Valuations don’t usually triple in weeks.
But that’s roughly what happened to General Intuition, an AI startup that TechCrunch reports is in talks to raise money at a $6 billion pre-money valuation. Weeks earlier, it closed a $320 million round at a $2.3 billion valuation. New backers reportedly include Valor Equity Partners and Point72. The stated direction for all that money: robotics.
If you’re not steeped in venture math, that jump is the part worth sitting with. A company can grow revenue fast. A company can ship a great product fast. Almost nothing about a company’s fundamentals changes enough in a few weeks to justify nearly tripling what investors think it’s worth. So something else is being priced in, and figuring out what tells you a lot about where AI is heading next.
What a foundation model actually is
General Intuition is building what’s called a foundation model. That term gets thrown around a lot, so let’s make it plain.
A foundation model is a single large model trained on a huge pile of data, meant to be general enough that you can point it at many different tasks afterward. Contrast that with older AI, which was built one job at a time. You’d train one system to spot spam, another to recognize faces, another to sort invoices. Each one was useless outside its lane.
Foundation models flipped that. Train something big on enough data and it picks up patterns broad enough to reuse. That’s why the same language model can draft an email, summarize a contract, and write code. The “foundation” part means everything else gets built on top.
Why robotics is the hard part
Text was the easy win. Language lives on the internet in enormous quantities, already written down, already labeled by virtue of being language. A model can read a staggering amount of it and get good at prediction.
The physical world offers no such convenience. A robot arm reaching for a mug has to understand that the mug has weight, that it might be hot, that liquid inside will slosh, that the handle is the sensible grip point, that a nudge in the wrong direction sends it off the counter. None of that is written down anywhere in usable form. Humans learn it by being toddlers for a few years and knocking things over.
This is the gap people mean when they talk about intuition in AI. Not intelligence in the test-taking sense, but the fast, wordless sense of how stuff behaves. You know a wobbling glass is about to fall before you could explain why. A robot needs that same instinct, and it needs it in milliseconds, because the glass isn’t waiting.
What this means for AI agents
Agent101 readers hear a lot about AI agents, so here’s the connection. An agent is software that takes actions toward a goal rather than just answering questions. Most agents today act inside computers. They browse, they fill in forms, they call other software, they move data around.
A robot is the same idea with a body attached. Same loop: look at the situation, decide what to do, do it, look again. The difference is that a software agent that makes a mistake can usually retry. A physical agent that makes a mistake breaks something, or hurts someone, or falls over.
That raises the bar considerably. It also explains why investors get excited about a company claiming progress on the physical side. If you solve the instinct problem for robots, you’ve probably built something that makes software agents more reliable too. The reverse isn’t true. Being great at text doesn’t teach you anything about gravity.
The part where I stay skeptical
A $6 billion valuation isn’t a product. It’s a bet, made by people who are paid to make bets and who are comfortable being wrong most of the time as long as they’re spectacularly right occasionally.
Valor and Point72 aren’t buying a finished robot. They’re buying a position in a category they expect to matter, at a moment when very little in AI is priced on current results. The $2.3 billion round from weeks earlier looks like the more sober number, and even that was ambitious.
None of this means the technology is hollow. Physical AI is a real problem with real progress behind it, and companies working on it deserve attention. But the valuation and the capability are two separate stories, and the first moves much faster than the second.
For anyone trying to follow AI without a technical background, that’s the useful takeaway. When you see a number like this, resist reading it as proof the technology works. Read it as a measure of how much money is chasing the same idea. Both are worth knowing about. They just answer different questions.
The robots, meanwhile, are still learning not to drop the mug.
🕒 Published: