Stress Testing $AAOI
Big projections...could they pull it off?
EDIT: I wrote this prior to the Citrini article. I have not personally read through the report, but I have been told of the bits from the AAOI piece. My views that I posted here remain unchanged. Happy to discuss any specific pieces and how I view them in the comments on this post!
To build conviction in my positions, I need to stress test the thesis. That means understanding the bear cases clearly, identifying the strongest ones, and figuring out what would actually prove them right or wrong.
That is what this report is. I am going to walk through the major risks I see in AOI, explain why they matter, and explain how I am thinking about them as an investor.
I do this with every company I own. Sometimes it builds my conviction, sometimes it breaks my conviction. But it always gives me a strategy and a game plan.
A few quick points up front. I am working from public information. I do not have visibility behind the curtain. It is always possible that there are issues public investors cannot yet see. That is true of any company. I also am not interested in pretending certainty exists where it does not. I do not know whether AAOI goes much higher from here or hits a wall. What I can do is identify the key risks, understand what matters most, and update my view as the evidence changes.
My major buys were in the $30s, $60s, and low $90s. That low-$90s range is where I liked the risk-reward for my portfolio and existing position size. That does not mean the stock cannot keep running. It can. It just means I am comfortable with my exposure and I want to keep forcing myself to challenge the thesis and that’s where I feel comfortable building up my position with the current information we have.
None of this is a recommendation to do anything. I own shares in $AAOI.
Management credibility
This is the most common bear case, and it deserves to be taken seriously.
AAOI has real scar tissue. The 2018–2019 period is the reason many investors discount management today. Revenue fell from $382.3 million in 2017 to $190.9 million in 2019. The company said the decline was driven primarily by lower 40G and 100G demand as a major customer changed the way it architected its network. The company also disclosed that in 2018 it experienced reduced 100G shipments because of concerns about product failures. That is the core of what people still refer to as the Amazon debacle.
That history matters for sure. But I do not think it should be treated as a clean apples-to-apples comparison with today.
Back then, AAOI was selling more standardized 40G and 100G transceivers into a different supply environment. Today, the setup is centered around 800G, 1.6T, and control of internal Indium Phosphide (InP) laser capacity during a period when hyperscalers are actively trying to secure optical supply. This backdrop today (and most likely into 2027-2028) is one of extreme supply:demand imbalance. That is really crucial.
The Amazon relationship itself shows that the current cycle looks different. AOI entered into a transaction agreement with Amazon and issued a warrant tied to aggregate Amazon purchases of up to $4 billion over the next 10 years. That is not a flat $4 billion blank check, but it is a massive strategic framework that validates AAOI has become relevant to Amazon again in this cycle. Also, the recent announcement of a contracted $200 million volume order for 1.6T transceivers from a major hyperscaler reinforces that demand is clearly converting from interest into contracted financial backlog.
However, we must also look at the sheer cost of management’s ambition. To fund their massive factory expansion, management heavily diluted shareholders in 2025. They executed four separate ATM offerings, selling 23.7 million shares to raise ~$500million That dilution is a real cost that investors have had to absorb. Obviously if you got in recently, you didn’t feel that impact. But they do have a current ATM. This is something that can be looked at from a few different perspectives. On one hand, yes dilution hurts. But if that dilution is necessary to fund massive growth, is it worth it? That’s what individual investors need to weigh. The growth could easily outpace the dilution if they succeed.
I also think it is useful to distinguish between the voices inside management. CEO Thompson Lin often speaks with very high conviction, noting on the Q4 call that their $1 billion revenue guide is entirely gated by capacity: “Let me say the demand is much, much bigger than $1 billion. That’s the number we feel comfortable... minimum 99% confident we can deliver”. CFO Stefan Murry, by contrast, sounds much more operational, focusing on line replication and supply-chain security.
This suggests the issue is not “management is making it up,” but rather a gap between founder-style narrative and the actual industrial cadence of revenue conversion. Are announced programs converting into qualified shipments? Are delays isolated or recurring? If those boxes get checked, management credibility improves. Rather than just assuming they are 100% accurate or chalking up what they are saying to over exaggeration, let’s actually follow the progress they are making and not get too ahead of ourselves.
Customer concentration
AAOI is still a highly concentrated business.
In the 2025 annual report, the company disclosed that its top ten customers represented 96.6% of revenue, with Digicomm at 53.1% and Microsoft at 28.8%. The company historically depends on recurring purchase orders rather than long-term contracts. That means one customer changing pace, shifting architecture, or pausing qualifications can still matter a lot.
