Summary
- Sivers is investing $30m in its indium-phosphide manufacturing operation in Glasgow.
- It is targeting annual capacity above 100 million continuous-wave DFB lasers.
- Expansion starts in H2 2026, with the additional capacity expected online in Q4 2027.
Sivers Semiconductors is investing $30 million to expand photonics manufacturing in Glasgow, targeting annual capacity above 100 million continuous-wave distributed-feedback lasers for AI data centre and optical-networking applications.
The expansion is due to begin in the second half of 2026 and is expected to become operational in the fourth quarter of 2027. Sivers said the programme will add manufacturing capacity, new process capabilities, and greater automation at its indium-phosphide facility.
The investment moves a component of the AI infrastructure supply chain that is often discussed as a chip issue into a physical manufacturing context. Large GPU clusters do not operate as isolated processors: the machines have to exchange large volumes of data at high speed, placing increasing demands on optical links within and between racks.
Distributed-feedback lasers are one of the components used to generate the optical signals needed for those links. As data rates rise and clusters become larger, the number, performance, power consumption, and availability of optical components become part of the capacity equation alongside GPUs, electrical power, and cooling.
Sivers describes the Glasgow programme as a shift from a Fab-Lite approach towards what it calls Hybrid Manufacturing. The company intends to expand selected in-house capabilities while continuing to use external foundry, packaging, and manufacturing partners. That keeps part of the production process under direct control without attempting to internalise the entire semiconductor supply chain.
The scale target is notable. Capacity above 100 million lasers a year implies a manufacturing operation designed for volume rather than specialist low-run photonics. The practical challenge will be maintaining yield, quality, and process consistency as output grows — particularly for components entering data centre networks where failure rates and power efficiency become consequential at fleet scale.
The Glasgow expansion also puts a piece of AI data centre manufacturing in the UK rather than limiting the infrastructure discussion to new server buildings. Data centres depend on a much wider industrial base: switchgear, transformers, power electronics, pumps, heat exchangers, optical components, cables, racks, and semiconductor packaging all have their own production constraints and investment cycles.
Those supply chains do not expand at the same speed. A facility developer may be able to secure land and planning permission while still waiting for a grid connection or critical electrical equipment. Similarly, a compute operator can secure processors but remain constrained by the networking equipment needed to assemble them into useful clusters. Manufacturing investment in photonics is therefore one part of the effort to keep the wider system from lagging behind processor deployments.
Sivers says expected AI data centre demand and customer production ramps are behind the expansion. Those demand forecasts still have to translate into orders, and the new capacity will not be operational until late 2027. The programme therefore carries the usual manufacturing risk of investing ahead of a market whose deployment schedules can move.
What is already concrete is the capital allocation. The Glasgow site is moving beyond its existing scale, with $30 million committed to equipment, processes, and automation rather than to another AI capacity announcement measured only in theoretical GPUs.
By the time the expanded facility is expected to come online, the question will be whether optical-networking demand has grown at the pace its customers anticipate. If it has, the constraint will not simply be producing more lasers, but producing them consistently enough, cheaply enough, and with the performance required by increasingly dense compute fabrics.
