Invalidating an optical-infrastructure thesis: make the disconfirming dashboard first

Start with the observation that could prove you wrong

An optical-infrastructure thesis is useful only if it can lose. “AI needs more bandwidth” is a direction, not a testable claim about a laser, module, switch port, or supplier. A testable thesis names a technical boundary, an operating mechanism, a time window and an observation that changes the conclusion. For example: a supplier may benefit if a qualified product reaches customer acceptance at a planned rate; the thesis weakens if the qualification is delayed, an alternate interconnect wins the relevant reach, or working-capital and pricing evidence contradict the assumed conversion.

The AAOI 2025 Form 10-K and Q2 2026 Form 10-Q are periodic reports with stated periods. Read their financial statements, MD&A and risk disclosures as period-specific evidence. The linked financial source library is a bilingual chronology of 70 actual AAOI documents; it is a reading index, not seventy independent confirmations of one claim.

1. 1Workload and topology
2. 2electrical reach limit
3. 3copper / AEC / optical choice

1. 1optical design
2. 2qualification
3. 3yield and test
4. 4shipment
5. 5customer acceptance

1. 1thesis assumption
2. 2dated primary observation
3. 3retain, narrow, or invalidate

Build four disconfirming lanes

The technical lane asks whether the assumed link boundary is still required. A shorter electrical path, an active electrical cable, a different fabric topology, or a newer optical packaging approach may serve the workload differently. A product announcement is evidence that a design exists; it is not proof of volume deployment or vendor share.

The customer lane asks whether a forecast converted into a firm order, a shipment, acceptance and revenue. The 2025 10-K says AAOI typically relies on individual purchase orders and that forecasts do not generally commit customers beyond firm orders. This makes cancellation, delay, mix change and second-source qualification direct invalidation candidates. It does not say that any particular future order will cancel.

The manufacturing lane asks whether the physical chain can deliver: laser and component availability, assembly, burn-in, test capacity, yield and returns. A capacity announcement does not establish qualified output. The accounting lane asks whether a stated revenue, receivable, inventory, margin or cash-flow movement is read with its period, currency, units and table context. A single measure can prompt a question; it cannot by itself settle a demand thesis.

Documents have different jobs

A 10-K is an annual periodic report; a 10-Q is a quarterly periodic report. They are the primary places to read the stated period's financial statements and management discussion. An 8-K is a current-event report. Item 1.01 can disclose entry into a material definitive agreement; it confirms the described agreement event, not revenue conversion. Item 2.02 concerns results of operations and financial condition and can point to an earnings release; read its status and attachment. Item 5.02 concerns directors, officers, appointments or compensation-related governance events; it is not an operating result.

The source library records a September 15, 2026 AAOI 8-K with Items 1.01 and 2.03, a financing/direct-obligation disclosure. It supports that disclosed transaction event. It must not be recast as a customer contract, product shipment or revenue confirmation. This form-and-item distinction prevents a common monitoring error: treating every new filing as equally strong proof of commercial demand.

Hypothetical dashboard and exercise

Suppose a research note assumes 800G module demand rises because a planned two-tier fabric needs 2,000 links. Make five rows: topology, customer release, qualified output, acceptance, and period financial result. For each row, specify one source, a date, a condition that would invalidate the assumption, and the action after observation. A delayed qualification may change the timing row but not the topology row; a financing 8-K may change capital-structure research but not demonstrate a link deployment. The 2,000-link number is invented for the exercise.

Run the exercise with one 10-K, one 10-Q and two 8-Ks from the library. Write what each document proves, what it cannot prove, and what next document would update the dashboard. Do not calculate a valuation or publish a growth rate without reading the relevant table and its units. The goal is a falsifiable research process, not an investment recommendation.