Unaudited Interim Results for Q2 and H1 2026

About SuperSeed Capital Limited

SuperSeed exists to back Europe's best B2B SaaS founders at the earliest stages and to help them build great companies. In the short term, our portfolio companies enable their customers to drive revenue growth and efficiency savings using next-generation software and AI. In the long-term, they have an opportunity to create category defining global technology companies. SuperSeed focuses on the fundamentals by helping founders build good companies with strong unit economics and sensible distribution models.

Forward-looking statements

This announcement contains statements that are or may be forward-looking statements. All statements other than statements of historical facts included in this announcement may be forward-looking statements, including statements that relate to the Company's future prospects, developments and strategies. The Company does not accept any responsibility for the accuracy or completeness of any information reported by the press or other media, nor the fairness or appropriateness of any forecasts, views or opinions expressed by the press or other media regarding the Group. The Company makes no representation as to the appropriateness, accuracy, completeness or reliability of any such information or publication.

Forward-looking statements are identified by their use of terms and phrases such as "believe", "targets", "expects", "aim", "anticipate", "projects", "would", "could", "envisage", "estimate", "intend", "may", "plan", "will" or the negative of those, variations or comparable expressions, including references to assumptions. The forward-looking statements in this announcement are based on current expectations and are subject to known and unknown risks and uncertainties that could cause actual results, performance and achievements to differ materially from any results, performance or achievements expressed or implied by such forward-looking statements. Factors that may cause actual results to differ materially from those expressed or implied by such forward looking statements include, but are not limited to, those described in the Risk Management Framework section of the Company's most recent Annual Report. These forward-looking statements are based on numerous assumptions regarding the present and future business strategies of the Group and the environment in which it is and will operate in the future. All subsequent oral or written forward-looking statements attributed to the Company or any persons acting on its behalf are expressly qualified in their entirety by the cautionary statement above. Each forward-looking statement speaks only as at the date of this announcement. Except as required by law, regulatory requirement, the Listing Rules and the Disclosure Guidance and Transparency Rules, neither the Company nor any other party intends to update or revise these forward-looking statements, whether as a result of new information, future events or otherwise.

Investment Manager's Review

Market Commentary

The models are not the hard part

In December, Andrej Karpathy wrote that he did not think the industry had realised anywhere near 10% of the potential of AI models at their present capability. Not their future capability. The ones we already had.

It matters who said it. Karpathy was a founding member of OpenAI, then spent five years running artificial intelligence and Autopilot Vision at Tesla, which is as close as anyone has come to putting machine learning into a million moving objects. He went back to OpenAI, left again, and was running a small education company when he wrote that line, which is to say he was the one person in the argument with nothing to sell. In May he joined Anthropic to work on pretraining.

So the man who has spent a decade at the front of this thinks we are only using a fraction of what already exists.

The industry appears to agree

Four announcements in eight weeks this summer. Anthropic set up a services joint venture with Blackstone, Hellman & Friedman and Goldman Sachs. OpenAI created a majority-owned deployment subsidiary with more than $4 billion behind it. Amazon committed a billion dollars to embedding engineers with customers. Microsoft committed $2.5 billion and six thousand people to the same idea.

There are two honest reasons to do this and both are in play. One is that you reach customers who cannot otherwise figure out how to implement your shiny new AI. The other is that sitting inside a working business is how you find out what your own technology can actually do.

OpenAI puts the second one in its job adverts, which ask forward deployed engineers to produce "eval-driven feedback that changes product and model roadmaps".

How new this is deserves a moment. Anthropic advertised for its first forward deployed engineering manager four days before I wrote this.

Why services are back

For decades, venture investors "just knew" what good looked like. Intellectual property heavy, people light. Services were low margin and awkward to scale, and the best technology was the kind you could package and sell with as little labour wrapped around it as possible. So it is worth asking why, with arguably the greatest technology of all time, services are suddenly back in fashion.

The answer is thirty-six years old. In 1990 two economists, Wesley Cohen and Daniel Levinthal, described what they called absorptive capacity: an organisation's ability to recognise the value of new information, assimilate it and apply it commercially. That ability rests on what a business already knows, so the capacity to take up something new is built out of everything it took up before. They had a warning for fastmoving fields too. Once a firm stops investing in that capacity, they wrote, it "may never assimilate and exploit new information in that field, regardless of the value of that information."

