Summary

  • Marvell's unconditional commitments to foundries and test-and-assembly partners rose by US$5.7621 billion in one quarter, from US$2.7568 billion at 2 May to US$8.5189 billion at 1 August 2026.
  • Almost all of the schedule change appears in fiscal 2028–2030. Those three buckets increased by a combined US$6.1964 billion, partly offset by a US$435.9 million decline in the remainder-of-fiscal-2027 bucket.
  • The company had announced new wafer and substrate capacity agreements after the May quarter, but it does not allocate the US$5.7621 billion increase by supplier, node, product, customer or agreement. A separate US$870 million deposit schedule is not the commitment increase.
  • Marvell says a significant portion of sales uses purchase orders rather than long-term customer commitments. Customers can cancel or defer on short notice, while Marvell's own supplier cancellations can require reimbursement, fees, forfeited advances or lost capacity priority.

Thirteen weeks changed the hard side of the ledger

At 2 May 2026, Marvell reported US$2.7568 billion of future unconditional purchase commitments to foundries and test-and-assembly partners. At 1 August, the same labelled table showed US$8.5189 billion. The difference is US$5.7621 billion, a 209% increase in one fiscal quarter. The new total is 3.09 times the previous one.

That aggregate is striking, but the fiscal-year bridge contains the useful evidence. In the fiscal-Q1 Form 10-Q, the remainder of fiscal 2027 carried US$2.2652 billion, fiscal 2028 US$174.4 million, fiscal 2029 US$68.4 million and fiscal 2030 US$66.3 million. The fiscal-Q2 Form 10-Q shows US$1.8293 billion, US$2.1254 billion, US$2.1783 billion and US$2.2018 billion in those respective buckets.

The near-term line fell by US$435.9 million as the quarter passed and commitments were used, revised or moved through the schedule. Fiscal 2028 increased by US$1.9510 billion, fiscal 2029 by US$2.1099 billion and fiscal 2030 by US$2.1355 billion. Together those three later years added US$6.1964 billion. A US$1.6 million change in fiscal 2031 and no change thereafter complete the arithmetic.

This decomposition matters because “US$5.76 billion added” does not mean one payment, one purchase order or one delivery date. It is the net change across a moving multi-year stock. The filings do not publish beginning obligations, new contract additions, purchases fulfilled, cancellations, price revisions and foreign-exchange effects as a formal roll-forward. The schedule proves the scale and timing perimeter. It does not supply the missing contract journal.

The historical ruler is equally revealing. The equivalent foundry-and-assembly total was US$1.5081 billion at 2 August 2025 and US$2.6658 billion at 31 January 2026. The latest figure is about 5.65 times the prior-year quarter's balance. But the “remainder” buckets are not identical periods, and the filing does not disclose wafer units or prices. The ratio measures the expansion of contractual exposure, not a 5.65-fold increase in capacity or demand.

New agreements are visible; their allocation is not

Marvell gave an important clue before the latest table appeared. Its May-quarter filing said that, after the quarter end, it had entered agreements to secure wafer manufacturing capacity through fiscal 2030 and substrate manufacturing capacity through fiscal 2033. In connection with those agreements, it committed to US$870 million of deposits, payable quarterly from fiscal Q2 2027 through fiscal Q2 2028.

The later commitment schedule adds most of its value in fiscal 2028–2030. It is therefore reasonable to read the newly disclosed agreements as context for the step change. It would be wrong to assign the whole US$5.7621 billion increase to those two agreements. Marvell does not say that. Nor does it say the increase is all payable to TSMC, all wafers, all substrates, all AI products, all 3nm production or all non-refundable.

The US$870 million deposit schedule is a separate instrument. Deposits can secure access before purchased components arrive. Purchase commitments measure future quantities or value that the company has agreed to buy. At 1 August, the balance sheet carried US$487.0 million of prepayments on supply-capacity reservation agreements, up US$208.2 million from US$278.8 million at January year-end. The balance, the future deposit schedule and the US$8.5189 billion purchase table are three different clocks.

Marvell's supplier language explains why the commitment can be economically hard without being an ordinary debt balance. It may cancel outstanding purchase orders, but must pay all costs and expenses incurred through cancellation. In some cases it may also face incremental fees, lose advances or lose priority to reserved capacity. Its TSMC reservation agreement requires target wafer purchases over multiple years; missing targets may forfeit a proportional share of advances.

This does not make US$8.5189 billion immediately payable. It does not put the whole sum in current liabilities or cost of goods sold. Components bought under the commitments can become prepayments, inventory and then product cost as their own conditions are met. Cancellation consequences can follow a different path. The useful description is a contractual purchase floor, not debt and not expense.

Customer choice remains the other side of the wager

Marvell's public customer contract is softer. The current 10-Q says products are typically sold through purchase orders rather than long-term purchase commitments. Some customers can cancel or defer on short notice without significant penalty. The company adds that customers who do have commitments may not honour them.

