Summary

  • Arista disclosed US$9.7 billion of legally binding, non-cancellable purchase commitments at 30 June 2026. US$9.4 billion, or 96.9%, is expected to arrive within twelve months; only US$0.3 billion sits beyond that window.
  • The customer ledger is not a matching hedge. US$8.4 billion of expected future revenue combines advance payments under cancellable contracts, deferred support and product revenue, and US$1.3 billion of other performance obligations. Purchase cost and gross revenue cannot be netted.
  • Current performance is strong: Q2 revenue was US$3.036 billion, GAAP operating margin was 45.4%, six-month operating cash flow was US$2.7765 billion and cash plus marketable securities was about US$13.3 billion. The issue is conversion discipline, not present distress.

The supplier clock has twelve months on it

At the end of June, Arista had US$9.7 billion of purchase commitments that were not recorded on its balance sheet. The orders cover finished goods and strategic components, including integrated circuits consigned to contract manufacturers. They are enforceable, legally binding and non-cancellable.

The timing is more revealing than the total. US$9.4 billion is scheduled for receipt within twelve months. That is 96.9% of the commitment ledger. Arista may have limited room to reschedule or alter requirements, but only if the relevant supplier agrees.

Three months earlier, the total was US$8.9 billion. US$7.6 billion sat inside twelve months and US$1.3 billion beyond. By June, total commitments had grown US$0.8 billion, but the current portion had risen US$1.8 billion while the later portion fell by US$1.0 billion.

That is a compression of time, not merely an expansion of volume. It does not disclose which component orders changed dates, whether prices moved or which product generation they support. It does show that substantially more of the procurement decision must pass through physical receipt, inventory and customer conversion before the next four quarters have elapsed.

The year-on-year change is larger. In June 2025, Arista disclosed US$3.6 billion of comparable non-cancellable commitments, with US$3.1 billion due inside a year. The present total is about 2.7 times that balance; the near-term amount is just over three times as large. Product mix may be different, so this is not a unit-demand index. It is evidence that Arista has placed much more capital and forecast responsibility on the supplier side of the AI-network cycle.

US$8.4 billion of future revenue is not the opposite entry

The tempting comparison sits nearby in the same filing. Arista expects US$8.4 billion of future revenue from contract liabilities, deferred revenue and other performance obligations. About 91% is expected over the next two years, with the remaining 9% in years three through five.

Subtracting US$8.4 billion from US$9.7 billion would produce an arithmetically correct US$1.3 billion and an economically false conclusion. The supplier number is a purchase-obligation amount. The customer number is gross expected revenue. They cover different goods, services, margins, contract rights and time periods. Neither filing supplies the cost attached to every future revenue dollar or the customer cover behind every supplier order.

The customer total also contains three states.

First, contract liabilities were US$278.4 million. Arista defines them as payments received before it has satisfied a performance obligation under a cancellable contract. Cash received is useful protection, but cancellation rights mean the balance is not identical to an irrevocable purchase promise.

Second, deferred revenue was US$6.8659 billion. It mainly represents unearned multi-year post-contract support and product deferrals under contracts with acceptance clauses. Some of the product may have moved physically while accounting recognition waits for another condition.

Third, other performance obligations were about US$1.3 billion. US$427.8 million related to unbilled multi-year support and service contracts; US$875.1 million came from binding contractual agreements primarily for future product shipments. That binding subset is firmer than an ordinary cancellable purchase order, but it is only one part of the customer-side total.

The correct comparison is therefore not balance against balance. It is state against state: what supply is irrevocably ordered, what has arrived, what is allocated to a customer, what remains in trial, what has passed acceptance, what has become revenue and what has become cash.

Acceptance is where demand becomes accounting

Arista is in a period of product introductions and expanded AI Ethernet use cases. That requires equipment to spend time with customers before revenue becomes final. The company explicitly connects customer trials and acceptance periods to the rising volatility of evaluation inventory and product deferred revenue.

Evaluation inventory reached US$616.0 million in June. It had been US$525.7 million in March and US$403.7 million at year-end. Total inventory rose from US$2.2471 billion in December to US$2.3800 billion in March and US$2.5353 billion in June.

None of those balances proves failed demand. Evaluation hardware is a normal way to qualify high-performance networking inside a real customer architecture. The risk appears if acceptance is delayed or requirements are not met. Arista says it may then have to accept a return, lose the associated revenue and write down inventory.

Deferred cost of goods sold shows another intermediate state. It was US$1.6219 billion at June, up from US$1.5465 billion in March and US$1.1970 billion in December. Arista attributes the increase to corresponding product deferred revenue.

This is economically important. Cost has not simply vanished while revenue waits. Product-related cost is held for release with the deferred revenue. The margin appears only after the relevant recognition conditions are satisfied. A shipment, a customer installation and a recognised sale are therefore not interchangeable proof points.

The one-year supply clock must pass through each of them. A component can arrive on time and still wait in raw materials. A finished switch can be assigned to an evaluation. An evaluation can succeed while revenue remains deferred under an acceptance clause. Revenue can be recognised while extended payment terms delay cash. Each transition is controlled by a different piece of evidence.

Customer flexibility is not evenly distributed

Arista says sales are primarily based on purchase orders, and some customers can cancel, delay, reduce or otherwise modify their commitments with little or no notice. That sentence establishes an asymmetry: supplier orders can be legally firm while portions of expected customer demand remain forecast-dependent.

It does not mean every customer contract is cancellable. The filing separately identifies US$875.1 million of binding agreements for future shipments, advance payments, support contracts and acceptance-based deferrals. Treating one general risk statement as the legal description of all US$8.4 billion would be as misleading as treating the whole amount as guaranteed cash.

Concentration raises the stakes. Two end customers supplied 16% and 26% of 2025 revenue. The checked record does not name them or disclose their Q2 2026 shares. It is not possible to assign current commitments, evaluation stock or deferred revenue to a particular hyperscaler.

What is visible is control. Large buyers can alter capital budgets, network architecture, order timing and efficiency plans. Some large-customer arrangements include extended payment terms. Their decisions can change both the speed of conversion and the amount of working capital Arista carries.

Arista has the opposite incentive. Tight memory and silicon supply and rapid AI-network deployment reward a vendor that reserves components early. A smaller commitment may preserve flexibility but lengthen lead times and sacrifice customer schedules. A larger firm order can protect delivery and market share, but transfers forecast error from suppliers to Arista.

The current numbers reject a crisis story

This commitment ledger has grown during a period of exceptional operating performance. Q2 revenue reached US$3.036 billion, up 37.7% year on year and 12.1% from Q1. GAAP operating income was US$1.3780 billion and operating margin was 45.4%.

GAAP gross profit was US$1.9103 billion. Gross margin was 62.9%, down from 65.2% a year earlier. That decline deserves monitoring, particularly because Arista says supply inflation, scarce materials and tariffs may pressure product economics. The filing does not isolate the effect of purchase commitments, and the margin remains very high.

Cash provides another boundary. Operating activities produced US$2.7765 billion during the first half, compared with US$1.8418 billion a year earlier. Cash, cash equivalents and marketable securities totalled about US$13.3 billion at June. Management says existing liquidity and operating cash are sufficient for working capital and growth for at least the next twelve months and thereafter for the foreseeable future.

The US$9.7 billion is therefore not evidence of an immediate funding gap, still less insolvency. It is an off-balance-sheet purchase obligation that can become inventory, accounts payable and cash use as suppliers perform. The size matters because execution can move the risk into those ledgers quickly.

Strong present cash also makes the decision more interesting. Arista can afford to reserve supply. The question is whether doing so earns a better delivery position and future margin, or merely converts liquid optionality into hardware before customer acceptance is secure.

Five receipts would make the build measurable

The first receipt is the supplier ledger. Investors need the total commitment, its twelve-month split, deposits, supplier delivery and any liability for quantities beyond forecast or products becoming obsolete.

The second is inventory state. Raw materials, ordinary finished goods and evaluation inventory should not be blended into one demand claim. Each has a different next decision and a different write-down path.

The third is customer state. Advance payments under cancellable contracts, support obligations, acceptance-clause product deferrals, binding future shipments and ordinary purchase orders need separate treatment. The label “future revenue” is a useful total but a poor description of rights.

The fourth is recognition. Deferred cost and deferred product revenue should convert together through acceptance. Returns, extended trials or a growing mismatch between the two would reveal pressure earlier than reported revenue alone.

The fifth is economics. Product gross margin, operating cash, customer collections and inventory provisions show whether reserved supply is creating profitable throughput. Revenue growth without these receipts cannot establish the quality of conversion.

Arista has made a rational strategic choice in a scarce market: place orders early enough to defend delivery. It has also accepted a measurable asymmetry. Suppliers hold a firm, near-term claim; customers occupy several contractual and operational states. US$9.4 billion inside one year is not a forecast of failure. It is the deadline by which confidence in AI-network demand must become accepted equipment, margin and cash.

Sources