Tech and AI
Fusion laboratory spins out an engineering AI built to preserve traceability and professional judgement, with a design firm as investor and first customer
By Staff Writer | 18 August 2026
Singular Machines has closed a strategic investment round led by a public seed fund and an Oxford investor, joined by a global engineering firm which has also contracted to use the platform. The pitch is unusual for an AI product: the claim is that it keeps the assurance trail intact rather than that it removes people from the work.
Fusion engineering has a property that most engineering does not: nothing about it can be approximate. The tolerances are severe, the assurance requirements are heavy, and every decision has to be defensible years afterwards. A company spun out of the national fusion laboratory on 17 August is betting that the discipline this forces is worth more outside fusion than in it.
Singular Machines has completed a strategic investment round led by the UK Innovation and Science Seed Fund, managed by Future Planet Capital, and Oxford Science Enterprises, joined by the engineering firm Arup and by Miraisozo Investments of Japan. Arup has separately contracted to use the company's coEngen platform, and will provide technical, commercial and strategic support, helping to shape and test it against real engineering workflows. The size of the round has not been disclosed.
The support from UKAEA's innovation team helped us get Singular Machines off the ground. Fusion is where we started, but the problem coEngen solves runs right across engineering: projects have become so complex that they are reaching the limits of what human coordination alone can handle. We are building coEngen so these projects can flourish.
Adrian van Arkel, Founder and Chief Executive, Singular Machines
A diagnosis worth reading twice
The stated problem is not that engineers work too slowly. It is that projects have become complex enough to exceed what human coordination alone can handle, which is a claim about interfaces rather than about productivity.
Anyone who has read a delay analysis on a large scheme will recognise it. The hours are rarely lost in the calculation. They are lost between disciplines, in information that arrived late, in a change whose consequences nobody traced across every drawing it touched, in an approval that could not be found when it was needed. Automating the calculation does nothing for any of that.
The company describes coEngen as an agentic platform designed to maintain the traceability, assurance and professional judgement that engineering practice depends on, so that engineers work more efficiently while engineering expertise stays at the centre of decision-making. That framing is the opposite of the usual one, and it is the framing a regulated industry can actually buy. An engineer who signs a design remains answerable for it, and a tool that cannot show its working is a liability rather than an asset.
The route from laboratory to industry
The spinout sits within the UK Atomic Energy Authority's 2026 to 2030 strategy of turning fusion research into commercial opportunity, so that public research spending produces industrial return rather than only papers. Heather Lewtas, Interim Chief Development Officer at the authority, said its science and innovation "is accelerating the creation of commercial opportunities in the UK" and that it was pleased to support the company in spinning out and developing its industrial partnerships.
The authority makes a broader claim for spinouts of this kind, that they develop skills, intellectual property and technology which generate value across several sectors, attract investment and create high-skilled jobs. The step that gives this one weight is the customer contract arriving with the investment. A design firm that puts its own money in and then agrees to run the product against live workflows is a harder endorsement to arrange than a funding round, and a considerably harder one to reverse.
What is not yet known
The record is silent on several things a buyer would want. It gives no figure for the investment and no valuation. It does not say what coEngen does in operation, beyond the description of its purpose. It does not say which of Arup's workflows will be used to test it, over what period, or against what measure of success. It names no delivery date for a commercial release.
Those gaps are ordinary at this stage of a spinout and none of them is a criticism. They do mean that the only substantive claim on the table today is the one about traceability, and that claim will be tested where every engineering software claim is tested, which is the first time somebody has to produce the audit trail for a decision the machine helped to make.