But I also think this risk needs to be looked at dynamically, not statically. The 800G and 1.6T cycles are actively broadening this base.
On the Q4 call, CEO Thompson Lin explicitly stated: “I would say we will have 3 hyperscale data center customer to be more than 10% or much more than 10% for the whole year”. He also noted that a brand new hyperscale customer began discussions about qualifying 800G and 1.6T products just within the last few weeks. Furthermore, the recent $200 million 1.6T order is expected to bring a long-time customer back into “10%+ status”.
If AAOI exits 2026 with three large hyperscale customers at meaningful scale, the concentration profile is still intense, but it is structurally much healthier.
Qualification, interoperability, and timing
This is one of the most important risks in the whole story, because this is where investor expectations can easily outrun reality.
With AAOI, customer demand, product qualification, interoperability, and recognized revenue are not the same thing. We saw this clearly in 800G. On the Q4 earnings call, management noted that 800G revenue came in below their $4M-$8M expectation. That was not a collapse in demand but rather a timing issue due to ongoing firmware optimization.
The counter here relies on Murry’s specific clarification of why this delay happened. Hyperscaler network application scopes are broadening, requiring AOI to interoperate with newly added switch platforms. Regarding the firmware fix, Murry noted: “The firmware issue itself, I wouldn’t even call it an issue... the ramp starting in Q2 2026 is still on schedule, and that’s really probably the most important point for investors to understand”.
We see the exact same dynamic in the recent 1.6T announcement. The $200 million order is officially booked, but shipments will only begin in early Q3 2026 and complete in Q4 following product qualifications. In a stock like this, timing matters almost as much as demand. If delays remain isolated and understandable (like standard interoperability tweaks) rather than recurring, the thesis is intact. But again, we NEED to track these developments and stay on top of them.
Capacity ramp and execution
This is where the story becomes industrial instead of conceptual.
AAOI is trying to build the machine that can serve this massive demand. Management exited 2025 with roughly 90,000 units per month of 800G capacity and plans to reach over 500,000 combined 800G and 1.6T units per month by the end of 2026.
The obvious bear case is that this is simply too aggressive. Can they really install equipment, expand facilities, and bring all of this online without operational mistakes?
AAOI’s answer is not “we will hire a huge manual workforce and hope for the best.” Their answer is a highly proprietary, 9-year automated manufacturing platform. This automation has decreased labor hours by over 85%, reduced manufacturing cycle times by more than 35%, and achieved a Defective Parts Per Million (DPPM) rate of less than 50 for highly complex 800G modules.
Because of this, Murry explained that scaling capacity simply requires ordering equipment and “replicating production lines” they have already perfected. That makes the 5x ramp drastically more predictable than a traditional labor-heavy assembly model.
It also enables their U.S. manufacturing moat. Murry noted AOI can produce in Texas for only about a 10% to 15% cost premium relative to Asia because of its automation. Hyperscalers view securing their AI supply chain as an “existential opportunity or crisis,” meaning they are terrified of geopolitical disruptions. They are more than happy to pay a slight 15% premium for secure, U.S.-made capacity.
The CATV buffer
This is one of the biggest missing pieces. The market talks about AOI like it is only an AI optics story. And while that is obviously where most of the growth will lie and what investors are interested in, it’s important to understand the foundation as well.
In 2025, CATV revenue nearly tripled year-over-year to $245.1 million and represented a massive 53.8% of total revenue. On the Q4 call, management stated they believe it is feasible that CATV could generate nearly $300 million annually if current momentum continues.
Beyond just the hardware cycle, there is a new, highly bullish angle to CATV. High-margin software revenue. Murry noted that this cycle of 1.8 GHz nodes are the first generation of “smart amplifiers”. AOI’s QuantumLink software uses machine learning to analyze network telemetry data in real-time, allowing MSOs to dispatch crews and fix problems before they cause a customer outage. Murry stated explicitly that they anticipate generating standalone revenue from these software solutions this year.
This provides a massive financial floor. If 800G slips by a month, the company does not suddenly become a near-death equity. The CATV business is churning out cash and developing new SaaS-like revenue streams in the background.
Overbuild and pricing pressure
The short argument says the supply response in optics is fast, broad, and structurally unconstrained. They argue that when 15-20 global assemblers add capacity, the aggregate supply increase is massive, which should crush pricing by 2026-2028.
