Salience Labs raises $125M to attack the data centre's power problem with light

Salience Labs, an Oxford chip company, has raised $125M (£94M) in a Series B led by Bullhound Capital, in a round whose investor list reads like a statement of intent: AMD Ventures, Intel Capital, SBI Group, Hudson River Trading and the British Business Bank all took part. The money goes towards scaling manufacturing and building out a Silicon Valley presence.

What Salience makes is a direct response to the constraint strangling AI infrastructure — the same one behind this week’s news that Microsoft reportedly walked away from the UK’s largest planned supercomputer over a site that cannot be powered until the 2030s. The problem is not only generating enough electricity; it is that a surprising amount of a data centre’s power is burned not on computation but on moving data between chips and servers.

Switching light, not electrons

Salience’s answer is photonic: optical circuit switching, or OCS. Instead of repeatedly converting data between light and electrical signals as it crosses a data centre, its chips keep it travelling as light, through an optical switch and a low-loss optical fabric. The claimed payoff is less power and less delay spent on data movement — which, as CEO and co-founder Vaysh Kewada puts it, means “OCS will play a critical role in AI datacenter networking.” In plain terms: do the plumbing with light, and more of the building’s finite electricity can go to running the models.

The timing speaks to how acute the bottleneck has become. Ofgem’s figures show the UK’s data-centre grid-connection queue tripling from 41 GW to 125 GW between late 2024 and mid-2025. When you cannot simply plug in more power, making each watt do more work stops being an efficiency nicety and becomes the whole game.

A deep UK hardware bench

Salience lands among an unusually strong set of British AI-hardware companies all chasing cheaper, more efficient compute from different angles — Graphcore’s accelerators, Lumai’s optical computing, Fractile’s in-memory inference. That the UK has a genuine cluster here is one of the more underappreciated strengths of its AI ecosystem, and exactly the kind of sovereign capability the government says it wants.

The honest caveats are the ones that apply to any hardware company at this stage: Salience is still scaling towards commercial-volume manufacturing and has not published deployment numbers or independent energy-savings figures, and competing to supply the hyperscalers’ networking is brutally capital-intensive. But a $125M round with both AMD and Intel on the cap table is a serious signal that the incumbents think light-based switching is part of the answer — and it is a rare piece of good news in a week when the UK’s compute ambitions otherwise ran into the wall of the national grid.