AI buying is shifting from just grabbing GPUs to making sure the whole system actually comes online on schedule. That raises the value of suppliers who can deliver. $AAOI’s edge is that it controls a lot of the lasers, optical components, light engines, module manufacturing, and testing itself. Vertical integration gives it more control over cost, quality, and how fast it can ramp. 1.6T modules still need external parts like DSPs and TIAs. And if customers don’t get memory, switches, or servers on time, the whole system can’t go live, so optical module pull-ins can get pushed back too. AAOI has already said some of its 1.6T revenue is constrained by key component supply. $SIVE sits further upstream. It sells CW DFB lasers and laser arrays. Once that part gets designed into a customer’s architecture, switching suppliers usually means going through reliability and qualification again, so the qualification itself becomes a barrier. The Jabil project is planned to hit beta build in Q4 2026, then customer qualification. Sivers expects production orders in H1 2027 and ramp in H2. On October 20 their strategic LiDAR customer already started formal production orders, with shipments beginning in Q4. At least shows they’re moving design wins into actual volume. What’s still tight in AI optical is a full supply chain that can deliver on time. Still constructive on the space.

Recalculated the LiDAR order $SIVE announced yesterday. Company’s lifetime revenue expectation on the project is $53M–$138M. For Sivers’ current size, that’s not a small project. Using the earlier orders they publicly named as a reference: ALL.SPACE $8.2M Tachyon $3M SemiNex $3.4M Those three add up to about $14.6M. So the LiDAR project’s potential is roughly 3.6x to 9.5x the total of those three. Looking at the stock here, it starts to feel more interesting. Have to keep the distinction clear though. $53M–$138M is potential revenue over the full customer product lifecycle. They haven’t disclosed the size of the first production order yet. Still need to track that. If the rolling production orders keep coming based on the customer’s multi-year plan, this LiDAR line has a real shot at becoming the first scaled revenue stream in Sivers Photonics. If that executes cleanly, I’d also feel better about the Jabil 1.6T project.

Learning about a company today: $SIVE. One of the optical names I’ve spent more time on. Servers move electrical signals. Fiber runs light. Optical modules do the conversion in between. Sivers sits further upstream — they make the InP lasers that actually generate the light. As AI clusters move from 800G to 1.6T and 3.2T, bandwidth keeps going up, and the requirements on laser power, efficiency, and stability go up with it. Sivers is already working on 70mW, 100mW, 200mW DFB lasers and laser arrays. That puts them in a position to supply pluggable, NPO, and CPO. That’s the part I find interesting about their setup. You don’t have to bet which optical architecture wins. As long as AI networking keeps scaling, the light source layer has to improve. On execution, the key project right now is with Jabil. They’re developing a 1.6T LRO module using Sivers DFB lasers. If that line gets traction, the business model changes. A big chunk of revenue used to come from NRE — customers paying them to do development work. They’re now trying to shift more toward product revenue. You can see it in the numbers if you look at the reports. Q2 revenue was only SEK 53.8M, so the top line still looks weak. But product revenue grew 18% year-over-year at constant currency. Opportunity pipeline is now at $1.2B, up 268%. They’ve also announced $30M to expand the Glasgow InP fab, aiming for over 100 million CW DFB lasers per year by Q4 2027. For a company this size, scaling capacity that aggressively suggests management is preparing for real AI optical volume. The risk is obvious — if orders don’t show up, that capacity turns into cost. Debt is also worth watching, especially with the cloud spending questions floating around. They did finish a financing round this year: the $12M convertible converted, and the remaining $5M term loan has been paid off. Near-term, the main question has shifted from survival to actually selling product. Market cap is around SEK 7.8B right now. TTM revenue is about SEK 278M, so roughly 28x sales. In the optical packaging space, worth putting this on the same watchlist as Marvell. Just my view.