Hardik Kabaria on AI for Hardware Design and the Foundation Model for Physics

What this conversation covers

• Why foundation models for physics are different from LLMs
• Why deterministic answers matter in engineering workflows
• Vinci’s work on heat transfer in semiconductors and electronics
• Thermal bottlenecks, throttling, warpage, and yield risk
• What it means to solve hundreds of millions of degrees of freedom in seconds and trillions in hours
• Why GPU infrastructure changes what is computationally possible
• Why Vinci deploys behind customer firewalls
• Why customer benchmarks matter more than vendor demos
• Whether general-purpose physics AI is realistic today or still years away

Why this matters

The bottleneck in engineering is not just model quality. It is the ability to compute trusted physics fast enough, often enough, and at sufficient fidelity to affect real decisions.

Across semiconductors, electronics, batteries, and data centers, teams are working against physical limits that directly shape product performance, reliability, and manufacturability. Traditional simulation workflows are too slow, too specialized, and too manual to support the volume of questions that modern engineering teams need answered.

This conversation points to a different model: physics as an always-available computational layer inside the product lifecycle, with determinism and solver-grounded validation as the minimum bar for trust.