Refiant AI Raises $5M Seed for AI Model Compression

In Short

Refiant AI has raised a $5 million seed round to build software that makes AI models smaller and cheaper to run.

The startup says model compression can reduce energy use without sacrificing performance.

What Happened

Refiant AI, a South Africa-founded company, announced a $5 million seed funding round led by California-based climate tech investor VoLo Earth Ventures.

Founded in 2025 by Viroshan Naicker, Siddharth Gutta, and Mathew Haswell, Refiant AI builds tools that compress and restructure AI models. In plain terms, it helps “shrink” large AI systems so they need less computing power to run.

The company says it compressed a 120 billion parameter open-source model to run on a MacBook Pro with 12GB of RAM. Parameters are the adjustable “knobs” inside a model, more parameters usually mean a bigger and more expensive model.

Refiant AI also says it has a team that includes a former Google Cloud architect, a Cambridge PhD researcher, and an engineer with NASA experience.

Why It Matters

AI is becoming more energy-hungry, mainly because of large language models and other generative AI workloads. Tech In Africa cited projections that global data centre power consumption could double by 2028.

For African companies, the cost problem is often more immediate than scale. Limited local data centre capacity and dependence on foreign cloud providers can make advanced AI hard to deploy, especially for teams that want data to stay within a country or region.

If Refiant AI can reliably run strong models on smaller devices or local servers, it could support “on-device” and “on-premise” AI. That means AI that runs on a laptop or a company’s own machines, not only in distant cloud data centres.

It could also help enterprises lower compute bills, while improving resilience in markets where power and connectivity are uneven.