$LITE runs five indium phosphide fabs and still ships more than 30% below what customers ask for. Its CEO spent July warning that this shortage is heading somewhere worse than DRAM and NAND. $SIVE sells indium phosphide lasers. Someone on its earnings call this week asked whether it had enough capacity. The CEO said capacity is not the bottleneck. The scarcest input in AI optics, and the company selling it says it has plenty. Doesn't that seem backwards? The answer is in how he described the size of his own business 01 · THE SHORTAGE Silicon is an indirect bandgap material. It does not emit light. Every silicon photonics design imports a III-V source, and in AI data centers that source is indium phosphide. Sivers put the chain on a slide, stage by stage. Refined indium, where China accounted for 69.1% of 2025 production. InP substrates, gated by export licensing and by how fast anyone grows crystals wider. Epi wafers. Then laser chips, where high-power continuous-wave parts need different processes than the parts telecom spent twenty years optimising. Packaging makes it tighter. A traditional pluggable runs on 70 to 100 milliwatt lasers. Extra-dense pluggables move to 200. Co-packaged optics needs 200 to 400 and adds optical amplifiers on top. Same photons, more power per source, more parts per board. On the call the CEO called pluggables a market that is available and severely supply constrained now. He was describing his own market. 02 · WHAT THE QUARTER SHOWED Net sales 53.8 million kronor against 61.4, down 12%. That is the wrong line. Underneath it, product and hardware revenue was 29.2 against NRE and support at 24.6. Product is 54% of the quarter. A year ago it was 25.8 against 35.6, so NRE was the business and product was the side. That inverted in twelve months, because management is shutting NRE down on purpose to free engineers. Adjusted EBITDA -35.5 against -20.9. Cash from operations -70.0 against -20.1. The company raised 825 million kronor gross across two directed issues in April and June, converted 12 million dollars of debt in July, and carries no debt. Asked about margins, the CEO said products carry higher margin than NRE and pointed at a financial model published nine months ago rather than a new number. 03 · WHAT THE PIPELINE ACTUALLY DID The pipeline is 1.216 billion dollars in July against 799 million in May, up 268% from the end of 2025. It breaks into three stages, and they moved in opposite directions. Pre design-in went from 343 million to 778. Design-in and design-win went from 369 to 354. Sponsored development went from 87 to 83. All of the growth sits in the earliest bucket. Both buckets nearest to revenue shrank. Someone on the call asked whether the pipeline is rolling over without converting. The CEO pushed back, then explained the window: once 2026 is finished there is no point showing 2026, so the frame moves to the next five years. That is accurate and it is also the mechanism. A pipeline measured over a horizon that advances gains room each time it advances. On Sivers' own chart a custom product takes 3 to 9 months to qualify, 18 to 36 through design-in, then 9 to 12 more before production. The bucket that grew sits three to five years from cash. 04 · WHAT THE CALL GAVE AWAY Asked directly how much of the pipeline it could capture before running into production limits, the CEO said he does not see production capacity becoming the limiting factor as the company grows from a small number to its growth targets. Small number is doing the work in that sentence. Capacity is not scarce for Sivers because Sivers is not yet buying much of it. The industry constraint and the company's comfort are the same fact viewed from opposite ends. The capacity itself is borrowed. Long-run the model is one third in-house, two thirds foundry, and Glasgow phase one is dated Q4 2027 and later, so the 2027 ramps run in someone else's fab. Asked what reserving that future capacity costs, the CEO said the partners are committed strategically and not by paying each other. No prepayment, in an industry where prepaid long-term agreements are how substrate supply has been getting locked. Set that against the cycle. $NVDA guided next quarter to 108 billion dollars and called its own 70% growth outlook supply constrained, and it answered the laser question with its balance sheet, funding Lumentum and Lumentum's largest competitor at once. Across the eight supply chain layers I track, ninety-day consensus revisions average positive 43%; optical interconnect is the only negative one at -5.1%, on the lowest delivery rate of the eight. Sivers is funding itself instead, 825 million kronor at a time, into a shortage it has told us it is still too small to feel. If the LiDAR order lands with a figure attached and the Jabil beta clears qualification in Q4, small number stops being the explanation and the borrowed capacity starts getting tested.