Fastino Raises $17.5M in Funding from Khosla to Train AI Models Using Budget-Friendly Gaming GPUs

While major tech companies showcase their trillion-parameter AI models that depend on vast and expensive GPU clusters, Fastino is taking a distinct direction.

Based in Palo Alto, the startup claims to have created a unique AI model architecture that is intentionally compact and specifically designed for targeted tasks. Fastino asserts that these models are so efficient they can be trained using budget-friendly gaming GPUs, costing less than $100,000.

This strategy is attracting attention, as Fastino has secured $17.5 million in seed funding, led by Khosla Ventures, famously known as OpenAI’s first venture investor, as reported exclusively by TechCrunch.

With this investment, the total funding for the startup has nearly reached $25 million, following a $7 million pre-seed round last November, spearheaded by Microsoft’s investment arm M12 and Insight Partners.

“Our models are quicker, more precise, and significantly more economical to train, all while surpassing leading models in specific applications,” says Ash Lewis, CEO and co-founder of Fastino.

Fastino provides a variety of compact models to enterprise clients, each crafted for particular tasks like redacting sensitive data or summarizing business documents.

Though Fastino has yet to disclose preliminary metrics or user details, it claims to have made a strong impression on early adopters. Lewis mentioned to TechCrunch that the compact size of the models enables them to deliver comprehensive responses in a single token, offering detailed answers within milliseconds.

It remains to be seen whether Fastino’s approach will achieve widespread acceptance. The enterprise AI landscape is crowded, with companies like Cohere and Databricks also championing AI systems that excel in specific areas. Additionally, smaller model creators focused on enterprise applications, such as Anthropic and Mistral, have emerged. It is widely believed that the future of generative AI in the enterprise space will likely hinge on smaller, more specialized language models.

While only time will tell, an early endorsement from Khosla is certainly advantageous. For now, Fastino is focused on building a leading AI team, aiming to attract researchers from top AI labs who are not solely driven by the ambition to create the largest models or exceed existing benchmarks.

“Our hiring strategy is geared towards attracting researchers who have a contrarian perspective on current language model development,” Lewis notes.