Micron's $22 Billion Deposit Is a Liability, Not a Demand Signal

Micron's $22 Billion Deposit Is a Liability, Not a Demand Signal

The number behind the record

Micron posted the biggest quarter in its history on June 24: record revenue of $41.46 billion in the fiscal third quarter ended May 28 — more than four times the $9.3 billion the same quarter produced a year earlier — GAAP net income of $28.2 billion, operating cash flow of $25.4 billion for the quarter, and guidance for the period ahead of $50 billion in revenue at roughly 86% gross margins. The shares jumped about 12% after hours.

The figure investors circled was not on the income statement. It was $22 billion: customer commitments under Micron's new class of multi-year "Strategic Customer Agreements," with roughly $18 billion of that arriving as cash deposits from customers paying ahead to lock in memory supply.

One month later, the chief executive who presented those numbers sold 40,000 shares for about $37.3 million. Four weeks after that, the same executive sat across from a television host who had flown to Boise and told viewers the stock was "radically undervalued" and "a national treasure."

Two instincts compete. The first reads the sale as a tell: executives sell when they know something. The second reads the deposit as unanswerable proof of a durable boom. Both misread the document on the table. The sale was executed under a plan written months earlier, and the $22 billion sits on Micron's balance sheet as a liability — cash received for chips not yet shipped — being marketed to the market as if it were an asset. That reclassification is the story.

What a customer deposit is

A customer deposit is cash a customer pays for product it has not received. Accounting files it as a liability: the company owes the customer either the goods or the money back. Micron's version is unusually strong on paper. The 10-Q filed June 25, 2026 describes take-or-pay contracts with price bands whose floor prices are expected to support gross margins "well above our peak quarterly margins in any past cycle." Translated: the customer has agreed to pay — in cash — whether it takes delivery or not, and the price floor is written to keep Micron's margin intact no matter what happens to spot memory prices.

That is the bullish reading, and it deserves a fair trial. Deposits that arrive ahead of shipment are the mirror image of the cash-flow anomaly this column normally chases. The usual problem is profit that never becomes cash. Here cash arrives before profit, which is better quality, not worse, provided the conversion holds. The deposits are "firm commitments" backed by "actual funds" customers have provided — real money, not letters of intent.

The liability that has not landed yet

Then there is the timing detail management let slip. On the earnings call that accompanied the June report, Micron said the deposits would "show up on our balance sheet more prominently in fiscal Q4."

The May 28 balance sheet backs that up. Cash and equivalents of about $25 billion. Total current liabilities of $19.5 billion, of which "other current liabilities" stood at just $3.4 billion. The roughly $18 billion of deposit cash was nowhere prominent. The celebrated figure is still walking toward the statements, and the quarter that ends in late August is when it lands.

When it lands, the same number will read two ways at once. As committed revenue, it dresses up a net-cash story — the company reports net debt of negative $20.3 billion and a current ratio above 340%. As a claim, it loads years of future delivery obligations onto the company: a contractual duty to convert $18 billion of prepaid money into shipped silicon, or to give part of it back under whatever cancellation and refund terms the contracts carry. The market gets to choose its favorite description; the audited statements will carry both.

Cash before profit

The cash-flow statement shows why this is a genuinely unusual setup rather than a gimmick. Trailing operating cash flow is $51.4 billion, free cash flow $26.2 billion, against roughly $25 billion of capital expenditure on the same twelve-month basis. The first nine months of the fiscal year produced $45.7 billion of operating cash against $11.8 billion in the year-earlier interval. Customers are effectively pre-financing the buildout with deposits, and the money concentrates exactly where the scarcity story lives: the HBM and server-DRAM unit, whose revenue grew 653% year over year to $11.5 billion at an 83% operating margin.

The question is no longer where the profit comes from. It is who is financing it, and whether the prepayment converts. Four things can break that conversion, and only one is visible in the press release.

First, coverage: what share of Micron's total volume the strategic agreements cover is not stated, and the entire "durability" case rests on that percentage. Second, cost: if the cost of producing a gigabit falls faster than the floor prices, the margin floor erodes even while the contract pays. Third, concentrated counterparties: a single customer amounted to 10% of total revenue in the first nine months of the fiscal year and 16% in the year-earlier period, meaning the deposits lean on a short list of names. Fourth, classification: whether the deposits carry refund obligations is not fully specified, and the document that should answer it — the fiscal-year 10-K, expected in October — has not been filed. None of these is an accusation. They are the pages the "durability" pitch has not shown yet.

There is also the premise underneath the premise. A class action filed in late June accuses Micron, Samsung and SK Hynix of conspiring to fix memory prices by limiting production of certain chips. That is an allegation, procedurally unproven, and it should be tagged as such. But it aims at the exact thing the floor margins depend on: the structural scarcity that lets a take-or-pay contract promise more than 80% gross margins. If scarcity is engineered rather than physical, the price bands lose their bedrock.

