Earnings Call Highlights: Applied Optoelectronics (AAOI) | Q3 FY2026 — A 1.6T LTA and a $400M Q4 Guide, but Year-End Capacity Drops to 500K and Texas Power Cuts the US Share From Half to 30%

This article is for technology and industry trend analysis only and is not investment advice.

1. Executive Takeaway: Three Signals to Remember

First, the news this quarter is not growth, it is a reshuffled timeline. The headline capacity target actually went up (end-2027 from 930,000 units to more than 1 million), but the near-term number came down, the far-term number went up, and Phase 2 slipped, which pushes the revenue curve out by about six months. On the Q4 guide, full-year revenue lands around $998M–$1.033B, below the roughly $1.1B cited last quarter. Management did not repeat that figure on this call.

Second, customers are changing generations faster than AOI can stock materials. The LTA customer moved orders from 800G to 1.6T earlier than expected, which set off a chain reaction: 800G growth came in below plan, gross margin fell below the guided floor, and 1.6T could not ship in full because there are not enough DSPs. That is good long term, since 1.6T carries higher pricing and margins. For the next two quarters it is a real gap.

Third, the Made-in-America story is getting thinner. The Texas share of end-2027 capacity went from more than half last quarter to nearly 30%. Taiwan fills in the total, which is good for shipments, but it deserves a second look at a company that sells itself on having the largest AI optics capacity in the US.

2. Revenue and Financials: Barely Past Breakeven, and the GAAP Profit Needs Unpacking

The card below is on a GAAP basis. Comparisons against guidance are marked non-GAAP:

• Total revenue $269.9M, up 127.5% YoY and 40.6% QoQ, inside the $255M–$290M guide and slightly below the $272.5M midpoint.

• GAAP gross margin 28.2% (+0.2pp YoY); non-GAAP gross margin 28.7%, below the 29% guidance floor, versus 29.8% last quarter and 31.0% a year ago.

• GAAP operating loss $1.4M, an operating margin of -0.5% versus -15.3% a year ago; non-GAAP operating income $7.2M versus a $10.3M loss a year ago.

• GAAP net income $6.3M, diluted EPS $0.07, versus a $17.9M net loss and -$0.28 a year ago; non-GAAP net income $14.8M and EPS $0.17, inside the $0.11–$0.26 guide.

• Non-GAAP operating expenses $70.3M, down from 39.7% to 26% of revenue, so operating leverage is starting to show.

By segment, Datacenter was $155.6M (58% of revenue, +254% YoY, +45% QoQ) and CATV $110.9M (41%, +57% YoY, +37.6% QoQ, slightly above the guided high end). Total revenue missed the midpoint while CATV beat, so CATV covered for Datacenter this quarter.

Management is right that AOI returned to GAAP profitability, but the number needs unpacking: the GAAP operating loss was $1.4M, and the profit came from $10.1M of non-operating income, of which "other income, net" alone was $8.2M. Remove that line and pre-tax income is about $0.4M, which turns into a loss after $2.4M of income tax. The press release gives no breakdown, so that waits for the 10-Q. The real operating turn should come in Q4: at the guidance midpoint we estimate non-GAAP operating margin jumps to about 12%.

▲ AAOI Q3 FY2026 key financials: revenue, gross margin, operating margin, net income, EPS, segment revenue and next-quarter guidance, GAAP basis. (Image source: Simple Tech Trend original; data from SEC 8-K earnings release)

3. Datacenter: 800G Missed Because Customers Jumped Ahead, Not Because Demand Faded

The Datacenter mix: 200G+400G 60.2% (400G alone $93.4M, nearly doubling QoQ), 800G 30.5% ($47.5M, nearly quadrupling QoQ), 100G 7.5%, 10G/40G 1.4%.

Against last quarter's commentary there are two gaps. The Q2 call pointed to nearly 5x sequential 800G growth; the actual was about 3.7x. 100G fell from about $41M to about $12M, a drop of nearly $30M and deeper than the $20–25M flagged then, still because one customer cannot get enough 100G switches amid the memory shortage.

The gross margin miss is the same story: less 800G means a heavier 400G mix. The Q4 direction is clear, with Datacenter guided to $270M–$300M (+281% YoY at the midpoint), led by 800G. Taking out $50–70M of 1.6T and about $100M of 400G, we roughly estimate 800G needs to reach about $110M, more than doubling again.

Concentration deserves a note too. The top 10 customers were 99% of revenue, and three exceeded 10%: one CATV customer at 41% and two Datacenter customers at 28% and 23%. Converted to dollars, the CATV customer is about $110.7M, essentially the entire CATV line, and the two Datacenter customers are about 88% of Datacenter revenue.