A business cannot buy its own absorptive capacity. It can rent somebody else's, which is precisely the business Anthropic, OpenAI, Amazon and Microsoft all entered this summer. What a deployment team sells is not labour by the hour. It is understanding accumulated somewhere else, applied to your problem, while you slowly build your own.

The Investment Manager wrote about this in its third-quarter letter last year, when the obstacles to enterprise AI were being reported as data quality, technical maturity and skills shortages. Those were symptoms.

The reason the work is so large is that none of it can be specified in advance. The tasks now being automated are the ones nobody ever wrote down, so there is no document describing the end state and no way to scope the job from a desk. You find out what is possible by trying it, watching it fail in a way nobody predicted, and carrying that into the next attempt.

Which is Karpathy's ten per cent, seen from the other side. The models are ahead of us, and the gap does not close on its own.

Whose understanding is worth renting

The Investment Manager ran into this themselves this year. They built autonomous pipelines that write code, check it, test it and deploy it, and they worked well enough that the checking stages raised corrections faster than the pipeline could absorb them. The whole thing seized. Nothing about the models was the problem. But it was only by putting them into production that they could see where the system broke at scale. There was no way to find that out except by running it.

Multiply that by twenty years and you have Palantir. Their annual report puts it plainly: "We embed directly with customers across numerous industries, tackling complex challenges while continuously enhancing our platforms' capabilities." Being in the room is how the product gets good.

The numbers that come out of the other end are worth sitting with. In the second quarter Palantir's revenue grew 93% on the same quarter last year. Over the same six months its headcount fell slightly, from 4,429 people to 4,401. Gross margin was 84.7%, and that is struck after paying the people who do the deployment work, because they sit in cost of revenue. You do not grow like that by only selling people.

Mistral is running the same play from the other direction. They used to build their own models. More recently they have been taking Chinese open source models and Europeanising them, retraining on their own corpus of data. This summer they began reselling GLM more or less vanilla. They are going from being a lab to being a hybrid between Palantir and a neo cloud.

That looks like the right move to the Investment Manager. Arthur Mensch understands what this technology can and cannot do better than the chief executive of any integrator. Everyone at Mistral is AI-native, which is true almost nowhere else. European enterprises would like a European partner to help them work this out. Being the people who can run any good model well is turning out to be a great business model in this transition.

Where that leaves us

If understanding is accumulated by doing, the advantage sits with whoever is furthest along on the AI maturity curve inside a particular industry. In physical industries that usually means somebody who started recently. The physical sectors have seen massive capability shifts in the last two years, and much of what is now possible is not driven by language models at all. It is world models and vision-language-action systems. Nobody is far down the track. Everybody is still figuring it out, which is what a genuinely new industry looks like from the inside.

Hive is one of the clearest examples we own. The company puts autonomy into industrial vehicles like forklifts, wheel loaders and diggers, so that one operator can run a fleet from a single station. Its founder has worked on autonomy in industrial systems for more than a decade, and that matters far more than it would have five years ago, because he knows where this breaks in a working yard, in the rain, next to people. The same pattern runs through OctaiPipe in industrial edge deployment and All3 in construction.

These companies don't just win by having good models. They win by understanding the operation and the technology at once, and being able to apply the AI in just the right way to solve customer problems.

So we back companies that are native to this technology and native to one industry, and we push our founders hard on wrapping services around what they sell. Not to turn them into consultancies. In enterprise you rarely sell a product on its own, you sell a working result, and getting the customer there is part of what you are selling. Distribution is destiny, and right now distribution means being the people who can make it work.

One last thing worth noting. Industrials now trade at a higher forward multiple than technology, 21.5 times against 17.7, with industrials above their own twenty-year range and technology back in line with its average. The companies that buy applied AI have been repriced above the companies that build it. So much potential as we continue the transition to the world of physical AI.

The Fund in Q2 2026

Net performance for the Fund's private investors as of 30 June 2026:

·   Net IRR: 20.25%

·   TVPI: 1.47x

·   DPI: 0.11x

The Fund's portfolio, as it stands

The Fund is fully deployed. The final Fund investment closed in January 2026 and the Fund now holds 24-active companies alongside its earlier exits.

Looking forward

Several companies are positioning for next-stage rounds over the next twelve-months.

As is always the case with a portfolio of this size, there is corporate M&A interest in a number of the Fund's portfolio companies. That said, the Investment Manager does not expect any material exits in the next quarter, but things can change quickly.

Other than that, the Fund's portfolio work continues, and the Investment Manager is genuinely positive about the Fund portfolio's potential as it looks to the remainder of 2026.