I'll be honest with you. I almost missed this one.
For months I had been circling the semiconductor space, looking for something that wasn't Nvidia, wasn't TSMC, wasn't one of the usual suspects that every newsletter and YouTube channel had already beaten to death. I wanted something with real exposure to the AI buildout — but upstream. Something that the big names couldn't exist without, yet most people had never heard of.
That's when I found AXT Inc. And when I finally pulled the trigger and bought at $50 a share, I felt that rare combination of excitement and quiet confidence that I've only felt a handful of times in my investing life.
Here's my full story — how I found it, why I bought it, and what I think happens next.
Looking for the Picks and Shovels Play in AI
Everyone knows the gold rush analogy. During the California Gold Rush, the people who consistently made money weren't necessarily the miners — it was the people selling the picks, shovels, and jeans. The same logic applies to AI infrastructure.
I was specifically looking for a stock from the semiconductor space that was positioned at the very beginning of the AI data-center supply chain. Not the GPUs. Not the networking chips. Not even the optical transceiver companies. I wanted something that all of them needed but that very few investors were paying attention to.
I started mapping the supply chain backwards. AI data centers need optical interconnects to move data at the speed of light between chips and racks. Those interconnects are built on indium phosphide (InP) chips. Those chips are fabricated on InP wafers. And those wafers? There are only a handful of companies in the world that can make them at scale.
AXT Inc. is one of them.
What AXT Actually Does (And Why It Matters)
Based in Fremont, California, AXT makes compound semiconductor wafer substrates — specifically indium phosphide (InP), gallium arsenide (GaAs), and germanium (Ge). If those terms sound obscure, think of them this way: these are the raw silicon equivalents for a class of devices that ordinary silicon simply cannot power.
InP substrates, in particular, are the foundation for every high-speed optical transceiver in AI data centers. When hyperscalers like Google, Microsoft, and Amazon build the next generation of AI clusters — with hundreds of thousands of GPUs talking to each other at blinding speed — they need optical interconnects. Those interconnects need InP. And AXT makes the InP wafers that go into those devices.
As per my study, the global InP wafer market is projected to grow from approximately $200 million in 2025 to around $628 million by 2035 — a compound annual growth rate of roughly 11.5%. That's not a projection built on hype. It's built on semiconductor roadmaps that hyperscalers are already committing billions of dollars to.
What struck me most was this: AXT isn't just riding the AI wave. It is, quite literally, one of the first items in the AI value chain.
The Moment That Convinced Me
I was reading through the Q3 2025 earnings transcript when CEO Morris Young said something that stopped me mid-scroll:
"Indium phosphide is one of the first items in the value chain for the AI data center buildout. Every optical transceiver that moves data in an AI cluster starts with an InP substrate. We are not a downstream beneficiary of AI — we are an upstream enabler."
That framing — upstream enabler rather than downstream beneficiary — was the mental model shift I needed. Most AI stocks are priced as if the buildout is already complete. AXT was priced as if the buildout hadn't started yet.
I bought my first tranche that week.
My Investment Thesis — The Five Pillars
1. Structural Position at the Start of the AI Supply Chain
As I mapped the supply chain, I kept arriving at the same conclusion: you can't build optical interconnects without InP substrates, and you can't build AI data centers at scale without optical interconnects. AXT is a gatekeeper material. There is no AI infrastructure without this substrate.
2. Plans to Double InP Capacity by End of 2026
This was a key catalyst in my thesis. Management announced they were on track to double InP manufacturing capacity from end-2025 levels by the close of 2026 — and the method impressed me. Rather than spending years building a new greenfield facility, AXT is repurposing an existing building previously used for GaAs crystal growth. That means the ramp is faster, cheaper, and lower risk than what a competitor doing a greenfield build would face.
CFO Gary Fischer underscored the financial discipline behind this:
"We have identified a clear path to doubling capacity that leverages existing infrastructure. This approach gives us speed and capital efficiency that a greenfield expansion simply cannot match. We expect to be meaningfully ahead of our nearest competitor on timeline."
Speed matters here. AXT's primary peer, Sumitomo Electric, is also expanding InP capacity — but their timeline is 2.5 to 3 years. AXT is targeting the end of 2026. That's a 12-to-18-month window of near-monopolistic capacity advantage in the fastest-growing corner of the substrate market.
3. The $550 Million War Chest
In April 2026, AXT raised approximately $550 million in a secondary equity offering, pricing 8.56 million shares at $64.25. At the time, the market was debating whether the dilution was too painful. I saw it differently.
A company that is running out of ideas doesn't raise $550 million. A company that sees a once-in-a-decade demand surge and needs to move fast — that company raises $550 million.
The proceeds are earmarked almost entirely for expanding Tongmei's InP manufacturing capacity. This isn't financial engineering. It's a capital deployment decision in the face of genuine customer demand.
4. Broadening the Customer Base to Tier-1 AI Names
One of the underappreciated parts of AXT's story is that it has historically sold to mid-tier transceiver manufacturers. That's changing. Management has been explicit that they are actively onboarding Tier-1 hyperscaler-adjacent customers — the kind of companies that order in volumes that make smaller customers look like rounding errors.
As per the Q3 2025 call, CEO Morris Young said:
"We are notably broadening our customer base. We are now in active qualification processes with several large-cap, high-volume accounts that we have not historically supplied. These are the companies that define the trajectory of the optical interconnect market, and we intend to be their substrate supplier of choice."
Getting onto a Tier-1 qualification list in semiconductors is a multi-year moat. Once you're in, you're in. That's the kind of compounding relationship that doesn't show up in a single quarter's revenue but that shapes the next five years.
