ARCVEREX

ARCVEREX builds reinforcement-learning environments that turn general LLMs into competent hardware engineers. The environments are graded, sandboxed tasks drawn from real chip-design workflows — reading specs, writing RTL, building verification stimulus and closing the design loop.

AI labs are racing to specialise LLMs for high-value technical domains, but hardware design is throttled by scarce, low-quality training data. ARCVEREX targets that bottleneck directly, at a moment when demand for chip-design talent far outstrips supply.

The company, founded in 2026 and based in San Francisco, was started to tackle the data-scarcity problem head-on rather than hoping general models would absorb hardware skills by osmosis. Its stated bet is that the right training environments — not more scraped code — are what teach models proper engineering methodology.

ARCVEREX is led by founder Ashish Tondwalkar, and the company is deliberately positioning as an infrastructure partner to AI labs rather than a competing model builder. That focus — environments over models — is a sharp, defensible wedge if executed well.