Extend Robotics Raises £2.6M to Expand Outcome-Based Industrial Robotics Model

Extend Robotics has closed a £2.6 million pre-Series A round led by Skyworks Venture Capital Fund, with Neo Venture Europe and Zip Capital participating. The money will be used to expand its "Result-as-a-Service" business in the UK and Europe and to build a network of remote robot operators.

The amount itself is not large, but the claim behind it is quite unconventional: under the Result-as-a-Service model, customers pay for completed output rather than buying equipment, and Extend says it bears responsibility for cycle time and task success rate. Founder and CEO Chang Liu, a former postdoctoral researcher in aerial robotics at Imperial College London, puts it this way: manufacturers have roles they cannot fill, and finance directors are unwilling to sign off on a robot that may not work, so the right approach is to sell "completed work," price it by output and take on the risk yourself.

In other words, this company is not selling robotic arms or humanoid robots; it is moving the uncertainty of deploying automation off the customer's books and onto its own. That is the most noteworthy signal in this round, and also the thing most in need of verification.

What supports this model is Extend's AMAS platform. As described in the report, it combines immersive teleoperation with AI assistance, letting people with no robotics experience remotely control robotic arms or humanoid robots through a 3D interface; AI handles the routine parts of a task, while human operators take on work requiring judgment or flexibility.

This combination determines Extend's target scenarios: manufacturing work that is hard to automate with traditional systems, such as low-volume production, high-mix operations and maintenance, as well as hazardous work in nuclear, aerospace and high-voltage environments. What these scenarios share is small batch size, high variability and dangerous environments, where traditional fixed automation struggles to amortize costs.

What needs to be distinguished is that the platform capability description comes from the company's own account, and the report provides no technical performance metrics, such as specific figures for task success rate or cycle time. So "AI handles the routine parts, humans handle the judgment parts" is currently a product positioning statement, not a verifiable performance conclusion.

Extend reports that it currently has 35 subscription customers across four sales regions, covering automotive, aerospace, nuclear, robotics and agriculture. The company also says its revenue has doubled year-over-year for three consecutive years.

These two sets of numbers are key to understanding the company's current state, but both must be read as "the company's own account": the report does not disclose the specific composition, geographic distribution or growth trend of the 35 customers, nor does it give the specific revenue base or financial data behind the doubling. With only one source, and that source relaying the company's statements, these numbers do not constitute independently verified operating facts.

Also worth flagging is the funding round itself: this is a pre-Series A, the amount is £2.6 million, and the report does not disclose valuation or follow-on funding plans. For a company claiming three consecutive years of doubled revenue, the report offers no explanation of how the funding size matches the growth narrative.

The report gives two specific deployment scenarios. First, Extend says the model has been used on a production line at Adrian Scripps, one of the UK's largest apple growers, where an Extend humanoid robot performs quality control work on a live packing line; the deployment was streamed live at the Google DeepMind Robotics Accelerator Demo Day in September 2026. Extend is one of 12 companies selected for that three-month program.

Second, at Leyland Trucks, Extend trained a robot to install a high-voltage main service disconnect unit, a task that previously involved live electrical work; Extend says a second deployment handles painting of custom parts.

The value of these two cases is that they pull "pay per outcome" from concept down to specific workstations: quality control and live electrical work are both tasks with high demands on reliability and safety. But the report does not disclose the actual operating duration, stability or economic benefits of these deployments, so for now they can only prove that something "happened," not that it "works."

Another use of this round's funds is to build a distributed operator network. Extend says operators can control industrial robots without being on the factory floor, for example remotely handling work in paint booths, gloveboxes or high-voltage work areas from elsewhere in the UK or Europe.

This network is not a nice-to-have; it is the premise of whether Result-as-a-Service can hold up: if the company is to be responsible for cycle time and task success rate, it must have a stable supply of human labor to cover the parts AI cannot handle. The report does not state the network's size, timeline or number of operators already online, nor whether it can meet industrial-grade real-time requirements across countries and time zones.

So the remote operator network is currently a plan under construction, not a proven delivery system. It is at once the most reasonable use of this funding and the part hardest for outside observers to verify in the short term.

The report mentions two forms of external recognition: Extend says its platform is hardware-neutral and was named an official integrator for Unitree Robotics in January; Nvidia selected Extend as one of five companies for its GTC 2026 robotics startup showcase.

Of these two, the Nvidia selection is presented in the report as a statement of fact, while the Unitree integrator status comes from the company's own account. Both can boost credibility, but neither can be directly equated with a technical partnership, an investment relationship or a commercial order — the report does not say whether the Nvidia showcase selection implies a technical partnership or investment relationship, nor does it disclose the specific terms and exclusivity of the Unitree cooperation.

Whether there is tension between hardware neutrality and official integrator status is also left unanswered by the report. For readers, the safer reading is: these are ecosystem signals worth noting, not commercial conclusions that can be extrapolated.

Stringing the above facts together, Extend's bet can be summarized as: in low-volume, high-mix, hazardous work scenarios that traditional automation struggles to cover, replace "selling equipment" with "human-machine hybrid + pay per output," and shift the risk of automation failure from the customer onto itself. The £2.6 million pre-Series A is the ammunition backing this bet, with a clearly stated use pointing to geographic expansion and the remote operator network.

But the verification chain for this bet is currently broken: customer counts, revenue growth and deployment results all come from the company's own account, and the single-source report provides no first-party or third-party cross-verification; the pricing mechanism and customer payment amounts are undisclosed; the scale and timeline of the remote operator network are undisclosed; and technical performance metrics are missing.

So at this stage, what can be confirmed is the funding fact, the investors, the use of funds, the company's founding year, the AMAS platform positioning, the target scenarios, and the Nvidia GTC 2026 showcase selection; all other judgments about operating quality and delivery capability should be held as unverified company claims. For those watching industrial automation business models, what is worth tracking about Extend is not how much it raised, but whether it can turn "pay per outcome" from two demonstration deployments into a replicable revenue structure.