Ex-Meta policy VP joins board as Worldmodeldata raises £7M seed. Can it hit 1M hours of data before its rivals do? — TFN

• Worldmodeldata closed its £7 million seed spherical final December, however the Cambridge startup waited seven months earlier than asserting it publicly this week.
• The corporate goals to gather a million hours of gameplay information by the top of 2026, however thus far, it has no finalised buyer contracts and a group of about ten folks, together with advisors and contractors.
• Worldmodeldata is becoming a member of a fast-changing business. Since its fundraising, opponents Origin Lab and Basic Instinct have raised $8 million and $454 million for comparable gaming information tasks.

Worldmodeldata, a Cambridge-based deeptech startup, goals to construct the biggest library of AI coaching information from video video games, concentrating on 1 million hours by the top of 2026. Up to now, it hasn’t made any gross sales.

The UK startup simply introduced its £7 million seed, led by Iona Star Capital. Lord Richard Allan, Meta’s former vice chairman of public coverage for Europe, the Center East and Africa, joined as non-executive chairman.

The spherical truly closed in December 2025. Founder and CEO Rhea Loucas tells Tech Funding Information, “We determined to remain nonetheless for a bit bit.” She stated the corporate used this time to find out which information was wanted for AI coaching earlier than launch.

Gerry Buggy, co-founder of Iona Star, provides, “That coupling between motion and consequence is the scarcest useful resource in AI at the moment. Worldmodeldata is manufacturing that lacking ingredient, and each firm constructing bodily AI or digital AI will finally face the identical problem.”

Loucas agreed, saying, “It’s not laborious, I might say, in comparison with others. We didn’t do a roadshow. After we talked to our present investor, they noticed this, and it was carried out in a short time.”

The story behind Worldmodeldata

Loucas spent 4 to 5 years within the video video games business earlier than beginning the corporate. The thought got here from seeing massive language fashions hit their limits in late 2024.

“Many capabilities we’re missing, like frequent sense and causality understanding, don’t come from pure textual content token understanding,” she says.

World fashions, that are AI techniques educated to foretell how environments reply to actions reasonably than simply reacting to inputs, require a unique form of information: recorded actions and their outcomes. Loucas notes that this information is uncommon exterior video video games, the place billions of gamers naturally create a file of actions, video frames, and engine states.

The corporate revealed a paper on its idea in August 2025, was based quickly after, and closed its seed spherical just some months later.

Competitors is a key challenge

Seven months later, the business has modified rather a lot. San Francisco’s Origin Lab raised an $8 million in a seed spherical in Could for the same thought, licensing gameplay information from studios to promote to AI labs. It says it already has partnerships with over 20 publishers. New York’s General Intuition has raised $454 million across two rounds in lower than a yr, backed by its Medal platform, which has tens of tens of millions of month-to-month customers.

Loucas factors out a key distinction. “Basic Instinct is a spin-off from Medal AI. They hold all their information in-house, prepare their very own AI fashions on it, after which promote the fashions. They’re competing with the likes of OpenAI or Anthropic. We’re information suppliers, extra like Scale AI. Our job is to offer high-quality, video game-generated information to the labs that want it. Basic Instinct doesn’t present its information to anybody,” she explains.

This distinction issues for anybody making an attempt to know the busy AI infrastructure sector, which TFN has lined in reviews on Odyssey’s $310 million round and Encord’s $60 million Series C for bodily AI information infrastructure.

Nonetheless, the primary query is: who’s paying Worldmodeldata for its information?

The worldwide artificial and licensed coaching information market was value an estimated $710 million in 2026 and is projected to achieve $3.67 billion by 2031, at a 39% compound annual development fee, based on Mordor Intelligence.

No contracts but, a small group, and one huge aim

Loucas says the corporate has a buyer pipeline and is in shut dialogue with one of many world’s main labs, however no contracts have been finalised, and he or she didn’t share any names.

The corporate has not but began incomes income. The group is about ten folks, together with workers, advisors, and contractors, and a lot of the new funding will go towards hiring.

This reveals a spot between the general public aim of 1 million hours, which Loucas says is 25 occasions larger than the present largest dataset at 40,000 hours, although this quantity comes from the corporate and isn’t independently verified, and the corporate’s present scenario: no finalised contracts, a ten-person group, and a seed spherical that’s already seven months previous, whereas opponents have raised rather more in much less time.

Loucas calls the one million-hour aim deliberate positioning: “The a million is form of our north star. That claims rather a lot.” Whether or not this implies speedy development or simply huge ambitions will grow to be clearer within the subsequent few quarters.

The corporate will keep based mostly within the UK and Europe by selection. Cambridge was picked on goal, and Loucas believes the UK’s video video games business provides the suitable components. The corporate might serve international clients as soon as the product is prepared, however Loucas declined to touch upon plans for a 2026 Collection A.