Summary

  • Arm no longer reports quarterly RPO in its shareholder-letter scorecard, saying the measure became less relevant as the business extended into production silicon. Its fiscal-Q1 financial statements still disclose US$2.1226 billion.
  • Approximately 24% is expected inside twelve months, 23% in months 13–24 and 53% later. The longest bucket can include obligations placed there because a customer's future action controls timing; it is not a dated delivery schedule.
  • RPO excludes potential future royalties, while royalties supplied US$715 million of the quarter's US$1.289 billion revenue. RPO therefore remains useful as a licence-contract ledger, but it cannot stand for Arm's whole revenue opportunity.

The row vanished; the obligation did not

Arm's fiscal-Q1 shareholder letter contains a small but consequential change to the investor scorecard. Annualised contract value remains. Quarterly RPO and the counts of extant Arm Total Access and Arm Flexible Access licences do not. The company says those measures have become less relevant to growth as it extends its business into production silicon.

That is a judgement about presentation and business mix, not an accounting extinguishment. The unaudited financial statements released alongside the letter still contain the RPO note required to explain unsatisfied or partly unsatisfied performance obligations. At 30 June 2026, the amount was US$2.1226 billion.

Both records can be true. A measure can become less representative of the way management wants investors to understand the future while remaining necessary to understand contracts already signed. The mistake is to choose between them. The scorecard says Arm's growth model is widening. The note says the older and continuing licence model still carries a material stock of work, billing and customer choices that have not finished converting into revenue.

The change is especially informative because the last shareholder letter that promoted RPO used it to describe “unearned revenue and amounts to be invoiced and recognised in future periods.” At March, the rounded headline was US$2.071 billion, down 7% year on year, which management linked to improved timing of revenue conversion. Three months later, the financial-statement balance was US$2.1226 billion, but the shareholder-letter row was gone.

This sequence is not evidence of concealment. Arm disclosed the policy change and kept the financial number public. It is evidence that metric governance matters: withdrawing a row can change which questions investors ask even when the underlying data survives elsewhere.

The 53% bucket is partly a timing convention

Arm expects to recognise approximately 24% of June RPO over the next twelve months, 23% over the subsequent 13–24 months and the remaining 53% thereafter. Those shares create a useful duration map, but not a project schedule.

The note supplies the reason. In some arrangements, satisfaction of the performance obligation depends on a customer action. When Arm lacks enough information to know when that action will occur, it places the transaction price beyond two years unless the contract or option expires earlier. The longest bucket therefore combines at least two possible states: work genuinely expected on a long timetable, and work whose timetable is not sufficiently observable because the customer still controls a step.

The filing does not quantify those states separately. It would be wrong to say all US$1.125 billion implied by the 53% share is customer-delayed. It would be equally wrong to treat that amount as delivery already scheduled for fiscal 2029 and later. The public evidence supports a mixed time class, not an appointment book.

Arm's licence design explains why the issue arises. Some customers receive rights to access a library of current and future IP over a contract period. These are stand-ready obligations. The customer controls when underlying IP is taken, and using the library in one period does not diminish the remaining obligation. Consideration is recognised ratably over the contract term. Other arrangements contain specified releases, version extensions, support or professional services with different recognition points.

A single RPO total therefore carries several kinds of waiting: Arm may still need to release IP, remain ready, provide support or perform services; a customer may still need to select or trigger delivery; time may simply need to pass. The recognition bucket records the accounting outcome of those conditions. It does not reveal the engineering state of each design.

The rounded bridge moved outward

At 31 March, Arm reported US$2.0714 billion of RPO and expected approximately 28% inside twelve months, 21% in months 13–24 and 51% later. By June, total RPO was US$51.2 million higher, but its nearest share was four percentage points lower and the longest share two points higher.

Applying the published rounded shares gives an indicative movement. The near-term amount falls from about US$580.0 million to US$509.4 million. The middle window rises from roughly US$435.0 million to US$488.2 million. The later bucket rises from about US$1.056 billion to US$1.125 billion. On that basis, almost US$70.6 million leaves the nearest window even as the total stock grows.

These are estimates, not disclosed subledger balances. Each percentage is approximate, so the arithmetic cannot support dollar-level precision or a claim that named contracts moved. It can support one disciplined conclusion: growth in the total did not increase the rounded next-twelve-month RPO. More contract stock and more near-term recognised revenue are not the same event.

