The bet that's dividing robotics
There's a quiet but increasingly heated debate shaping where tens of billions of dollars in investment are flowing: when will robotics have its "ChatGPT moment"? The answer from someone who has been inside the world's largest artificial intelligence hardware factory is not optimistic at all — and it throws into question the thesis that has been attracting Wall Street investors to the sector.
Amulya Vishwanath, who spent seven years leading Nvidia's robotics product marketing and is now the founder and managing partner of Techable VC, made her position clear in an interview with TechCrunch's Equity podcast: robotics is far from its ChatGPT moment. According to her, the sector is still in a "Lego era," where the fundamental building blocks — actuators, robotic hands, software and safety standards — are being rebuilt piece by piece rather than integrated into a unified platform.
This statement might seem like a bucket of cold water on enthusiasm, but it carries specific weight. Vishwanath is not an outside observer doing theoretical speculation. She spent seven years inside Nvidia, working alongside Jensen Huang, Dr. Deepu Talla, Ian Buck and key researchers, seeing firsthand which parts of the robotics stack enterprise customers actually struggled to integrate.
What the 'Lego era' means in practice
Vishwanath's analogy is revealing because it describes the current situation of the industry precisely: every robotics startup is building its own solution for problems that, in other technology sectors, have already been solved by standardized platforms. When Nvidia launched CUDA, developers didn't need to reinvent accelerated computation engines — they could focus on what to do with GPUs, not how to make GPUs work.
In robotics, that hasn't happened yet. Efficient actuators are still niche components. Robotic hands with enough dexterity to manipulate varied objects exist only in laboratories and high-cost prototypes. Safety standards for human-robot interaction vary by region and sector, and there is no universal framework that allows any manufacturer to certify its product for use in shared environments.
"Every robotics startup is building its own Lego," Vishwanath explained on the podcast. "Instead of plugging in ready-made modules, each team needs to design its own actuators, its own perception system, its own control software. That makes each solution expensive, slow to iterate on and difficult to scale."
This fragmentation is especially relevant because Nvidia, ironically, is trying to do for robotics exactly what it did with CUDA: create a platform around which the entire ecosystem organizes. Its investments in Halos for Robotics, Isaac Lab and Cosmos are attempts to provide the Lego bricks — simulation, safety, physical simulation — so that startups stop reinventing the wheel.
The debate: general model vs. specialized pieces
Vishwanath's thesis directly challenges an alternative view that has also gained traction among investors. Pim de Witte, CEO of General Intuition, argued on the same Equity podcast that a single general foundation model would render much of today's specialized robotics work obsolete — and would, in fact, be robotics' ChatGPT moment.
The difference between the two views is fundamental and defines where capital should flow. If Vishwanath is right, the most promising startups are the ones solving specific layers of the stack — better actuators, cheaper perception, faster safety certification. If de Witte is right, those specialized bets will become commodity once a general-purpose model matures, the same way that specialized NLP startups got squeezed when large language models matured.
What makes this debate so important is that the industry is betting on both directions simultaneously. Venture funds are financing both general model startups and specialized hardware companies. And that creates a real risk of misallocated capital — money going to the wrong solution for the wrong problem.
Why the ChatGPT moment is different for robotics
The comparison with ChatGPT is inevitable but misleading. When ChatGPT was launched in November 2022, the language model had already existed for years — researchers knew transformers were approaching something special. What happened was a convergence: models large enough, data good enough and a use case anyone could understand and use.
In robotics, the convergence hasn't happened because physical problems are fundamentally harder than language problems. A chatbot can simulate reasoning about an abstract world. A robot needs to manipulate an unpredictable physical world governed by laws of physics that forgive no errors.
This explains why ACE Robotics, a Chinese embodied AI startup, predicts its own "ChatGPT moment" will arrive only by late 2027 — and that another four to five years would be needed for broad commercial deployment. And it explains why Wang Xingxing, founder of Unitree, estimated up to ten years before humanoid robots could have a similar breakthrough.
The investment numbers tell another part of the story. Nvidia's physical AI revenue exceeded US$ 9 billion over the trailing twelve months, up from US$ 6 billion the previous year. Jensen Huang calls humanoid robotics a "multibillion-dollar economic opportunity." But in the same period, Nvidia's total revenue was US$ 215.9 billion — and the line that grouped automotive and robotics (almost entirely autonomous driving) accounted for only US$ 2.3 billion. Robotics is still a footnote in Nvidia's revenue.
Techable VC's strategy: betting on the right pieces
Vishwanath founded Techable VC in 2025, focusing on Physical AI, robotics, materials science and life sciences — a broader scope than most robotics-only funds. The thesis is clear: invest in the specific layers of the stack that are blocking overall progress.
The Techable VC website lists areas such as brain-computer interfaces, neurotechnology, neuromorphic computing, industrial robotics, surgical robotics, soft robotics, sensor stacks, metamaterials and quantum as investment focuses. There is no mention of general-purpose AI models for robotics — which, by itself, is a positioning on where real innovation is. The Berkeley Haas newsroom, where Vishwanath completed her EMBA, reported on the fund's launch, highlighting its eight-company portfolio and Vishwanath's vision of building "civilization systems."
The portfolio already has one name: Alpha Z, a pre-seed field robotics startup in Los Angeles. Vishwanath has not disclosed check sizes or other investments, which makes it impossible to verify whether she is putting her money where her mouth is.
What to watch
The debate between Vishwanath and de Witte reflects a structural divide in how the industry is approaching robotics. There is enough money being allocated to both views for one of them to eventually prove more correct — but the question is how much time and capital will be spent before the dust settles.
If fundamental hardware blocks continue being rebuilt from scratch by each startup, the ChatGPT moment for robotics will take longer to arrive. If, on the other hand, a general foundation model emerges and proves it generalizes well enough for the physical world, the story could be much faster — and companies that bet on specialized pieces may discover their investments have become commodity.
The real question is what will be the catalyst that forces this bifurcation: a hardware breakthrough that makes Lego bricks accessible enough, or a general model breakthrough that makes the bricks irrelevant?
Sources: TechCrunch, Value Add VC, Berkeley Haas
✓ Independent sources cross-checked and verified before publishing
