Key points
- Glasgow fab targeted at 100 million lasers
- July's raise named InP capacity expansion
- Test throughput looks tighter than wafer supply
- Broadcom is tripling its own laser fabs
Sivers Semiconductors (SIVEF) is putting $30 million into its Glasgow fab to reach more than 100 million continuous-wave DFB lasers a year, with the line meant to be running at that rate in the fourth quarter of 2027. The release names no customer, and Sivers has still not disclosed a purchase order for the lasers. SIVEF closed Thursday at $2.85, up 26.16%, and the Stockholm listing closed at SEK 25.46, up 15.20%. The release doesn't say where the money comes from, though July's SEK 700 million raise named InP capacity expansion among its uses of proceeds.
What the July raise actually said
On July 1, Sivers resolved a directed share issue of 12,280,701 shares at SEK 57, about SEK 700 million, oversubscribed. The release said proceeds would go to "expanding manufacturing capacity for InP lasers and optical amplifiers" and that the company would "make proactive CAPEX investments to increase InP manufacturing capacity for our Photonics business given the long equipment lead-times." That release doesn't earmark this project, and cash is fungible, so this is a close match rather than a straight line from one to the other. One note for balance-sheet math: the SEK 62.7 million of cash in the second-quarter report is dated June 30, before the issue was resolved on July 1, with settlement later still.
The number everyone is going to quote
Here it is, conditions first, because the number will travel further than they will. At full utilization, if every laser shipped inside an eight-channel array, and if the $50 volume price still applied at that scale, the installed capacity would correspond to roughly $625 million in annual sales. Call that a mechanical throughput ceiling at the bottom of the cited price range. Sivers has never published it, and it isn't guidance.
The price comes from the company. A May 2024 Sivers presentation to the NMI carries a slide reading "Sivers Photonics' Laser Array has ASP of $50-100," pointing at the 8-channel O-band part inside Ayar Labs' SuperNova light source, and a second slide describing "volume pricing $50-$100 per array depending on volumes." Eight lasers to an array puts 100 million lasers at 12.5 million arrays.
Every condition in that sentence is load-bearing. Yield has to hold. Utilization would have to remain near full, which fabs rarely sustain. The mix has to stay in arrays, and it won't entirely, since Sivers also ships single lasers into automotive LiDAR. The deck is more than two years old. And a price the company describes as depending on volumes points down at 12.5 million units, which is why the $100 end of that band is the harder half to defend. Q4 2027 is also when the capacity is meant to exist, not a date when $625 million of revenue arrives.
What actually caps the line
The tighter number sits underneath all of that. The same deck lists Glasgow as a 100mm fab at 5,000 wafer starts a year with up to 125,000 die sites per wafer. Die sites aren't saleable lasers. Yield, utilization, process mix, packaging and qualification all sit in between, so that's headroom rather than capacity. But the deck also puts high-volume test capacity at ">2M lasers/month" and describes Sivers as an "established volume supplier, shipping over 1 million lasers" a month. Both are floors rather than exact figures, so the multiples are ceilings. A 100 million unit year is less than about 4.2 times the disclosed test minimum, and less than about 8.3 times the disclosed shipping minimum. That makes test throughput the clearest disclosed constraint, and a likely target of the automation spending.
12.5 million arrays shows up one more place, in Sivers' own model. That deck assumes four arrays per GPU and sizes the addressable market at 75 to 82 million arrays a year in 2027 and 2028. That's about 15% to 17% of it, depending on the year. The model has a 15% penetration row and prices it at $564 million to $1.1 billion for 2027. Here's the catch. The fab only runs at rate from the fourth quarter of 2027, and a line starting in the last quarter can't serve a full year of demand. So this is really 2028 capacity, and the deck's 2028 row assumes 30% penetration, about double what 12.5 million arrays would cover.
Broadcom cuts both ways
Broadcom (AVGO) helped the timing along. On Wednesday evening's earnings call, chief executive Hock Tan said "Demand for lasers, whether it is EML lasers, CW lasers, is far surpassing supply out there in the industry." CW lasers are exactly what Sivers makes, named by one of the market's largest participants, the night before Sivers said it was building more of them.
That reads well for Sivers, but the same call cuts back. Broadcom's semiconductor solutions president Charlie Kawwas said the company's own indium phosphide factories, "both in the U.S. as well as in Singapore, we are actually more than tripling them year-on-year." That matters. Scarcity is worth the most to a small supplier when nobody else can relieve it, and Broadcom is moving to relieve it at scale.
What's still missing
So: the Glasgow spend matches a use of proceeds named in July's raise, the capacity sits near the company's own 15% case, and test throughput is the tightest constraint disclosed. What's missing is what's been missing all year. We've covered the four-session collapse, the quarter behind it, and July's dilution. Partnerships, programs, a raise, now a fab. No disclosed customer order for any of it.
Q4 2027 is the date the company picked. That's the one to hold it to.
