Two robotics companies raised roughly $200 million each in the same stretch of August, and only one of them can point to trucks already running on real customer routes with no one behind the wheel. The other raised its $200 million on the promise that a single model will eventually drive, walk, or grip its way across just about any robot body you hand it. Both bets might turn out fine. But they are not the same bet, and the gap between them is the clearest read yet on what “boring” actually buys a hardware founder right now.
Two hundred million dollars, and only one of them has trucks on the road
Start with Gatik. The self-driving trucking company just closed a $200 million Series D led by the Qatar Investment Authority and Koch Disruptive Technologies, with Millennium Management and ARK Invest also in the round, taking its total funding to roughly $500 million, according to TechCrunch’s report on the raise . The round landed two months after a multiyear PepsiCo deal went live. Gatik now runs 41 fully driverless box trucks, no safety driver on board, moving Frito-Lay product across Dallas, Phoenix, and Northwest Arkansas to roughly 250 retail locations, including Walmart and Dollar General stores. The company says it’s sitting on $600 million in contracted revenue and has completed more than 85,000 driverless orders at a 99% on-time delivery rate.
Every one of those numbers is checkable. Trucks either drove the route or they didn’t. A contract either exists or it doesn’t. That’s the point: Gatik’s fundraising story is built entirely out of facts a skeptical analyst could go verify, not a narrative someone has to take on faith.
Now put Gravis Robotics next to it. The 2022 ETH Zurich spinout raised a $200 million Series A, SoftBank as the sole investor, at a $1 billion valuation, reportedly the largest Series A round in construction-robotics history and ETH Zurich’s fifth unicorn overall, its first in robotics, per Bloomberg’s coverage of the deal . Gravis doesn’t build new autonomous excavators. It retrofits the ones construction companies already own with self-driving hardware and software, and it’s already deployed across four continents, with customers including cement giant Holcim and HD Hyundai. It recently added an $8 million retrofit contract with the UK government.
Gravis’s story is a little less “boring” than Gatik’s, since retrofitting an entire industry’s existing equipment fleet is a genuinely ambitious wedge. But the evidence underneath it is the same kind: real deployments, on real machines, for large industrial buyers who did their own diligence before writing a check. Holcim and HD Hyundai are not the kind of customers that sign on to a pilot because a demo looked impressive. They run capital-intensive operations where an unreliable retrofit costs real money in downtime, so a construction giant putting Gravis’s kit on its own excavators is itself a form of due diligence that a term sheet can’t replicate. That’s also the quieter argument behind SoftBank writing a $1 billion valuation into a Series A: it’s not a bet on a roadmap, it’s a bet on a wedge that’s already proven it can clear a large industrial buyer’s own internal bar.
The other kind of $200 million
Now consider Generalist. Founded in 2024 by ex-DeepMind researchers Pete Florence and Andy Zeng along with ex-Boston Dynamics engineer Andrew Barry, the company reportedly reached a $3 billion valuation through a roughly $200 million extension led by 8VC, layered on top of the $400 million Series B it disclosed in June at a $2 billion valuation, according to TechCrunch’s reporting on the round , reporting that, notably, is still sourced to unnamed people familiar with the deal rather than a company confirmation. The pitch, as described by an 8VC partner, is that a single foundation model can become useful across many different robot bodies without engineers hand-coding each new task. Generalist’s Gen 1.5 model claims to pick up new tasks from video demonstrations as short as three to twelve seconds, an ambitious version of the idea that robotics is about to have its own large-model moment the way language models did.
Here’s the part of the picture that doesn’t get much airtime in a story about a valuation jumping from $2 billion to a reported $3 billion in about two months, on reporting the company itself hasn’t confirmed: in the same window, a joint Stanford and Google DeepMind study ran more than 1,600 real-world robot trials across fourteen axes of generalization and found that today’s best “generalist” manipulation policies (the AI systems that decide how a robot arm should move to grasp or manipulate an object) still struggle with small perturbations. A camera angle shifts a few degrees. An instruction gets reworded. An object moves a few inches from where the model expects it. Performance drops. None of this means Generalist’s underlying research is bad science, and academic benchmarks always lag what a well-funded lab can achieve in the months after a paper is published. But it does mean the gap between how confidently “robotics is having its ChatGPT moment” gets said in a pitch deck and what independent researchers can currently replicate is real, and it’s widening, not closing.
What “boring” actually buys you
None of this is an argument against ambition. Somebody is going to build the foundation model that generalizes across robot bodies, and whoever gets there first captures an enormous prize. The argument is narrower: Gatik and Gravis didn’t need a manifesto to justify a mega-round. They needed a customer running their machines every single day, and they had one. That is a fundamentally cheaper story to tell an investor, because it doesn’t require anyone to extend credit on the strength of a vision. It requires them to read a delivery log.
That distinction matters most for founders who are capital-constrained and don’t have the luxury of a multi-year runway to prove out a generalization thesis before revenue shows up. It’s also worth noting that the biggest checks in deeptech right now, the kind increasingly structured against contracted revenue and physical production capacity rather than a growth narrative , are being written for companies that can already show the same kind of boring, checkable proof Gatik and Gravis have. If your fundraising narrative currently leans on “this changes everything” language ahead of a boring, checkable number, you are borrowing conviction you haven’t earned yet, and eventually somebody runs the 1,600-trial version of your own pitch. The founders who get ahead of that aren’t the ones who avoid ambitious claims. They’re the ones who go find their own version of Gatik’s delivery data or Gravis’s four-continent deployment list before a diligence team asks for it, and who are honest, in the meantime, about which parts of their story are proven and which parts are still a bet.
What to watch next
Gatik and Gravis both have public, dated facts to hold them to going forward: Gatik’s contracted revenue and delivery volume, Gravis’s retrofit contract pipeline. Generalist is a different kind of watch item. Two things would meaningfully change how much weight to put on its valuation: a company-confirmed announcement of the reported $200 million extension and $3 billion mark, since the current reporting is still attributed to unnamed sources, and a public, third-party benchmark of Gen 1.5 against something like the Stanford/DeepMind generalization axes rather than Generalist’s own demo reel. Until one of those shows up, the honest read is that Generalist’s valuation and Gatik’s delivery numbers are being priced by two very different standards of evidence, which is really the same gap between a claim and a claim someone has actually gone and checked that shows up across deeptech right now, and it’s worth knowing which standard is backing whichever robotics story you’re betting on next.
Clement Chen is a startup founder, advisor, and investor who writes about deeptech and building companies at clementchen.co .