A contract is not a revenue event
Optical-networking commentary often compresses a long conversion chain into the word “demand.” That loses the decision points that determine whether a supplier can build, ship, invoice, and recognize a module sale. The AAOI 2025 Form 10-K says the company typically sells through individual purchase orders rather than long-term customer contracts. It also says customers' forecasts generally do not commit them to quantities beyond firm purchase orders, and that orders may be increased, reduced, cancelled or delayed. Those are issuer statements for the year ended 2025, not a general rule for every optics vendor.
The filing gives a concrete reason to separate these stages. It reports that its ten largest customers represented 96.6% of 2025 revenue, and identifies Digicomm at 53.1% and Microsoft at 28.8%. These are reported revenue shares for that fiscal year, not current run-rate, module volumes, bookings, margins, or promises about a future period. The same filing warns that order timing, order size, contractual acceptance criteria, pricing, and shipment delays can affect revenue. A forecast is therefore useful for capacity planning, but it is not an accounting result.
- 1Network design and forecast
- 2quotation / qualification
- 3firm purchase order
- 1firm purchase order
- 2component allocation
- 3assembly and test
- 4shipment
- 1shipment
- 2contractual acceptance / transfer of control
- 3invoice
- 4revenue recognition
- 1forecast changes at any earlier gate
- 2capacity or inventory exposure, not automatically revenue
The module and the document move together
At the factory, a module starts with an approved specification: host interface, lane rate, reach, optical budget, temperature class, firmware and test limits. Procurement reserves laser, driver, receiver, package and fiber inputs. Assembly and burn-in produce a serial-numbered unit; final test decides whether it meets the release criteria. A customer can require a qualification lot or acceptance condition after shipment. The commercial record needs to follow the same path: quote, purchase order, shipping terms, acceptance evidence, invoice and accounting treatment.
That is why “backlog” needs a definition before it is compared with revenue. Does it include cancellable purchase orders, forecasted releases, unaccepted shipments, or only enforceable obligations? Does a customer hold the right to reschedule? Is a module configured for one customer's firmware or label? Without those facts, dividing a stated backlog by a quarterly revenue figure creates a number that looks precise but describes no reliable conversion schedule.
A labelled hypothetical example
Assume a customer forecasts 1,000 800G modules for a quarter. The supplier orders long-lead components for 700, receives firm purchase orders for 500, builds and tests 450, ships 430, and documents acceptance for 400 before the reporting cutoff. The remaining units are not automatically revenue: some may be in test, transit, customer evaluation, or subject to a stated acceptance term. If the forecast falls to 600 after component commitments, the operational issue is excess material and capacity; it is not proof that 1,000 units were ever revenue.
This example has invented quantities. It is intentionally not an estimate for AAOI. Its purpose is to show that yield, allocation, delivery terms and acceptance can each change the relation between a planning signal and reported revenue.
Tradeoffs and a reading exercise
Short purchase-order cycles let a buyer adapt to changing port architectures and deployment schedules. They push demand uncertainty upstream to the supplier, which may carry material commitments or reserve production capacity. A longer commitment can improve planning but can make a buyer reluctant to adopt a new module, change a reach, or qualify an alternate vendor. Customer acceptance protects the buyer against nonconforming hardware but makes documentation and test traceability part of the supplier's cash-conversion process.
Open the cited filing and construct a gate log for one imaginary module program. For each gate, write the evidence that would exist, who owns it, and the failure that could stop the next gate. Then mark every statement in a news release as forecast, order, shipment, acceptance, or recognized revenue. Keep fiscal period, currency and unit beside any company number. The exercise is offline; it does not establish a current contract or forecast.
NETWORK / HYPOTHETICAL INPUTS
Port labels and useful throughput differ.
An 800 Gb/s port multiplied by the selected useful fraction. This toy fraction combines idle time and overhead; it is not a measured link or a protocol model. Tail latency, topology, retries and collective algorithms need separate measurements. It cannot predict a supplier’s sales.
SOURCES
01YOUR NOTES