There is evidence of this already. Tower Semi is aggressively expanding silicon photonics capacity, and other standard CMOS foundries are entering the fray.
But here’s the thing. We are currently in a severely supply-constrained environment. When asked about the historical standard of 15% to 20% annual price reductions, Murry explicitly stated: “historical trends in terms of price reductions are probably not really on the table at this point”. He noted current pricing holds firm at roughly $0.50 per gigabit - meaning an 800G transceiver is selling for roughly $400, and 1.6T is between $700 and $800.
The recent NVIDIA agreements (investing $2 billion into both Lumentum and Coherent) support this broader view. Hyperscalers are signaling that advanced optical capacity matters deeply right now.
To be clear, I believe overbuild is a real risk, but in my opinion it is an ~18 month issue, not a next-quarter issue. But nonetheless, must track it.
The laser bottleneck (And architecture shifts)
I think this is a good counter to the assembly is easy thesis.
AOI’s argument is that final module assembly is not the true bottleneck. The absolute bottleneck is upstream InP laser capacity.
Even as the industry shifts to Silicon Photonics (SiPh) and Co-Packaged Optics (CPO) where modulation shifts to the silicon chip, the system still fundamentally requires a continuous high-power “pump laser” to feed it light.
Murry explained the physical math behind why this is breaking the industry: “those optical devices are physically larger in size, significantly larger in size than the earlier generations of devices... As the die sizes get multiples in terms of size, well, that has implications on the amount of capacity that needs to be brought online”.
Because these specific lasers are so much larger, they eat up manufacturing wafer real estate at a massive multiple of previous rates. AOI is highly differentiated because they are vertically integrated down to the atomic level, growing their own raw semiconductor crystals using proprietary MBE and MOCVD processes in Texas. To meet this specific bottleneck, AOI already has line of sight on the equipment to “more than triple” their in-house laser manufacturing in Texas by mid-2027.
What I am really interested to see is how the future architectures shift where value accrues. But again I think this will be more in focus in 2027 and onward. Something to track but at this time we don’t have concrete visibility. We also know that pluggable transceivers are here to stay for many years.
Profitability inflection
Aside from the big revenue projections, there is a lot of promise in profitability.
On the Q4 call, AOI explicitly guided for over $1 billion in 2026 revenue and non-GAAP operating profit of over $120 million. They also expect to achieve sustainable non-GAAP profitability beginning in Q2 2026.
More importantly, the gross margin expansion targets are incredibly bullish. Lin stated: “by Q2 next year , we believe the overall gross margin will be 35%-38% just for transceiver. By end of next year , we believe we can achieve more than 40% gross margin for all the transceiver by Q4 2027”.
If they get anywhere close to that, the company changes materially. It means less dependence on external capital dilution, more internal cash generation, and a complete rerating of how investors view their operating leverage.
Will they get there? Again, no one knows yet. So we’ll track it. And even if they get 70% of the way there, that is still significant.
Valuation versus duration
Even if AAOI executes perfectly over the next year or two, the market can still get ahead of itself if it starts treating current shortage economics as if they will last indefinitely.
That is why I separate the near-term and the medium-term. The near-term question is whether AOI converts demand into real shipments, ramps capacity, and moves into profitability. The medium-term question is whether the industry remains attractive once more capacity arrives and architecture shifts change where value accrues.
This is why it is important to stay dynamic.
To sum it up
I am bullish AAOI and have a significant position, but I do not think that means looking past the risks.
The biggest near-term risks for me are customer concentration, qualification timing, and execution on the massive 5x capacity ramp. The biggest medium-term risk is overbuild and pricing pressure.
What keeps me constructive is that current evidence suggests the opportunity is real, demand is still outrunning supply, hyperscalers are acting like secure optical capacity is an “existential” priority, and AOI appears to have massive structural differentiation. The internal MBE/MOCVD laser fab matters. The physical wafer real estate math matters to me. The proprietary 85%+ labor-reducing automation matters. The $300 million CATV software-revenue buffer matters. The profitability inflection to 40% gross margins matters. And the fact that customers are already placing contracted $200 million 1.6T orders matters.
I will keep pressure-testing the thesis, watching the checkpoints, and updating as the evidence changes. That is how I am approaching AAOI.
Again, none of this is a recommendation to do anything. You have to understand that everyone will view investments differently and it’s important to take all the information that is presented to you and make your own internal decisions.
Great and balanced review.