After a week of sovereign anchors and seven co-lead pileups, today's tape is small, European, and — for once — you can actually read the cap tables. This is Startup Fundraising. We've got a Cambridge seed racing a year-end deadline, a Series B that grew up without a new name muscling in, and an oversubscribed defense fund. Adam's already circling. Smaller clothes, same moves. Let's start with Worldmodeldata — because somebody promised a million hours by December. Hit follow and you won't have to come looking for the next episode. BusinessCloud writes:
Worldmodeldata, a Cambridge-based startup building the world’s largest database of video game-generated training data for next-generation AI, has raised £7m in seed funding as it emerges from stealth. The round was led by Iona Star Capital, a London-based venture capital fund focused on early-stage companies operating at the intersection of AI, data and technology.
A Cambridge seed: £7 million, one named lead — Iona Star out of London. After a week of club deals, here's a round you can actually read top to bottom. And they pulled in Lord Richard Allan — Meta's former VP of Public Policy — as chairman. For a company whose whole product is game data scraped into training sets, putting a policy heavyweight on the board this early tells you where they think the fights are. Here's what I want to know about that £7 million: they're promising 1 million hours of training data by the end of 2026. That's a six-month clock. So what's the burn rate that makes it real — is this a lean pipeline or a very confident press release? And it's testable, which I love. December rolls around, and either the million hours exists or it doesn't. Compare that to Tripo pulling $350 million over a month for generative 3D — this is a fraction of the check with a hard number attached. Here's Ventureburn:
Aligned, the AI Deal Workspace where sellers, buyers, and their AI agents run deals together, today announced $60 million in Series B financing, bringing its total funding to $73.8 million. The round was led by PeakSpan Capital, with participation from all existing investors, Hetz Ventures, JAL Ventures and NFX.
$60 million Series B, $73.8 million total — so back that out and the prior rounds add up to roughly $13.8 million. PeakSpan leads, and every existing investor followed, with Hetz, JAL, and NFX all re-upping. That cap table compounded without a new name muscling in to reset the terms. Coming right after Worldmodeldata, it has the same clean-read quality, just one stage later. 70,000 sellers and a million buyers — those are usage numbers, not revenue. But they say ARR tripled over twelve months, so somebody's paying. PeakSpan's a B2B SaaS shop; they know the ARR math cold. My question is whether an 'agentic execution layer' is something Deel and WordPress pay for as a net-new line item, or just a workflow wrapper replacing three tools they were already buying. Deel, SimilarWeb, and WordPress are on the logo wall, and they claim 30% faster cycles. Those are customer-supplied numbers, so grade accordingly. Here's David Cendon Garcia at geoSurge:
geoSurge, the deep-tech AI company helping brands shape how they are represented inside generative AI systems, today announced an oversubscribed $12m Seed round - led by AlbionVC, with participation from Play Ventures, Octopus Ventures, Celero Ventures, Boost Capital, existing investors Passion Capital and Tuesday Capital, alongside angel investors from Google DeepMind, Microsoft AI and Signal AI.
geoSurge, London — a $12 million seed, oversubscribed, AlbionVC leading. Note the currency split in the headlines: dollars in the company release, €10 million in the EU-Startups version. Same round, two framings. The pitch is that they help brands shape how they show up inside ChatGPT and Gemini. So — SEO for the models that ate SEO. What does the money actually buy me as a customer? They say the differentiator is a trademarked methodology — Corpus Engineering — meant to go beyond citation tracking and influence what the models already understand. Founded just over a year ago, so this moved fast. Trademark symbol on the methodology, angels from DeepMind and Microsoft AI on the cap table. Very well-decorated seed. Two years from now, the bet only works if OpenAI and Google don't just close the door on external actors reshaping their corpus — which is exactly what they'd want to do. That's the structural risk — you're building a business on top of platforms that can change the rules overnight. Compare that with the Worldmodeldata seed we hit earlier: single lead, clean thesis, a testable data target. This one has a syndicate and a category that barely existed twelve months ago. From PR Newswire:
Marlinspike Partners, a venture capital firm investing in dual-use technologies with national security and commercial applications, today announced the final close of Marlinspike Disruptive Technology Fund II at $127 million, surpassing its $75 million target.
Marlinspike closed Fund II at $127 million against a $75 million target — 69% oversubscribed. It's a dedicated dual-use, national-security fund, and LPs came in wanting more than the manager asked for. And here's what I actually like about it — after a week of Aramco anchor checks and Beijing-adjacent AI rounds, this is explicitly American defense-adjacent capital. No dancing around it. It says 'Rearm and Rebuild America' right in the headline. Right, no euphemism budget spent there. And it fits the pattern that built all week — sovereign and strategic money showing up across the stack. This is the pure-play version, and it closed hot. The oversubscription is the tell, though. At $127 million for early-stage defense, the fund has to survive long procurement cycles. You need exits that clear a very slow buyer — the Pentagon. That's the math I want to see, not the mission statement. Fair. Enthusiasm at the raise and reality at the exit are two different clocks. Tech.eu's Tamara Djurickovic is tracking this. A €4.5 million pre-seed for optical interconnect. That's deep-tech hardware — photonics — and pre-seed money doesn't buy you a fab. So what does €4.5 million actually get Aylight? Tape-out, a demo chip, or eighteen months of payroll and a slide deck? Optical interconnect is the pick-and-shovel play under every AI data center buildout — moving data between chips faster and cooler than copper. At pre-seed, though, the term sheet matters more than the thesis. I want to see who led and whether the round was clean or a party pile. Right, and photonics has a graveyard of companies with beautiful physics and no path to volume. Two years from now, does this check get them a hyperscaler design win, or just enough proof points to raise again? It fits today's tape — Sherpa's $18 million, this €4.5 million, all European, all readable. After a week of sovereign anchors and club deals, a straight pre-seed in photonics is a number you can actually parse. If you track capital markets beyond startup rounds, try Infrastructure Secondaries Daily. It covers LP stake sales, GP-led continuation vehicles, and discount-to-NAV pricing every day. It's a useful lens on liquidity and investor behavior. Find it wherever you listen to podcasts.
We'll be watching Worldmodeldata's push to generate 1 million hours of training data by the end of 2026. Links to the stories we covered are in the show notes, so if one caught your ear, you can go straight to the source. That's Startup Fundraising for today. This is a Lantern Podcast.