←2026-08-30 · VM Genius
Analysis
AXTI doesn’t make GPUs, doesn’t make switch chips, and doesn’t produce 800G or 1.6T optical modules; optical communication chips aren’t even its main business. Yet in this AI optical communications rally, AXTI has been showing up more and more often on investors’ radar.
There’s only one truth: InP, indium phosphide.
More precisely, AXT sells InP wafer substrates. That sounds obscure, but the point actually worth noting is exactly here: while the market is chasing optical modules, silicon photonics, and CPO, InP wafers at the very upstream may be turning from a niche material into the most easily squeezed link in the entire AI optical interconnect expansion chain.
So when money trades AXTI now, what it’s really betting on isn’t “AI,” but:
If 1.6T, silicon photonics, and CPO all ramp together, will upstream InP be enough?
That’s the core.
For the past two years, the core bottleneck in AI infrastructure has been insufficient compute, so the entire market has been scrambling for GPUs. But once GPU counts scale from a few thousand units to tens of thousands—and then past a hundred thousand—the problem starts to change. You can have as many GPUs as you want; if data can’t move between those GPUs fast enough, cluster efficiency still won’t improve.
So AI infrastructure is now gradually shifting from a “compute race” to a “connectivity race.”
Network speeds are also upgrading across the board: 100G → 400G → 800G → 1.6T → and eventually 3.2T. The higher the speed, the more obvious the limitations of traditional copper interconnects become in distance, power consumption, and signal integrity. As a result, more and more data transmission must shift from “electrical” to “optical.” This is why 800G, 1.6T, silicon photonics, and CPO are becoming the new key themes in the AI supply chain. But the market often focuses only on optical modules and overlooks a more upstream question:
Where does the light come from?
One answer is InP lasers. And before you can produce InP lasers, you first need high-quality InP wafers.
AXT sits exactly at this layer.
The entire supply chain can be simplified as:
Raw materials
→ InP single crystal
→ InP wafer substrate
→ Epitaxy
→ DFB, EML, lasers, photodetectors
→ Optical engines, silicon photonics PIC
→ 800G, 1.6T optical modules
→ AI data centers.
AXT sells the most upstream InP wafer substrate. In other words, if downstream players like Lumentum and Coherent want to expand laser and optical component capacity, the precondition is that the upstream first has enough wafers.
This is what makes AXTI genuinely interesting. It does not directly feed on AI end-demand; it feeds on the upstream materials demand created by capacity expansion across the whole supply chain:
More GPUs, more data traffic;
More data, more optical connections;
More optical connections, more lasers;
More lasers, more InP wafer consumption.
So AXTI’s real investment logic is actually very traditional: upstream materials supply-demand mismatch.
When many people research AXTI, they ask: if the future is Silicon Photonics, why is InP still needed? The issue is that silicon photonics does not mean “everything uses silicon.” Silicon is very good at waveguides, modulators, and large-scale photonic integration, but it has one inherent weakness: silicon is not good at emitting light efficiently. So many silicon photonics approaches still require external laser sources, and those laser sources are largely built on the InP material system. The more likely combination in the future is not “silicon replaces InP,” but: InP handles light emission, and silicon handles light processing. So rising silicon photonics penetration does not necessarily suppress InP demand. In some architectures, it may actually require more high-performance lasers.
The same applies to CPO.
CPO solves the problem of high-speed electrical signals traveling too far on a PCB: it places the optical engine next to the switch chip and converts electrical signals into light as quickly as possible. But the light source does not simply disappear. Switch ASIC + Silicon Photonics + Laser can still become a core architecture.
And InP remains one of the most important material systems for high-performance lasers.
One of the biggest future things to watch for AXTI is 6-inch InP wafers. Traditional InP is mostly concentrated at 2-inch, 3-inch, and 4-inch, and once 6-inch matures, the economics improve significantly. The area of a 6-inch wafer is roughly 2.25× that of a 4-inch wafer. In theory, this means each wafer can produce more devices, automation can be higher, and unit costs have a chance to come down.
But the hard part is: InP is not silicon. Silicon can already be reliably made at 12-inch, but InP is a compound semiconductor. The larger the size, the harder it is to control crystal defects, dislocations, uniformity, and yield. So 6-inch is not simply about making the wafer a little bigger. It is actually a technical and manufacturing barrier. Whoever first makes 6-inch stable could gain stronger cost and supply advantages in the next wave of optical-communications capacity expansion. This is also why AXT's 6-inch progress deserves to be watched separately.
