Substack作者更新POET Tech观点：现金更充足、客户管线扩大，但Marvell订单取消后仍需等待H2出货验证

It’s been a while since I last spoke about POET Tech.

The Marvell blunder made it hard for me to trust management enough to hold any real size. But there’s a lot going on, so let’s talk about it.

The balance sheet is much larger. The customer list is broader. Lumilens gives the 2027 story a clearer revenue target. Management is talking more directly about Malaysia capacity, H2 shipments, Blazar, external light sources, and the design-win funnel.

The bear case is also still very present. Marvell cancelled the Celestial purchase orders, taking out one of my main validation points. I moved that part of the thesis to zero, and it stays there until I see a repaired relationship, reinstated orders, or a new Marvell agreement. They still haven’t executed in the way that I would really like to see.

So the story is bigger. The proof still has to show up in shipments.

Let’s unpack everything on my brain.

*The rest of this post is for paid subscribers. Below the paywall: the balance sheet and Lumilens detail, what I think happened with Marvell, the real H2 test I’m watching, Malaysia’s capacity numbers, and where I land on valuation given the dilution.*

*This report is for informational and educational purposes only and should never serve as the basis for an investment decision. I may hold positions in companies mentioned, including POET, and my views may change at any time without notice. Readers should conduct their own due diligence and consult a qualified financial professional before buying or selling any security.*

Wow, they have a ton of cash. I feel like I keep saying this and it keeps going higher. POET now has roughly $830 million of pro forma cash and short-term investments after the May financing. That’s practically half the market cap, so there’s no imminent financing risk. I thought that two raises ago too, and they keep raising. If the capital goes toward hypergrowth, great. If not, not great. Why are you taking so much of my growth potential? But right now it gives them real room to hire, buy equipment, secure supply, and support customer programs. Management plans to deploy roughly $50 million into capex in H2, which says they’re building capacity ahead of the ramp rather than waiting for every order to be fully visible. Right move if demand is real. Higher execution risk if it isn’t.

**Lumilens**: an initial $50 million purchase order, products expected to deploy in 2027, with management framing the relationship at more than $500 million over five years. To be precise, the $50 million is commercial visibility, not realized revenue. The products still need development, qualification by Lumilens, qualification by Lumilens’s own customers, manufacturing, shipment, and acceptance. Real positive. Still has to be executed. That’s always what this comes down to me. Execution proof.

**Customer breadth**: Lumilens, ADTRAN, LITEON, Lessengers, NTT Innovative Devices, unnamed system integrators, and external-light-source customers. Management says the current funnel could translate into more than $100 million of annualized revenue over the next two years. That’s run-rate language, not reported revenue, and the “if those programs qualify and ramp” is doing a lot of work in that sentence.

**Now Marvell**. Celestial connected POET to optical scale-up architectures, Marvell acquired Celestial, and that created a great read-through. Then Marvell cancelled every Celestial purchase order for POET. POET disclosed the written cancellation came April 23, announced it publicly four days later, and the stated basis was alleged disclosure of purchase-order and shipping information, not product performance. That suggests this was about communication and legal process rather than the technology. Although, you could push the argument that Marvell wanted out anyway and used this as a reason to get out. But who knows. Doesn’t matter because that’s pure speculation that I can’t support given what I know. But the cleanest AI-scale-up validation point in the thesis is gone, and it gets zero credit from me until it’s repaired or replaced. The AGM gave no cleanup on it either (as far as I heard). I expected something and didn’t get it. Replacement validation now has to come from the existing $5 million-plus order, ADTRAN, Lumilens, LITEON, Lessengers, Blazar, or a new customer entirely.

Which gets to what I actually care about. The $5 million-plus 800G order, disclosed last year, expected to ship in H2, separate from the cancelled Celestial orders. Management again confirmed delivery against existing purchase orders this half. I want to know if POET actually ships. Please, POET. Ship on time. Partnerships and demos build a foundation, but the thesis only really changes once product leaves the factory, gets accepted, revenue gets recognized, and follow-on orders show up. Shipped. Accepted. Qualified. Repeat order. Gross margin. I’ll wait and pay up for that. I am happy taking a bigger position once I get more proof, even if it’s at a higher price. That’s just me.

