How deeptech companies can use AI for smarter R&D bets

Britain's technology funding figures look reassuring at first. UK companies raised £14.4 billion in the first half of 2026, while the average round reached a record £5.4 million.

But this was not a broad-based funding boom. Just three AI companies accounted for 29% of all capital raised, the number of deals fell, and total investment remained 21% below the H1 2022 peak in nominal terms.

Deeptech also faces a real capital shortage. The Royal Academy of Engineering estimates that closing Britain's late-stage funding gap would require an additional $4 billion to $11 billion a year.

More capital is necessary, but it is not sufficient. In capital-intensive technology companies, I have repeatedly seen another gap: the distance between raising enough money to scale and being operationally ready to scale.

💡

A large round does not remove technical, commercial or manufacturing risk. If the capital is deployed in the wrong sequence, it can allow a company to make expensive mistakes faster.

AI introduces another dimension to this problem. As research teams gain new ways to analyse experimental data and explore potential approaches, deciding where to allocate limited R&D capital becomes more difficult.

Deeptech does not scale like SaaS

The familiar software playbook is deceptively simple. Once a company has found product-market fit, new capital can fund more engineers, more salespeople and more marketing. If the economics work, the company tries to repeat what already works, only faster.

Deeptech adds a different set of constraints. Growth may require laboratories, specialised equipment, pilot production, manufacturing capacity, new suppliers, certification and lengthy enterprise sales processes.

The Royal Academy of Engineering describes deeptech companies as requiring significant R&D, capital and time investment, and identifies access to pilot production lines and large-scale manufacturing facilities as a particular challenge when moving from research to commercialisation.

I have seen this firsthand at SPhotonix. Building a laboratory meant coordinating equipment purchases, suppliers, payments and installation, while also helping recruit a technical and scientific team.

💡

Hiring specialists from outside the EU added another layer: permits and approvals had to be handled alongside the technical work. None of this resembles adding another software seat or cloud instance. Each dependency can affect when the company can run experiments and move to the next stage.

A funding round therefore creates commitments as well as options. Hiring can often be slowed or reorganized. Building a production line or buying specialised equipment is much harder to unwind.

A technical decision made today may commit capital for years, while the commercial assumptions behind it may still be unproven.

Founders should resist treating a funding round as permission to scale. Use capital first to remove the next layer of risk.

What goes wrong after the money arrives

One problem is how companies budget. A conventional company budget might allocate money to engineering, sales, marketing and operations. That tells you which department is spending the money. It does not necessarily tell you whether the thing they are spending it on deserves to exist.

For capital-intensive technology businesses, I prefer to look at the product or R&D project as the unit of investment. Imagine that a company is developing three technologies simultaneously.

Engineers may work across all three.

The same laboratory may support all three. Salespeople may be testing several markets. Looking only at departmental budgets can hide the fact that one project is absorbing disproportionate resources without moving closer to either technical validation or a customer.

Fundraising can make the problem worse. Suddenly, the company can afford to hire the team, order the equipment and expand capacity. So it does. But technical progress is not the same as commercial progress.

AI can intensify this problem because it can speed up experimentation without improving the quality of the investment decision behind it. A team may test more ideas or process more data, but that does not mean every experiment deserves funding.

A prototype can perform beautifully and still have no viable market. A customer can praise a demonstration and still refuse to pay for it. A successful laboratory result does not automatically mean the technology can be manufactured reliably or economically.

Technical, commercial and capital risks have to be managed at the same time. McKinsey finds that deeptech ventures typically take longer than conventional technology startups to progress through early funding stages, partly because development often depends on specialized equipment, facilities and hardware as well as software.

The worst response is to keep spending simply because the company has already spent so much.

Give every project its own investment case

A workable approach can remain fairly simple.

For every significant product or R&D initiative, I want to see three things:

A separate project budget. Not simply "£2 million for R&D", but how much capital this particular project is expected to consume over the next stage.
Measurable milestones. These need to include technical and commercial milestones. "Prototype completed" may be a technical milestone; a successful paid pilot, customer commitment or evidence of acceptable production economics may be a commercial one.
Monthly plan-versus-actual analysis. What did we expect to spend? What did we actually spend? More importantly, what did that spending achieve? A project being within budget is not enough if it is three months behind the evidence required to justify its next stage.

AI can make this project-level discipline more useful by letting companies work with a much larger volume of experimental and operational data. In R&D-heavy businesses, models can help teams compare previous test results and decide where additional testing may be worth the cost.

💡

The UK's Advanced Research and Invention Agency is already backing AI Scientist programes that use autonomous systems to generate hypotheses and plan experiments.

That only works if the experimental record is good enough to learn from. If previous tests are incomplete or hard to compare, AI has little reliable evidence to work with. This is why traceability inside the laboratory becomes part of capital discipline as well.

Consider a company deciding between two experimental approaches. One requires additional specialized equipment, while the other can be tested using existing laboratory infrastructure.

If AI-assisted analysis of previous results suggests that the second approach has stronger supporting evidence, management can prioritize a lower-cost experiment before committing capital to new equipment.

At SPhotonix, we have tried to bring software-style traceability into experimental work by recording the stages of an experiment, the settings that changed and the results that followed.

That information can then be analyzed with an internal AI tool to help the team review what has already been tried and where the next experiment might focus.

The important management point is not the AI itself. Experimental spending should leave a usable record of evidence, not a sequence of expensive trials that are difficult to compare.

Finally, decision points must be agreed before reaching them. At each significant milestone, management should have clear evidence-based criteria for deciding whether to continue funding the project or reconsider its direction.

Stopping projects is psychologically difficult.

💡

Once a team has spent a year and a significant amount of capital developing something, another six months can always sound reasonable. Predetermined criteria make the decision less dependent on optimism, internal politics or sunk costs.

Funding should buy down risk

The UK is actively trying to put more capital behind science and technology companies. The British Business Bank says it has now invested more than £600 million directly into more than 50 UK scale-ups and plans to deploy more than £400 million a year into businesses spanning deeptech, life sciences, AI and other sectors.

More public and private capital helps, but it cannot substitute for the ability to allocate it well.

After a funding round, the useful question is: which uncertainty are we paying to remove next?

AI gives deeptech companies new ways to analyse experimental evidence and identify where further research may be justified. But faster experimentation does not automatically lead to better investment decisions.

The real advantage comes when companies use AI-assisted analysis to decide which projects deserve further funding.

Igor Konopliastyi is the Head Of Business Development and CEO of SPhotonix with a demonstrated history of working in the IT industry.