Fastino raises $17.5M to develop efficient task-specific AI models using low-cost GPUs

Fastino, a Palo Alto-based startup, has secured $17.5 million in seed funding led by Khosla Ventures to advance its innovative approach to artificial intelligence. Rather than relying on massive, expensive GPU clusters, Fastino develops small, task-specific AI models that can be trained on affordable gaming GPUs valued at under $100,000 total. This strategy enables significantly lower training costs while delivering high performance on specialized enterprise tasks such as data redaction and document summarization.

The company, which previously raised $7 million in a pre-seed round from Microsoft’s M12 and Insight Partners, now has nearly $25 million in total funding. CEO Ash Lewis emphasizes that their models are not only faster and more accurate but can also respond in milliseconds, often in a single token. This allows for practical real-time applications in enterprise environments.

While the enterprise AI space is increasingly competitive—with players like Cohere, Databricks, and Anthropic also offering task-optimized models—Fastino differentiates itself by focusing on small model architecture and unconventional AI talent. The startup is now concentrating on expanding its technical team by recruiting researchers who challenge mainstream trends in language model development. Despite not sharing user metrics yet, Fastino claims strong early feedback and is positioning itself as a disruptor in cost-effective enterprise AI.