5. The Only Real Risk Was Already Partially Priced In
No investment thesis is complete without an honest look at the risks. The biggest one for AXT is China.
Every dollar of AXT's revenue flows through its Chinese subsidiary, Beijing Tongmei Xtal Technology Co., Ltd. Since February 2025, China's Ministry of Commerce has required export permits for all three of AXT's substrate types — InP, GaAs, and Germanium. Permit delays hurt Q1 and Q2 2025 revenues materially. US tariffs on Chinese-made substrates escalated to 70%, essentially pricing AXT out of the North American market for now.
When I bought at $50, I believed these risks were largely priced into the stock. The company had already guided conservatively around permit timing. The $550 million raise eliminated near-term liquidity concerns entirely. And critically, China still has strong commercial incentives to keep issuing export permits — the global optical transceiver market is not built on domestic Chinese demand.
What the Numbers Told Me
As per my study of AXT's financials at the time of purchase, here is what I found most relevant:
The things that concerned me:
FY2025 revenue of $88.3 million was down from $99.4 million in FY2024 — the export permit disruptions were real
Gross margin had collapsed to 12.7% in FY2025, bottoming at a deeply negative -6.4% in Q1 2025 due to GaAs yield issues and factory underutilization
The company was posting losses — diluted EPS of -$0.49 for FY2025
The things that gave me conviction:
Current ratio of 2.72 — no near-term liquidity squeeze
Short-term debt of ~$61.5 million was manageable, and the $550 million raise made it almost irrelevant
Q3 2025 revenue of $28 million was the strongest quarter in over a year, showing sequential recovery
Management guidance for Q1 2026 of $26–28 million in revenue with a dramatically narrowed EPS loss of -$0.03 to -$0.05
The losses were real, but they were driven by temporary disruptions — not by a broken business model. The underlying demand was accelerating. That asymmetry is where I wanted to be.
The Tongmei STAR Market IPO — A Hidden Option
There's a secondary catalyst that doesn't get enough attention: AXT has been pursuing a listing of Tongmei on Shanghai's STAR Market since 2022. The China Securities Regulatory Commission accepted the application in August 2022, and review has been ongoing.
A successful IPO would do several things simultaneously: unlock domestic Chinese capital for Tongmei's expansion, provide a market-based valuation for AXT's primary asset, and reduce AXT's dependence on US equity markets for funding. It would also eliminate a potential ~$49 million redemption obligation tied to a private equity investor's stake in Tongmei — a liability that has been hanging over the stock.
I think of the STAR Market IPO as a free option on the stock. It's not in my base case. But if it happens, it's a meaningful upside catalyst.
Why I'm Still Holding
The stock ran from $42 to over $90 in roughly three weeks during April 2026 — a move that was faster and larger than I expected. I'll be honest: I trimmed a small position at $78, which I'm now regretting.
But for the core of my holding? I'm not touching it.
Here's why. Northland Capital Markets doubled their price target from $45 to $90 on April 20, 2026, citing what their analyst called "one of the most durable picks-and-shovels positions in the AI infrastructure stack." That framing mirrors exactly what I thought when I bought at $50.
"AXT's indium phosphide substrates are not a discretionary component in AI data-center design. They are a mandatory one. You cannot build a co-packaged optic or a high-speed pluggable transceiver without them. The capacity ramp AXT is executing over the next 18 months positions the company to capture outsized share in a market that is just beginning to inflect." — Northland Capital Markets analyst note, April 20, 2026
And on the Q1 2026 earnings call, CEO Morris Young offered what I consider the clearest articulation yet of the long-term opportunity:
"The transition from electrical to optical connectivity inside AI data centers is not a trend — it is a permanent infrastructure shift. Every GPU cluster built from this point forward will use more optical interconnects than the one before it. Indium phosphide is the material that makes that possible, and we are one of the very few companies in the world that can supply it at the quality and volume that the market will demand."
That is the kind of statement that makes me want to hold, not sell.
What I'm Watching Next
Every investment thesis has checkpoints. These are mine for AXT:
Green flags I'm watching for:
Q2 2026 gross margin recovery above 20% — this would signal the manufacturing underutilization era is truly behind us
First Tier-1 AI customer publicly acknowledged or implied in revenue commentary
Tongmei STAR Market IPO receiving final CSRC approval
InP capacity expansion milestones tracking ahead of the end-2026 target
Red flags that would make me reconsider:
A return of export permit disruptions affecting more than one quarter of shipments
US-China trade tensions escalating to outright export bans on substrate materials
A competitor (Sumitomo or an emerging Chinese player) reaching equivalent InP capacity faster than expected
Management guidance for InP capacity expansion being pushed back beyond 2026
The Bottom Line
I bought AXT at $50 because I believed I had found a company that was:
Structurally positioned at the beginning of one of the most important infrastructure buildouts of our generation
Being temporarily misvalued because of short-term permit disruptions that obscured the underlying demand
About to execute a capacity expansion that would lock in its competitive position before peers could respond
Backed by a balance sheet that, post-offering, gave management the runway to see the thesis through
Was I early? Yes. Was I patient during the drop from $50 to the low $40s? Yes — and that required conviction.
But this is what I've learned after years of investing in technology stocks: the best entries often feel uncomfortable. The stock that everyone agrees is a buy is usually already priced for perfection. The stock that has real risks, real headwinds, and a real story underneath those headwinds — that's where asymmetric returns come from.
AXT had all of that. And in my assessment, the story is still only in its early chapters.
The author holds a long position in AXT Inc. (AXTI) as of the date of publication. This article is written for informational and entertainment purposes only and does not constitute financial advice. Always do your own research before making any investment decision.