AI Mega-Rounds Chase Agents, Memory, and Real-World Robots - June 26, 2026

A $2.3 billion bet on video games a $200 million seed before there's a product and a debut fund that closed a third over target - all on one tape. This is Startup Fundraising. Today: who actually set the terms on these mega-rounds, and whether the cap tables underneath hold up the headlines.

And one round I'll let you talk me into - but first, the most circular pitch of the week. Let's go. Hit follow and you won't have to come looking for the next episode. Rebecca Bellan, writing in Rebecca Bellan: An agentic model that can generalize from gameplay to simulation to embodiment is General Intuition’s raison d’être.

And that model’s ability to figure out its place in the world has secured the backing of some heavy hitters. On Thursday, General Intuition said it raised $320 million at a $2.3 billion valuation, confirming TechCrunch’s previous reporting. $2.

3 billion for General Intuition per Rebecca Bellan at TechCrunch - the pitch is that an agent that's played Fortnite-style sims for a hundred hours straight can drive a real quadruped robot with the same brain. And in the demo, the robot bumped into chair legs with its one eye. So the brain that conquered Fortnite is currently losing to office furniture. The number is what makes me sit up.

$2.3 billion is enormous for one bet - and if there was an earlier round on that cap table those early checks just got crushed against the biggest number we've seen all week. Here's what I want pinned: running Fortnite-scale simulation nonstop isn't free. What's the monthly burn on that compute, and does $2.

3 billion disappear in 18 months before one real robot deployment proves any of it? Right - sim-to-real is the whole game, and a camera-only, bug-like bot clipping furniture isn't there yet. Valuation aside, the terms underneath this are what tell you whether anyone believes the timeline. Unite.

AI writes: Former Anthropic researchers have launched Mirendil with an ambitious objective: developing artificial intelligence systems capable of accelerating AI research itself. The San Francisco-based startup has emerged from stealth with a$200 million seed funding round at a $1 billion valuation one of the largest seed financings ever announced by an AI startup. Mirendil - $200 million seed, $1 billion valuation, and per Unite.AI, one of the largest seed rounds ever announced by an AI startup.

But the line that matters to me is a16z and Kleiner co-leading. All week the big names showed up as participants. This time, Andreessen is actually setting the terms. Setting terms on what, Hope?

There's no product. Two former Anthropic researchers, a pitch to build AI that makes AI better - that's the most circular thing I've heard all week. It's pedigree priced at a billion dollars. NVIDIA's in the round too, which tells you who benefits if the compute thesis is real.

So here's my actual question - not a rant. What does Andreessen Horowitz need to see in two years to not write this down? Because right now they're underwriting two resumes and a sentence about democratizing frontier research. From The Next Web: The Berlin-and-New-York startup has raised $110m in a Series C round.

Growth Equity at Goldman Sachs Alternatives led it. Balderton Capital, Index Ventures, Tiger Global, Y Combinator and Dig Ventures all joined. The deal brings Taktile’s total funding to $184m. The company declined to share its valuation.

After the $2.3 billion game-training thesis and a $200 million seed for software that doesn't exist yet this is the one with an actual customer problem. Banks spend billions having people screen risky transactions, and Taktile wants AI agents to make the call. And here the lead is clean - Growth Equity at Goldman Sachs Alternatives, $110 million Series C.

A bank arm writing the check into the layer it could theoretically build in-house. Right, but the pitch is narrow on purpose - they don't want to be ChatGPT, they want to be the trusted layer on top of the big labs. Fine. So when the AI screens a transaction wrong and the bill's in the millions, who eats it - the bank, or Taktile?

$184 million total funding, and they declined to share the valuation. For a Series C this size, that silence tells you something - either the number's gaudy or it didn't move the way they'd like. Here's AI Insiders: Engram, which emerged from stealth Monday, disclosed the raise to CNBC alongside a roughly $600 million valuation. General Catalyst Kleiner Perkins and Sequoia backed the round with participation from Andrej Karpathy the former Tesla AI director and OpenAI founding member whose personal investments have become a closely watched signal in the AI infrastructure space.

Engram came out of stealth Monday: $98 million, roughly a $600 million valuation. General Catalyst Kleiner Sequoia - and Andrej Karpathy writing a personal check which in AI infrastructure is its own little signal flare. Thirteen people. Do the math - that's about $7.

5 million per head before anyone's shipped a thing you can independently verify. And the headline claim is 100x fewer tokens to match frontier models. Beautiful number. There's just no benchmark from anyone who doesn't have equity in it.

Right, the whole pitch is, 'we separate memory from reasoning so you stop re-billing every token.' Fine. Show me one third-party run. Karpathy's name is not a benchmark.

Here's Fund Momentum: Deep33 a US-headquartered Israel-rooted deep-tech firm founded at the start of 2026 by investor Lior Prosor and entrepreneur Michael Broukhim has closed its debut fund at $200M roughly a third above its $150M target. The fund backs the unglamorous capital-intensive layer beneath the AI boom: power compute quantum advanced materials and the physical infrastructure that models actually run on. After a tape full of pedigree and slide decks this one's the picks-and-shovels play - Deep33 closing $200 million for power compute materials.

The unglamorous layer everything else we covered today actually runs on. And credit to Fund Momentum for the useful detail - they targeted $150 million and closed a third above it. Most first-time deep-tech managers can't get to a first close, let alone beat target by 33%. The part I keep chewing on: six months ago, Israeli institutions called hardware too complex, too risky.

Now they're writing the checks. Either the thesis got better, or the AI-software story finally scared them into the real economy. It's a clean counter to the 80-percent-to-US-firms number we've been hammering - US-headquartered Israel-rooted with more than half the LP base from overseas. Non-US money found a US wrapper and still got to set its own terms.

If you're tracking where startup capital is flowing next check out The Data Center Daily - a daily briefing on AI compute hyperscaler capex power semiconductors and energy markets reshaped by AI at scale. Find it wherever you listen to podcasts. We've put links to all of today's stories in the show notes, so if one of them deserves a closer look, you can head there after the episode. That's Startup Fundraising for today.

This is a Lantern Podcast.