South African AI startup Refiant raises $5 million to cut AI energy and data costs

South African artificial intelligence startup Refiant raised $5 million in seed funding, aiming to reduce the cost and power use of large AI systems. The round was led by VoLo Earth Ventures, a climate-focused investor with $225 million deployed across 35 companies, at a time when major technology groups are expected to spend nearly $700 billion this year building data centers for artificial intelligence. Refiant said the funding will be used to expand its engineering team, build out its platform, and deepen partnerships with corporate customers.

AI efficiency addresses rising energy demand

The startup already counts a former Google Cloud architect, a Cambridge PhD researcher, and an engineer with NASA experience among its early hires, underscoring its focus on technical depth as it seeks to enter a competitive market. The company said it is in discussions with multinational technology firms interested in reducing AI computing costs while keeping data processing closer to home. The push for efficiency is gaining urgency. Companies such as Meta and Microsoft are pouring billions into graphics processing units and cooling systems to support more powerful AI models, while also facing pressure to manage energy use.

Refiant says its software takes a different approach, compressing AI models so they can run on less hardware. The company said it successfully reduced a 120 billion-parameter model to run on a standard laptop, cutting energy requirements by more than 80 percent while maintaining comparable performance. “AI’s growing energy footprint is one of the most urgent and underappreciated challenges in the climate space,” co-founder Siddharth Gutta said. “The industry’s default answer is to build more data centers and consume more power. Ours is to make the AI itself dramatically more efficient.”

AI sustainability coexist efficiency key message

Founded in 2025 by Viroshan Naicker, Siddharth Gutta and Mathew Haswell, Refiant develops tools that restructure AI models by reducing computational weight and retraining them to preserve performance. This allows companies to run advanced systems on smaller machines or within local environments instead of relying on cloud infrastructure. Larger technology firms are pursuing similar strategies. In March, Google introduced TurboQuant, an algorithm that cuts memory requirements, while Refiant targets compression at lower compute levels.

Businesses adopting artificial intelligence face pressure to curb emissions. Refiant’s founders say both goals can align. “AI adoption and sustainability commitments can coexist, but only if the technology becomes more efficient,” Haswell said, adding organizations should not choose between AI deployment and energy targets. The position drew backing from VoLo Earth Ventures. Managing partner Joseph Goodman said AI’s main constraint is energy demand, noting more efficient computing could help meet rising power needs.

Compressed AI for local enterprise use

Refiant said its technology could be particularly useful in markets where computing infrastructure is limited or cloud access is costly. In sectors such as banking, telecommunications, and government services, companies may be able to run AI locally without constant cloud connectivity. For Africa, where roughly 250 data centers account for about 0.6 percent of global capacity, the company says compressed AI models could lower the barrier to adoption while allowing organizations to keep sensitive data closer to home.