Nvidia’s Les Karpas to address robotics’ AI gap

At TechCrunch Disrupt 2026, Nvidia Inception’s physical AI lead will discuss why robotics still lacks language AI’s data scale.

At TechCrunch Disrupt 2026, Nvidia Inception’s Global Head of Physical AI, Les Karpas, will take on a question that has hovered over robotics for years: why the field still has not had its own ChatGPT moment. His session is set to look at both the limits slowing robotics down and the methods companies are using to move past them. That matters beyond a single conference, because the gap between digital AI and machines in the physical world shapes how robots are built, trained, and deployed.

The talk also reflects Nvidia’s growing focus on robotics, an area CEO Jensen Huang has highlighted in recent keynotes. At Disrupt, Karpas is expected to lay out the main bottlenecks in physical AI — and the opportunities that could open up if those barriers are reduced.

Robotics still lacks the data scale that lifted language AI

The problem, as Karpas frames it, is simple to describe and hard to solve. Large language models were able to draw on enormous internet-scale datasets, helping systems from OpenAI, Anthropic, and others move from research into everyday use.

Robotics does not have an equivalent source of data. General-purpose robots need information about the physical world, but there is no internet-wide dataset that gives them training at the scale language models received. Progress, in turn, is slower and more complex.

The source points to self-driving vehicles as a partial comparison. Companies such as Waymo built their datasets over years of road miles, and those datasets keep expanding market by market. Robotics, by contrast, still lacks that same long-term accumulation of real-world data, as IT-PUB News notes.

Startups are turning to simulation and synthetic data

Because that dataset does not already exist, more startups are trying to build it through other means. The source says they are using simulation, synthetic data, and foundation models trained across multiple robot types at once.

The goal is to narrow the gap between digital and physical AI. Rather than waiting years for enough real-world experience to build up, companies are trying to create the scale physical AI needs much faster.

That is one reason TechCrunch created the Real World AI Stage for Disrupt. The event will bring together founders from Shield AI, Colossal Biosciences, FieldAI, and Foxglove to talk about the specific challenges and opportunities involved in building AI systems for real-world use.

Nvidia sees the problem across multiple industries

Karpas is positioned as a notable voice on the issue because of his role at Nvidia Inception. As Global Head of Physical AI, he works with startups across robotics, automotive, manufacturing, mobility, smart cities, and other sectors.

That gives him a broad view of the ecosystem trying to solve the same data and training problem from different directions. The companies working to close the physical AI gap are also the ones he regularly engages with through Nvidia’s startup network.

His background reflects the range of experience robotics often demands. The source says he has held roles at Stanley Black & Decker, Intellectual Ventures, Herman Miller, iRobot, and Cirque du Soleil, and has worked as an architect, manufacturing engineer, startup CEO, venture studio executive, and corporate venture capitalist.

Disrupt will pitch the session to founders and investors

TechCrunch is presenting Karpas’ session as a chance to hear directly how Nvidia views the biggest bottleneck in physical AI and what it may take to get past it. For attendees, that could offer practical insight into where robotics remains constrained and where new business opportunities may emerge.

The event is also being promoted as a place for founders and investors to connect. TechCrunch says the lineup will include more than 10,000 tech leaders, founders, and investors at Moscone West in San Francisco from October 13 to 15.

The conference is also tied to ticket sales. TechCrunch says buyers can save up to $200 before September 25, after which prices rise. Group passes are also being offered with savings of up to 30%.

For the wider tech industry, the larger issue is whether robotics can eventually reach the kind of breakthrough language AI saw in 2022. The source does not suggest that moment is near. What Karpas’ session is meant to examine is what still stands in the way — and whether the next step in physical AI depends on better data, better models, or both.