A new artificial intelligence venture is developing software to run large models directly on consumer devices to eliminate the need for cloud-based processing. The company, called Underdog, announced backing from a16z, Khosla Ventures, and the Anthology fund—which includes partners like Anthropic and Menlo Ventures—as well as investors including Patrick Collison and Naval Ravikant. Underdog aims to provide free and reliable AI services to billions of users by optimizing the co-design of models and inference engines for better power and data efficiency.

The startup claims that the latest frontier-level intelligence will be accessible on local hardware within 6 months of its release, a principle it calls Underdog's Law. To support this, the company is conducting research into confidential inference and agentic commerce to evaluate how shifting AI processing from the cloud to the device will reshape the existing internet economy.