The movement need not be adverse. Arm could have signed longer arrangements, recognised near-term obligations, added customer-controlled options or changed the mix of stand-ready work. The nearest share can fall because conversion is working, because replacement bookings are longer, or because timing remains open. Without a gross additions-and-recognition bridge by time class, demand and duration cannot be separated.

RPO omits the largest current revenue stream

Arm explicitly excludes potential future royalty receipts from RPO. That boundary is central to reading the number after the company moved into production silicon.

Fiscal-Q1 revenue was US$1.289 billion. Licence and other revenue supplied US$574 million; royalty revenue supplied US$715 million. The larger current revenue stream therefore sits outside potential future RPO until the relevant customer products are sold or used and the royalty exception permits recognition. Arm estimates the accrual using customer reports, shipment trends, product mix and industry evidence, then adjusts when later information arrives.

The financial statements also report US$735.8 million of revenue from performance obligations satisfied in prior periods, primarily royalties earned in the current quarter. That phrase can look backward while the economics are current: Arm transferred the underlying IP earlier, and the customer's subsequent chip shipment now creates royalty revenue.

RPO consequently answers a narrower question. It records fixed transaction price allocated to remaining contract performance. It does not include the future economic yield of chips that licensees have not yet shipped. A declining, rising or retired RPO row cannot by itself describe the royalty engine.

Annualised contract value is different again. Arm reported ACV of US$1.732 billion, up 13% year on year. ACV annualises committed fees for active licence agreements and also excludes future royalties. It smooths contract duration under Arm's stated methods and is not GAAP revenue. Keeping ACV while dropping RPO tells readers which licence signal management now prefers, but it does not make the two numbers interchangeable.

Cash and performance have their own bridge

RPO is not the same as cash received. Arm's contract liabilities fell from US$1.046 billion at March to US$939 million at June. During the quarter, customer prepayments and advance billing added US$110 million. Revenue recognition removed US$138 million from the opening balance and another US$79 million from amounts billed or paid during the period.

That bridge shows why a single backlog-style number cannot carry the cash story. A customer can pay before Arm performs, Arm can perform before it has an unconditional right to invoice, and a customer action can leave timing open even when consideration is fixed. Contract liabilities, contract assets, receivables and RPO meet at the contract but occupy different moments.

Production silicon adds more clocks. A physical processor business introduces manufacturing partners, capacity, inventory, purchase commitments, delivery acceptance and product gross profit. Some customer demand for silicon may not fit the licence-oriented operating metrics that helped describe Arm's earlier model. That supports management's relevance argument. It also increases the need for replacement disclosure rather than less disclosure: readers need to see which commitments belong to licences, which to royalties, which to silicon and which remain non-binding demand signals.

The US$2.1226 billion RPO should not be added to management's production-silicon demand statements. The public record does not show their overlap. It should not be subtracted from ACV, netted against contract liabilities or treated as revenue lost because its scorecard row disappeared. Each measure has its own inclusion rule.

Metric retirement creates a replacement obligation

Companies rightly retire operating metrics when the business changes. A metric maintained by habit can become more misleading than one withdrawn with an explanation. The test is whether the new scorecard preserves the decision-useful information the old measure carried.

RPO contributed three things. It showed a fixed contracted stock, offered a coarse recognition horizon and created a bridge to contract balances. Production silicon makes it less comprehensive, but not less real. ACV covers active committed licence fees but not the same accounting perimeter. Revenue guidance covers a near period but not contract duration. A demand headline can include intentions that are not performance obligations.

The best replacement would separate business models. For licences, disclose ACV, RPO, recognition windows and customer-dependent timing. For royalties, disclose architecture and end-market conversion without pretending future variable receipts are contracted. For silicon, disclose binding orders, capacity, deposits, cancellation rights, inventory and accepted delivery. Then a reader can see why one metric lost relevance without losing the contract information it once made easy to find.

Arm's first post-retirement quarter leaves the evidence available, but more dispersed. The market can still reconstruct the contract ledger. The harder question is whether future quarters will make the widening business easier to reconcile, or merely remove another row from the front page.

Sources