The other number I’m watching is the 30,000-plus optical engine shipment target for 2026, light-source production in Malaysia first, then 800G engines. I want to know if that target still stands after Marvell. Reaffirmed is a good sign. Walked down, pay attention and definitely a yellow flag.

**Malaysia**: the transition from China is complete, the first tranche of capacity is qualified for production, and more space is coming online at Globetronics and NationGate. The capacity gap is the real story though. Management is framing roughly a ten-fold expansion into 2027, from about 1 million optical engines per year today to demand exiting 2027 that could require 1 million per month. That’s the upside and the execution risk in one number, not unlike AAOI for anyone in that name too. Real demand plus clean Malaysia scaling gets POET into a much bigger commercial ramp. Slower scaling or uneven yields push the curve right.

**Revenue visibility today**: still tiny in reported terms. Not a problem. It’s still in a ‘pre-commercial, pre-inflection’ bucket. Q1 2026 revenue was about $503,000. The future sits in buckets. Near-term from the 800G order and undisclosed ADTRAN orders, development-stage from the $1 million-plus external-light-source NRE, 2027 commercial from the $50 million Lumilens order, pipeline from the $100 million-plus claim, and longer-dated upside from the $500 million-plus Lumilens potential, Blazar, and OCI. Better setup than a few months ago. Still heavily weighted toward future execution of course.

One thing worth noting is that management said they expect to be selective with new customer announcements over the next 12 months. I wonder how much of that is downstream of Marvell. But read the other way, it’s a fine signal. The pipeline is already large enough that the job now is development, qualification, and ramping current customers rather than collecting logos. That’s how I’ll judge them too. Less interested in another partner announcement, more interested in whether the current ones convert. Show us the money POET!

Several products are qualified for production and headed there imminently. Lumilens is a different bucket, still in development, still needing qualification by Lumilens and then by Lumilens’s customers. Some products are close though.

Blazar is the interesting long-dated option, a high-power multi-channel external laser platform tied to OCI, external light sources, CPO, and future AI scale-up. Could be large. It’s a 2028 story: sampling in H2 2026, design wins and qualification in 2027, production in 2028.

PFIC and redomicile, from the short report a couple months back, is still unresolved. Board approved a U.S. redomicile path, but at the AGM Tom Mika said they’re still assessing with legal and tax advisors and won’t know 2026 PFIC status until year-end. Not core to the thesis, but it deserves clarity given how much of this stock’s base is U.S. retail. The legal and promotion overhang from the Marvell cancellation is also still hanging around I guess.

My view is constructive but disciplined. The setup genuinely improved, and Marvell was a genuine hit, both at once. The bull case comes down to whether this story turns into shipped engines, recognized revenue, repeat orders, and a revenue curve that actually supports today’s valuation. That’s why H2 carries so much weight. Ship against existing orders, show real product revenue, prove Malaysia can run at volume, and confidence improves a lot. Slip on shipments or stay vague on customers while the valuation holds, and the market starts asking how much of this pipeline is actually real.

So for me, it remains a very small position. Like I said, I will pay up for proof.

Does this mean the stock won’t go up? Absolutely not. I would be surprised if it didn’t go up. Especially if the sector catches a tailwind. But I just want more proof for myself.

Not a recommendation of course.

I know I have some really informed POET investors that sub. Please feel free to add any additional notes or some points I might be missing! I haven’t been tracking this daily like I have in the past.

*This report is for informational and educational purposes only and should never serve as the basis for an investment decision. I may hold positions in companies mentioned, including POET, and my views may change at any time without notice. Readers should conduct their own due diligence and consult a qualified financial professional before buying or selling any security.*
The story is progressing