4. The 1.6T LTA: $1–1.5B Looks Good, but What Does It Actually Buy?

The terms of the agreement signed in October are the most concrete information from the call:

• Counterparty: an existing Tier 1 hyperscale customer, 1.6T only. AOI's first 1.6T product recently qualified with this customer.

• Commitment: subject to conditions including AOI's ability to supply, the customer will buy at least 10% of its 1.6T needs from AOI from June 2026 through May 2029.

• Value: AOI expects $1–1.5B, though this can vary significantly with the customer's actual demand.

• Price paid: AOI commits to volume and agreed pricing, including reductions over the term. The customer committed no capital.

Three points to unpack. First, $1–1.5B over 36 months is about $28–42M a month, only 6–9% of the $471M monthly revenue target. Its value is visibility and confidence to expand, not carrying the growth curve. Second, working back from 10% implies this customer's three-year 1.6T spend is about $10–15B. That is our inference from the company's estimate and will move with demand. Third, there is no prepayment. The CFO acknowledged that customers' own balance sheets are under scrutiny, which makes prepayments harder than before and more likely around newer technologies like CPO and NPO. Compared with Coherent's five-year $20B single-customer LTA, this one is an order of magnitude smaller.

The real variable surfaced only in Q&A. CEO Thompson Lin said a second LTA is in the final stage and expected within a few weeks, five times the size of the first or more, covering 800G, lasers and other products. If it happens, that means roughly $5–7.5B or more. But it is a verbal statement, absent from the press release and unsigned, so for now it is a signal to verify.

The near-term 1.6T issue needs to be clear as well. Orders in hand doubled from more than $200M last quarter to about $400–450M, yet Q4 can deliver only $50–70M, versus $70–80M indicated last quarter. The limit is the DSP. The CFO said this is more about shifting priorities and demand than an absolute shortage, that a few months' notice would have secured the parts, and that he expects dramatic improvement in Q1. AOI has qualified two 1.6T DSP suppliers and is working on a third. For why this part of the supply chain is tight, see MaxLinear's 1.6T content story this quarter.

5. Capacity Timeline: Texas Is Short of Transformers and Cables, Not Equipment

Put last quarter and this quarter side by side and the changes are obvious:

Metric	Last quarter	This quarter
End-2026 800G+1.6T monthly capacity	650,000 units	More than 500,000 units
End-2027 800G+1.6T monthly capacity	930,000 units	More than 1 million units
Texas share at end-2027	More than half	Nearly 30%
$471M monthly revenue target	Mid-2027	End-2027
1.6T orders in hand	More than $200M	About $400–450M
1.6T Q4 revenue	$70–80M	$50–70M
Year-end gross margin	32–33%	Q4 guide 29.5–31% (non-GAAP)
2026 full-year revenue	About $1.1B	About $998M–$1.033B implied by guidance (STT estimate)

The total went up while Texas went down: 30% of 1 million is about 300,000 units, half of 930,000 is about 465,000, so the US capacity target is roughly a third lower. The 210,000 sq ft Phase 1 facility still starts operating in December. Phase 2 is what slipped, because obtaining electrical power is taking longer than expected. The CFO was blunt: it comes down to how transformers and cables get to the site, and "at some level" AOI is competing with all the data centers. To a degree, the optics factory's power is being taken by its own customers. He even noted that if Texas restricted data center growth, AOI might get power sooner, though that is not in the plan.

The CEO offered a simple way to model it: Datacenter at about $100M a month in Q4, $470M a month by end-2027, and roughly linear in between. On that line, mid-2027 is about $280–300M a month and full-year 2027 Datacenter revenue is around $3.7B. That is our rough estimate, not company guidance.

One more caution: these are figures for qualified production equipment capacity, and the CFO specifically warned that output depends on qualification schedules, staffing, materials and cycle time. About $95M a month in Q4 against more than 500,000 units at year-end works out to under $190 of revenue per unit of capacity; the end-2027 target is about $470. The gap has to be closed by 1.6T pricing and utilization, so do not track equipment numbers alone.

On demand, management gave no ground: 800G and 1.6T demand exceeds capacity through 2027, and by the end of next year AOI and the other major suppliers combined can meet less than 80% of it. The CFO estimated that even 1.2 million units a month would barely keep up. Asked about politics and market doubts over AI spending, management said no customers are second-guessing and that its customers are well-capitalized hyperscalers. All of this is management's view and has not been independently verified.

6. Lasers and InP: A 180,000-Wafer Order Is Not About Next Year's 800G

AOI did not name the supplier on the call, but the counterparty is public. Yunnan Germanium announced (in Chinese) that its subsidiary Yunnan Xinyao signed a long-term agreement with AOI for 4-inch N-type InP substrates: a 2027 annual contract for 180,000 wafers worth $194.4M, which is 15,000 wafers a month at about $1,080 each. Expected volume over 2027–2031 is 900,000 wafers, but volume and price from 2028 are negotiated annually, and performance depends on a Chinese Ministry of Commerce dual-use export license.