That is not an admission that current demand is fictitious. It is a description of control. Marvell can reserve supply from manufacturing partners; it cannot unilaterally make a customer finish product qualification, preserve an end market, accept a shipment or maintain its own inventory plan. Customers can change architecture, shift schedules, integrate vertically, choose a competitor or react to their buyers. A design win, booking, purchase order, shipment and recognised revenue are successive receipts, not synonyms.

The company's fiscal-Q2 earnings release provides strong demand signals. Quarterly revenue reached a record US$2.7393 billion, up 37% year on year. Data-centre revenue rose 46% to US$2.1715 billion and represented 79% of the total. Management called AI-related bookings “exceptionally robust,” raised its fiscal-2027 and fiscal-2028 revenue outlooks and guided fiscal-Q3 revenue to US$3.150 billion, plus or minus 5%.

Those statements explain the incentive to secure scarce advanced wafers, packaging and substrates. They do not fill the contract ledger. The release gives no bookings amount, cancellation schedule, customer deposit or allocation of committed capacity. Management's confidence may prove right. The evidence discipline is simply that a forecast and an unconditional supplier obligation do not acquire the same legal weight because they appear in the same quarter.

The asymmetry is not unique to Marvell; fabless semiconductor economics often require capacity decisions before downstream demand is fixed. What is unusual enough to demand attention is the disclosed scale and speed. The hard supplier-side stock increased by more than twice its entire May balance in thirteen weeks. The public customer side still offers no comparable non-cancellable amount.

Concentration makes the missing bridge more important

Aggregate data-centre growth can conceal who controls conversion. One distributor represented 44% of Marvell's fiscal-Q2 revenue; one direct customer represented another 16%. Four customers accounted for 72% of gross accounts receivable at 1 August. The filing does not name them in these tables.

Those percentages cannot be added into one concentration score. Revenue is a flow over the quarter; gross receivables are an end-date asset. A distributor may also represent end-customer programmes that are not visible in the aggregate label. The evidence supports a narrower conclusion: a small number of counterparties materially influence both recognised sales and collection exposure while Marvell is lengthening its own supplier commitment horizon.

That control structure can work well when demand converts. Long-lived supply reservations protect access in a constrained environment, reduce the risk that competitors receive scarce capacity first and allow Marvell to promise a credible production path for custom silicon. The same structure becomes costly if customers reschedule after Marvell has crossed a supplier cancellation threshold, if a design qualification fails or if product mix leaves reserved capacity unsuitable for the demand that survives.

Inventory is the intermediate receipt. Marvell held US$1.3606 billion of inventory at 1 August: US$1.0570 billion in work in process and US$303.6 million in finished goods. That stock was slightly below January's US$1.3880 billion, even as commitments rose. The contrast does not prove efficient conversion or hidden excess. Commitments extend years into the future; inventory records components already inside the production cycle. Future filings need to connect them before either can be used as a verdict on the other.

Liquidity is also separate. Marvell had US$3.9328 billion of cash and cash equivalents and US$5.0 billion of senior notes outstanding. It generated US$605.5 million of operating cash in the quarter. None of those figures can be netted mechanically against US$8.5189 billion to produce a “funding gap.” Purchases will arrive across fiscal years and may be accompanied by customer collections. But the expanding commitment schedule is a claim on future financial flexibility, and therefore belongs beside cash, working capital, debt, dividends and repurchases in board decisions.

Five ledgers, not one AI number

The market story becomes clearer when each state keeps its own field:

  1. Supplier commitment: US$8.5189 billion of future unconditional foundry/test/assembly purchases.
  2. Capacity funding: US$487.0 million of balance-sheet prepayments and a separately disclosed US$870 million deposit schedule associated with post-quarter agreements.
  3. Customer signal: purchase orders, design qualifications and management's unquantified bookings language, much of it cancellable or adjustable.
  4. Operating conversion: US$2.1715 billion of quarterly data-centre revenue, inventory movement, gross margin and cash collection.
  5. Concentration: 44% distributor revenue, 16% direct-customer revenue and 72% gross-receivable exposure, each on its own perimeter.

Calling the first item backlog would turn Marvell's obligation into somebody else's. Calling it debt would skip component delivery and accounting recognition. Calling the fourth item proof of the entire multi-year schedule would project one strong quarter across agreements whose customer support is not disclosed. Calling the third item empty would ignore a record quarter and a raised outlook.

The missing receipt is more practical than a prediction. It would show, by capacity class and fiscal year, how much supplier commitment is supported by non-cancellable customer demand, customer prepayments, shared cancellation economics or product programmes that have passed qualification. It would show utilisation, unused-capacity cost and conversion to shipment and cash. Until that record exists, the US$5.7621 billion increase is evidence of Marvell's decision—not proof of the demand outcome.

Sources