Fastino Raises $17.5M to Scale Compact AI Models for Enterprise
While much of the AI world continues to pursue ever-larger models powered by vast GPU clusters, Fastino is charting a more streamlined path. The Palo Alto-based startup has created a new class of AI model architecture that is intentionally small, highly specialised, and hardware-efficient.
These models are trained on consumer-grade GPUs costing less than $100,000 in total, offering an alternative to the resource-heavy approach taken by many in the industry.
Fastino recently closed a $17.5 million seed funding round led by Khosla Ventures, bringing its total funding to nearly $25 million. This follows a previous pre-seed round that included backing from major venture firms.
Also read: ServiceNow and NVIDIA Reveal Apriel Nemotron 15B to Advance Enterprise AI
Enterprise Tools Built for Precision
Fastino develops and sells task-specific models tailored to enterprise needs—ranging from document summarisation to redacting sensitive information. Each model is designed to perform one function exceptionally well rather than providing general-purpose capabilities.
The startup says its compact models not only reduce the cost of training and deployment but also deliver faster results, often responding in milliseconds. Because of their narrow focus, these tools are said to achieve higher accuracy in specialised domains when compared to larger, general-purpose models.
Contrarian Vision for AI's Future
Fastino appears to be betting on a shift in enterprise demand toward models that are lean, fast, and effective for distinct tasks. As the generative AI landscape matures, there is growing interest in smaller models that offer operational efficiency without sacrificing performance. Rather than competing on size, Fastino is positioning itself as a provider of smarter infrastructure and task-optimised intelligence.
Looking ahead, the company plans to invest in team growth and further research, with a focus on attracting talent that aligns with its belief in precision over scale.