Run the numbers with Dr. Lin's parameters from the call: each good wafer yields 8,000–9,000 100mW lasers, and high-power lasers divide that by 5–6. Fifteen thousand wafers a month is about 120–135 million 100mW lasers, or 20–27 million high-power lasers. Even if 1 million 1.6T modules each used 8 lasers, that is only 8 million a month. This uses best-case yield and full loading, but the gap is wide enough to make the point: this order maps to high-power CPO and NPO lasers, not today's pluggables. That is our inference. The company said only that the need will be "much, much bigger."

The timing fits. The CFO said InP is fine through 2027 and not a current limitation, while Dr. Lin said this agreement is mainly for 2028 because the supplier is still expanding. AOI has six substrate suppliers and is also working with two in Japan and one in Germany. The whole InP chain is moving toward allocation, as AXT's backlog of more than $500M this quarter shows. It is worth noting that for lasers marketed as US-made, one important substrate source depends on a Chinese export license, so diversifying is not excess caution.

On CPO, several customers returned positive preliminary test results, and qualification is under way with four or five customers. NPO uses 100–200mW and CPO 300–350mW. Dr. Lin said customer feedback puts AOI's power efficiency ahead of all suppliers, which is a company claim. Laser gross margin exceeds 60%, but it takes 18–24 months from prepayment to volume shipment, so this will not show up in next year's first-half numbers.

As for Marvell buying Lumentum, the CFO said it highlights the value of both DSP and optics, and that being "one of the remaining independent suppliers" of optics is a good position. We covered the industry logic in our W44 weekly review.

7. CATV and Funding: One Foot Steady, the Other Burning Cash

CATV was the stabilizer: record revenue of $110.9M, the one millionth 1.8 GHz amplifier shipped, and a 3 GHz network demo and the QuantumWave 25G tunable module at SCTE TechExpo. The Q4 guide of $105M–$115M is roughly flat, amplifiers remain the vast majority, and software revenue begins next year.

Funding is the next thing to watch. Q3 capex was $501M (including $142.8M of equipment prepayments), yet cash rose from $508.8M to $621.5M because additional paid-in capital grew about $565M in the quarter, mainly from at-the-market sales (the third $600M program this year finished in early October). The Q4 guide uses about 96 million diluted shares versus 87.8 million in Q3. Accounts payable climbed from $143.9M at the start of the year to $494.3M, and inventory reached $378M. AOI signed a $200M bank loan last week, Dr. Lin said the US facility could grow to $400–500M if needed, and the company says it is leaning more on debt than before.

There was a contradiction on the call that should be pointed out. In prepared remarks, the CFO said capex would rise in Q4 and "increase further in 2027." In Q&A, Dr. Lin said next year will not be higher, stressed "we never said" that, and said it could be lower because the US sites and their payments are delayed. The two statements differ and neither was corrected, so the next call needs a clear number.

8. Management Outlook: Guidance and Language

Q4 FY2026 guidance (non-GAAP): revenue $385M–$420M (about +200% YoY and +49% QoQ at the midpoint), gross margin 29.5%–31%, non-GAAP net income $45.9M–$60.4M, EPS $0.48–$0.63 on about 96 million diluted shares. By segment, Datacenter $270M–$300M and CATV $105M–$115M. Non-GAAP operating expenses are expected at $68M–$75M per quarter.

The language falls into three tiers. Confirmed: the LTA terms, Q3 numbers, Phase 1 starting in December. Expected: the Q4 guide, DSP improving dramatically in Q1, and $471M a month by end-2027, which management described as "achievable," not a commitment. Verbal, not yet delivered: the second LTA, a net margin near 20% by end-2027, and long-term gross margin of about 40%. Watch gross margin closely: last quarter the talk was 32–33% by year-end, this quarter's guide tops out at 31%, and the LTA has price reductions built in.

Next up is the Raymond James conference on December 9, a facility tour in mid-December, and the Q4 call on February 25, 2027.

Summary

This quarter confirms last quarter's read: AOI is a supply story, not a demand story. The difference is that the constraints now extend beyond its own lines to the Texas grid, DSP allocation and the pace of customer transitions. Order visibility has improved, but delivery has moved out by six months. Three things to track next: first, whether the second LTA actually gets signed, at what size and on what funding terms; second, whether 1.6T steps up in Q1, which tests both DSP supply and the margin recovery; third, Q4 operating profit, and whether GAAP operating income can turn positive without help from non-operating income.

Demand is not in question. What AOI has to prove now is that when everyone else's data center is finished, its own factory has power.