Hardik Kabaria is the co-founder and CEO of Vinci, which makes physics analysis fast and accessible to anyone designing hardware. It starts with thermal and thermo-mechanical simulation for semiconductor and electronics companies. His background is in physics and geometry software: 12+ years in the space, a PhD, and other startups before starting Vinci about two and a half years before the episode. He joined us in Episode 50 (July 10, 2026).
His one-line pitch, from the top of the episode: today you don't need to be an AI researcher to get language reasoning, you just open ChatGPT or Anthropic's models and ask. Vinci is building the same thing for physics.
For decades, Hardik said at 06:19, hardware teams have checked physical performance two ways: simulation tools from vendors like Cadence, Siemens and Synopsys, and physical experiments in the lab. Both are slow. A semiconductor tape-out can cost millions of dollars and take months, and the people who design a part, test its physics and check manufacturability often sit on different teams.
Vinci puts a physics model into that loop. An engineer drags in a design file, asks a question such as how a part will heat up, and gets an answer in seconds or minutes. Its customers today are tier-one semiconductor, electronics and foundry-equipment companies, from memory to fabless to foundry. Customers are sensitive, so he named none.
At 03:15 Sam played a Vinci demo video on warpage in semiconductor packaging. Warpage decides whether a package ships: if it falls outside tolerance, measured in microns, bonds don't form and there is no electrical connection. Physical experiments take months per iteration, and simplified simulations throw away the geometry that causes real warpage.
The demo ran the native design, every layer, metal line and via, at three-micron resolution and a 225°C bonding temperature. The result was 392 microns of warpage, and the point was its shape: it followed the actual layout, metal distribution and residual stress. At 05:29 the video gave the numbers: 1.2 billion degrees of freedom at manufacturing resolution in under four minutes.
Now instead that engineer, she might be performing 10 to 20 analysis per day, exploring different configurations that came to her from design. With us, she's able to do thousands.
— Hardik Kabaria, Episode 50
And of course, now that you are producing gigabytes of data, humans don't make decisions on gigabytes of data, we make decisions on two by two charts.
— Hardik Kabaria, Episode 50
So it's a half a billion parameter model, We have trained it ground up. So we are the pre training team and we are the product team.
— Hardik Kabaria, Episode 50
Sam tested Meta Muse Image, which he found similar to Nano Banana, and even asked it for a diagram of how Vinci works. Hardik expects images to follow the path language models took: raw capability first, then guardrails and trust. For AI-generated physics and designs that may drive robots, he said, the question of who generated a design matters even more. For more on chips and hardware from the show, see AI in semiconductors and AI in manufacturing and hardware.
Hardik Kabaria is the co-founder and CEO of Vinci. His background is in physics and geometry software; he has spent 12+ years in the field, including a PhD and other startups, and started Vinci about two and a half years before his July 2026 appearance on Built This Week.
Vinci makes physics analysis, starting with thermal and thermo-mechanical, accessible to anyone designing hardware. It started with semiconductor and electronics customers and returns answers in seconds or minutes instead of a day.
In the demo, Vinci solved a semiconductor package at three-micron manufacturing resolution, 1.2 billion degrees of freedom, on the native design in under four minutes.
Yes. Hardik said Vinci deploys behind the customer's firewall, so their IP stays inside and they don't have to worry about data leaking.
Vinci trained its own half-billion-parameter physics model from the ground up for thermal and thermo-mechanical phenomena. Hardik compared its capability level to GPT-2 and said the team does both pre-training and product.
