the company has a strong roadmap for future HBM products and is working with Nvidia on what he described as the industry’s first custom high-bandwidth memory implementation. Custom parts are harder to substitute than commodity DRAM. If memory becomes something designed jointly with the customer rather than bought off a price list, the old cycle logic weakens.

I find that argument plausible and unproven. It is the single most important open question in the sector, and it will not be settled by one quarter, however good.

The detail I keep coming back to

Micron is the smallest of the three high-bandwidth memory makers. SK Hynix leads, Samsung is larger in overall scale, and Micron sits third. It is also the only American one, which in 2026 is a commercial asset as much as a geographic fact. Mehrotra attended the White House state dinner for Xi Jinping last week and a summit on AI regulation two days later. Memory executives did not used to spend their time this way.

Here is the thing about those prepayments, though. Micron ended the year holding roughly $12.9 billion of other companies’ money as noncurrent customer contract liabilities, up from $568 million three months earlier, because those companies are more afraid of not having memory than of tying up cash. That fear is the real product of this cycle. It is also the thing that will disappear first when supply catches up, long before revenue does.

Watch that line next quarter. If it keeps growing, the shortage is intact. If it flattens while revenue still climbs, something has started to change, and you will have seen it a couple of quarters before the income statement shows it.

What we are watching

Noncurrent customer contract liabilities on the balance sheet, as an early read on how tight supply really is
Whether the custom HBM work with Nvidia produces a named, dated commitment rather than a roadmap reference
Gross margin against the roughly 86.25% guide, given that 87.0% leaves far more room to fall than to rise
Capital expenditure across the industry, since this is where the next cycle is being built right now
Treasury yields, which did more to move chip stocks in September than any earnings report did

The Orion View

This was a very good quarter from a company in the right place at the right moment, and the market treated it as such without getting carried away. That seems about right to me.

AI needs memory. That part is not in dispute. What I still cannot tell is whether HBM has changed the economics of this industry permanently, or merely made this particular cycle unusually profitable. Micron’s prepayments, custom work with Nvidia and multi-year agreements make the structural-change case stronger. Twenty years of memory cycles make me reluctant to declare the old model dead. I would want to see one more year of evidence before deciding. Until then, a stock trading at a modest multiple of peak earnings is neither obviously cheap nor obviously expensive. It is a bet on which of those two stories turns out to be true.

Disclaimer: This publication is independent research and reflects the views of the author. It is provided for informational purposes only and does not constitute investment advice or a recommendation to buy or sell any security. Investors should conduct their own analysis before making investment decisions.

Sources

About the Author

Ieva Bružaitė is the Founder of Orion Capital, an independent research platform covering AI, semiconductors and long-term investing.

Ieva Bružaitė - Founder of Orion Capital

← Back to Orion Capital