Optalysys
UK startup using silicon photonics for matrix multiplication in deep-learning operations.
Last refreshed: 1 May 2026 · Appears in 1 active topic
Will Optalysys's decade of photonics research translate into a government fab commitment in June?
Timeline for Optalysys
Named by Kendall as British AI hardware supply-chain candidate
UK Startups and Innovation: Kendall names UK chip five at RUSI- How does Optalysys use light to speed up AI?
- Optalysys uses silicon photonics — routing light through CMOS-compatible chip fabrication — to perform the matrix multiplications that dominate deep-learning inference. Optical paths execute these operations with lower energy and latency than electronic equivalents.Source: Lowdown reporting
- How old is Optalysys and when was it founded?
- Optalysys was founded in 2013, making it one of the longest-operating companies in the British photonic AI compute space, pre-dating the deep-learning wave that created commercial demand for its approach.Source: Lowdown reporting
- What is silicon photonics and why does it matter for AI chips?
- Silicon photonics routes light signals through chips made with standard CMOS semiconductor manufacturing, making it possible to build optical compute using existing fab infrastructure. For AI, it offers lower-heat, lower-latency matrix multiplication than conventional electronic chips.Source: Lowdown reporting
Background
Optalysys was named by Secretary of State Liz Kendall on 28 April 2026 at RUSI as one of five British AI hardware supply-chain candidates for the government's incoming AI Hardware Plan, to be announced at London Tech Week in June. The ministerial endorsement was the first time any of the five had been named publicly by a Cabinet minister, despite Optalysys having been one of the longer-operating companies in the British photonic compute space .
Founded in 2013, Optalysys uses silicon photonics to execute the matrix multiplications at the core of deep-learning inference. Silicon photonics routes light through standard CMOS-compatible wafer fabrication rather than specialised optical media, making the manufacturing pathway more accessible than free-space optical approaches. The technique is particularly suited to the convolution and attention operations that dominate modern neural-network architectures, where the same matrix multiply is executed repeatedly and benefits from the low-latency, low-heat properties of optical paths. Optalysys has been one of the earliest commercial actors in this space globally, pre-dating the deep-learning wave that created mass demand for its approach.
The company sits in a British photonic compute cluster with Lumai and Salience Labs, all three named in Kendall's speech. The cluster's combined ministerial endorsement is significant not just for each company individually but for the photonic compute architectural bet as a whole: the AI Hardware Plan's June launch will reveal whether sovereign backing takes the form of purchase commitments that can underwrite a production line, or advisory endorsement that does not.