The $37 million sale, graded

Now the sale, treated as evidence rather than atmosphere. On July 24 — four weeks before the Boise interview — the CEO sold 40,000 shares worth about $37.3 million. The mechanics matter. A portion of the block, 8,715 shares, was executed under a pre-established Rule 10b5-1 plan — a schedule set in advance so the insider is not trading on information at the moment of sale — at prices between $942.73 and $965.85, with another 31,285 shares changing hands the same trading day. Filings showed he still held roughly 600,000 shares afterward.

A scheduled sale is an alibi, not an acquittal. It tells you the transaction date was fixed before the record quarter could inform it, which is precisely why this trade conveys almost nothing about what the executive knew on July 24. On the evidence ladder this sits at Level Two — an anomaly repeated often enough to matter, since filings had already shown more than $45 million of CEO sales for the year by early July — and not a single level higher. There is no restatement, no regulator finding, no charge, no admission. A securities-fraud class action did surface in July, but its class period runs between March 29, 2023 and December 18, 2024 — a different era from these contracts. An earlier shareholder derivative suit over roughly $70 million of sales ahead of a weak quarter was refiled in Idaho federal court in early 2025; a complaint is an allegation about a different year.

What is fair to hold against the timing is the corridor it chose. The sale lands between the June 24 record release and the August 20 praise session, in the same stretch in which a sell-the-news tape took 6.9% off the stock on July 2 despite record earnings — a session that mixed the price-fixing class action, insider-selling headlines and a pullback across Asian memory names. Even the most recent tape leaned to the sell side, with retail, medium, large and block orders all netting out as outflows.

The December 9 hole

The third date explains why the sale stings even at Level Two. Micron cannot buy back its own stock until December 9, 2026, the second anniversary of its CHIPS Act funding agreement, when the restriction attached to that subsidy expires. So while the CEO's planned sales executed this summer, the company was contractually barred from the offsetting action, even as rivals returned cash: SK Hynix announced a $29 billion buyback, and Samsung may announce more than $78 billion in returns.

The company's answer to that arithmetic is a commitment, not a document. The chief financial officer has said Micron will return 100% of excess cash to shareholders beginning December 9, and one published estimate gives the unlock real size: with nearly $24 billion of net cash, post-December repurchases could retire roughly 12% of the shares. The market has already priced part of the promise — Micron jumped about 6.6% on August 13 on reports of the coming capital-allocation plan.

The shareholder invoice

Price the position as of the latest session: about $966 a share, a $1.09 trillion market capitalization, roughly eight and a half times the 52-week low of $114, up about 134% in four months and 239% since January. The trailing earnings multiple is about 21 times, and the stock trades near 12 times sales. The "radically undervalued" label, by contrast, rests on a forward multiple near 6 times — and that switch in basis is the entire argument, because the cheapness claim only works on next year's deposit-subsidized earnings, not this year's. Even on EV/EBITDA, at 15.7 times, Micron screens below memory and storage peers Sandisk, Seagate and Western Digital, which is the kind of relative bargain that forms exactly when the market doubts the forward numbers.

Three ways this pays out.

Resolved. The take-or-pay contracts convert as written, the floor margins hold through the cycle, the deposits finance the buildout, and the December unlock starts retiring shares. At the forward multiple the bulls cite, the stock is cheap, and the CEO's $37 million is a rounding error inside a much larger fortune — diversification, not information. This is the case the current disclosures most resemble.

Persistent but lawful. Memory spot prices normalize, or per-gigabit costs fall faster than the floors. The deposits still get worked off and earnings still print, but the premium compresses — and at 12 times sales, the multiple has more room to fall than the earnings have room to surprise. The "national treasure" framing then describes a great company at a cyclical peak, which is the historically normal condition of memory stocks, floor contracts notwithstanding.

Misstated. No evidence supports it today. It would require the deposits to be less firm than described, the coverage claims to overstate, or the scarcity to be manufactured. The fiscal-year 10-K in October, the late-August balance sheet where the deposits finally land, and the resolution of the price-fixing case are the records that could establish any of it. This case is a monitoring item, not a conclusion.

The accounting does not need the drama

The balance of the evidence reads this way. The $22 billion figure is a liability doing triple duty: it finances the buildout, it dresses the balance sheet, and it underwrites the forward valuation — all while the company's own conference call concedes the money has not fully arrived. The CEO's sale is scheduled, small relative to his stake, and legal, which is exactly why it is no evidence of wrongdoing and also no reason to relax. The deposit is the fact worth watching: whether it converts into years of floor-protected, high-margin revenue is the question the late-August balance sheet and the October 10-K will answer. The next settling event is a document, not a price.