The AI data hunger got so big it went shopping in three aisles at once — scraping, video games, and defense. Same appetite, wildly different checks. If you're just joining, the gaming-data-for-world-models story really started when Worldmodeldata came out of stealth on seven million pounds in seed, led by Iona Star Capital. The pitch — aggregate the behavioral data video games generate, hit a million hours of training data by the end of 2026, and reframe games from toy dataset to actual infrastructure. This is Startup Fundraising. Today: a scraper that never took a dime is suddenly a unicorn, a $320M bet on Grand Theft Auto teaching robots — and a half-year VC number that explains all of it. Oxylabs first. This one's from SiliconANGLE:
Data scraping startup Oxylabs UAB has broken into unicorn territory after raising $130 million in funding from the private equity firm Warburg Pincus LLC. The round is the first time the bootstrapped company has ever sought outside funding, and lifts its valuation to a cool $3.6 billion, it said today.
Oxylabs, out of Lithuania, $130 million from Warburg Pincus — and here's the part that reframes the whole thing. Eleven years bootstrapped, profitable, and this is the first outside check they've ever taken. So there's no prior round to read this against. No preference stack, no dilution runway. First money in comes as growth equity, straight to a $3.6 billion valuation. Credit to SiliconANGLE on the details. And it's Warburg — PE, not VC. That changes how you read it. Nobody writes a $130 million check into a profitable eleven-year-old business hoping for a Series B markup. They're engineering an exit. So what's the timeline, and who's the buyer for a web-scraping business at three-and-a-half billion? That's what I want on the whiteboard. And the pitch has quietly upgraded — 'premium proxy service' in 2015, now it's a 'web intelligence platform' feeding the frontier-model data pipeline. Same billions of requests, better adjective. This one's from Silicon Canals:
General Intuition, a startup building what it describes as a foundation model for embodied AI, has raised $320 million at a $2.3 billion valuation on the thesis that robotics is approaching the same inflection point language AI crossed with GPT-3. The company’s approach — training on millions of hours of video game data rather than real-world robot telemetry — was detailed by TechCrunch this week.
$320 million on the idea that Grand Theft Auto trains your robot better than an actual robot. I love it, genuinely, I want it to be true. Silicon Canals literally called the thesis 'sounds absurd' in the headline — and the check came anyway. The gaming-data-for-world-models idea just got a $2.3 billion price tag stapled to it. Here's my problem — a foundation model that hasn't shipped to a single paying robotics customer, priced at $2.3 billion. What revenue multiple is that? There is no revenue. It's a multiple on a sentence. The pitch is: the product is the generalization itself — one base model instead of everyone collecting their own robot telemetry. And they're not alone; Worldmodeldata's coming out of stealth chasing the same game-data substrate. Maria Deutscher, over at SiliconANGLE, has the details. Kraken. Autonomous ships, $175M, billion-dollar tag. And this is the one that makes me sit up, because maritime autonomy is capital-heavy, regulation-heavy, long-cycle — everything the AI infrastructure crowd is running away from right now. Right — after the General Intuition round we just hit, $320M on synthetic robot data, this reads almost quaint. Actual steel in actual water. So who's the paying customer? A single autonomous voyage — what does that cost, what does it save, and who's signing a check for it today, not in a demo? And at a billion, the terms matter more than the number. Hardware-heavy companies burn cash between raises — I'd want to see how much of this $175M is runway versus fleet. SiliconANGLE, with Duncan Riley:
U.S. venture capital deal value hit $412.7 billion in the first half of 2026, nearly 30% more than investors put to work in all of last year — and a small cluster of giant artificial intelligence rounds accounted for almost the entire jump. That’s according to the second-quarter PitchBook-NVCA Venture Monitor report released Wednesday night.
$412.7 billion deployed in six months — nearly 30% more than all of last year. And AI took $355.9 billion of it, 86 cents of every venture dollar. Eighty-six percent. AI has swallowed the venture market. Everything else is a rounding error. And the money's stratified. Rounds of $100 million-plus took 87.5% of all capital. The sub-$100 million deals — still the bulk by count — split $51.4 billion among them. Their share went from 43.8% in 2024, to 33.1% last year, to 12.5% now. So the little guys keep doing most of the deals and get a smaller slice every year. Seven billion-dollar rounds in Q2 alone — Anthropic, Anduril, Cognition, the usual — worth $87 billion together. The Oxylabs check we just talked about? A profitable business taking $130 million, and it doesn't even register at this altitude. PitchBook's calling it structural, not cyclical — cheaper AI coding tools, foundation models as a base layer. That's them saying: don't wait for this to revert. From Nicholas Pfosi at CTech:
The rapid proliferation of inexpensive attack drones is reshaping the economics of modern warfare, prompting investors to back startups seeking alternatives to conventional air-defense systems. Skapion, a defense technology company developing systems to counter coordinated drone attacks, has raised a $36 million Seed round as militaries increasingly confront the challenge of defending against large numbers of unmanned aerial vehicles simultaneously.
Skapion, $36 million seed, co-led by UP Partners and Khosla — Israeli air-defense veterans building counter-drone-swarm systems. And their pitch is honest, actually: existing air-defense wasn't built for the economics of cheap mass UAV attacks. That's the part I respect. You're shooting a million-dollar interceptor at a thousand-dollar drone — the math loses even when you win. So I want to know what that $36 million actually buys toward flipping the cost curve. And put it next to the Kraken maritime round on the board today — two hard-tech defense bets in the same rundown. Different asset, same reality: long cycles, heavy regulation, real hardware, not a slide deck. Here's what makes this one land for me — the founders ran the actual systems. The pedigree here is operating history, not just a headline metric. I want to see the cost-per-intercept number before I get excited, but I'll take veterans over narrative any day. Have feedback on today’s episode, a fundraising question we should tackle, or a correction? Send us a note at startupfundraising at lantern podcasts dot com. We’d love to hear what would make the show more useful for you.
If you want to dig deeper, we’ve put links to every story from today’s briefing in the show notes. Take a look at the pieces that matter most to your fundraising plans.
That’s Startup Fundraising for today. Have a great Friday, and thanks for listening. This is a Lantern Podcast.