When I look at upstream materials companies, what I care about most has never been hearing “AI demand is strong” on a conference call. There is only one signal that actually matters:
Whether customers are willing to put up money to lock in future capacity ahead of time. This is also what makes AXTI most worth watching lately. Coherent has already begun working with AXT on 6-inch InP and is providing funding to support the related capacity expansion. Lumentum has also signed a long-term InP wafer supply and capacity reservation agreement.
This signal is very important.
Because once a customer is willing to lock in capacity several years ahead, what they are usually worried about is no longer price, but: whether there will be supply in the future.
This means the industry logic is starting to shift from “demand is strong” to “supply may tighten.” And the phase where upstream materials stocks are most likely to produce the biggest upside is often exactly this moment. Once customers move from negotiating price to fighting over capacity, pricing power may begin to shift toward the materials side.
InP isn’t an ordinary silicon wafer. Its market itself isn’t especially large, but the process is difficult, customer qualification cycles are long, and high-quality suppliers have always been concentrated in a handful of companies—just these three: Sumitomo Electric 40.2%, Beijing Tongmei (AXT) 38.3%, JX Metals 13.1%. Combined, the three account for 91.6%.
This creates a very typical structure:
The market used to be small, so nobody expanded capacity aggressively;
Now AI has suddenly pulled up demand for optical communications;
Downstream players are starting to lock up capacity early;
But upstream expansion isn’t that fast.
That’s where the problem comes in.
What the materials market fears most isn’t big demand; it’s:
Demand suddenly accelerates, but supply can’t keep up. Because this kind of capacity can’t be brought online immediately just by throwing money at a factory. Crystal growth, slicing, polishing, defect control, customer qualification—every step takes time. So AXTI’s greatest elasticity may not come from how large the InP market itself is, but from: very low supply elasticity. In a market that wasn’t originally that large, as soon as a supply-demand gap appears, profit elasticity can be just as extraordinary.
If you strip away all the storylines, there are really only five things to watch with AXTI from here.
First, whether InP revenue can continue to accelerate.
Second, whether capacity expansion stays on track.
Third, the yield, customer qualification, and production ramp progress on 6-inch InP.
Fourth, whether major customers like Lumentum and Coherent keep adding purchases, prepayments, and long-term locked-in orders.
Fifth, whether China supply-chain and export-license risk has changed (who is AXT's upstream supplier? Go look up Beijing Tongmei first—then you'll get it; there are a lot of stories there too).
The first four determine the upside; the last one determines the tail risk.
What’s actually worth studying about AXTI isn’t whether it’s an AI concept stock. The real question is: after 800G, 1.6T, silicon photonics, and CPO all expand together, how much InP will the entire industry need? And: can the existing group of suppliers bring high-quality capacity online fast enough? If demand just grows at a normal pace, AXTI is simply an ordinary upstream materials company. But if AI data centers really push high-speed optical interconnect into a new order of magnitude, and InP supply doesn’t expand fast enough, then the market is no longer trading an “AI concept.”
It is trading a very classic upstream cycle: demand suddenly accelerates, supply expansion lags, customers lock in supply ahead of time, and pricing power shifts toward materials suppliers. That is the real reason capital has recently begun to take a fresh look at AXTI. It’s not because the company suddenly learned how to tell an AI story. It’s because the supply chain has already started telling the market with real money:
InP, as an upstream material, may not be as ample as everyone assumed.
Q2 2026 revenue jumped 77% quarter-over-quarter to $47.6 million, InP sales set a fresh record, and gross margin climbed to 44.9%. Operating leverage has fully kicked in: Non-GAAP net income was $11.9 million, versus a Non-GAAP loss of $0.15 per share in the year-ago period. Q2 revenue of $47.6 million was a record high, InP revenue also set a record, GAAP gross margin reached 44.9%, and GAAP net income was $11.1 million.
Valuation is definitely aggressive right now. The FY1 Non-GAAP P/E of 77.71x is not low, but it falls to 29.28x in FY2 and to 19.81x in FY3. As long as InP demand from AI infrastructure remains strong and AXT’s capacity comes online smoothly, the high valuation has a chance to be digested by results.