South Africa-founded Refiant AI raises $5 million to build leaner AI models

Refiant AI, a South Africa-founded startup that uses algorithms to compress artificial intelligence (AI) models, has closed a $5 million seed round to build its platform, grow its team and support enterprise partnerships. 

VoLo Earth Ventures, a California climate technology fund, led the round, backing what it considers a more efficient approach to developing smarter and leaner AI models.

Founded in 2025 by Viroshan Naicker, Siddharth Gutta, and Mathew Haswell, Refiant AI is building tools that restructure and compress AI models by reducing computational weight and retraining them to maintain performance, so that they can run efficiently on smaller or local machines.

As global companies, such as Meta and Microsoft, race to deploy more efficient AI models, they are investing heavily in building data centres equipped with graphics processing units (GPUs) and cooling systems. 

That energy infrastructure is both expensive and resource-intensive. In the first quarter of the year, both companies each committed nearly $50 billion in additional data centre leases to support AI. Bloomberg reported that those pledges among the largest cloud computing companies helped push commitments to data centre leases above $700 billion. 

“AI’s growing energy footprint is one of the most urgent and underappreciated challenges in the climate space,” Gutta, one of Refiant AI’s co-founders, said. “The industry’s default answer is to build more data centres and consume more power. Ours is to make the AI itself dramatically more efficient.”

Instead of relying on building more infrastructure, Refiant AI intends to make the AI models lighter so they require fewer resources to operate. The company said it had successfully compressed a 120-billion-parameter model to run on a standard laptop with just 12 gigabytes (GB) of RAM, a task that would typically require at least 80GB of memory. The company said that the compressed model retained between 95% and 99% of its original performance, using more than 80% less energy.

Refiant said its technology is suited for markets where computing infrastructure is limited, and cloud access can be expensive. In sectors like banking, telecoms, and government services, its tools could allow organisations to run AI systems without needing constant access to the cloud. 

Nigeria’s Central Bank recently embedded AI and machine learning into new baseline standards that require financial institutions to deploy automated anti-money laundering systems. Tools like Refiant’s could enable banks to deploy these systems locally without relying on costly foreign infrastructure or moving sensitive data across borders.

The implications of compressed AI models vary. If powerful AI models can run on everyday hardware, African organisations may rely less on global cloud providers and expensive infrastructure. For the continent whose 250 data centres only account for 0.6% of the global capacity, the AI compression technology could lower the barrier to adopting advanced AI models while keeping data local.

At the same time, it introduces a new tension. Local data centre operators across Africa have been expanding their capacity to meet the rising demand for cloud services, estimated to reach $25.46 billion by 2029. In July 2025, telecom giant MTN completed the first phase of its $235 million Tiier III data centre in Lagos, Nigeria.  In August, Airtel announced that its 38-megawatt data centre will come live in 2026. If more workloads shift to smaller devices, it could slow the pace at which these local facilities are needed.

Global companies are already experimenting with reduced compute capacities in AI models. In March, Google launched TurboQuant, a compression algorithm designed to shrink how much memory AI models need to run. Refiant AI’s work sits in the same direction, but goes further by ensuring compression happens at much lower compute levels.

On its path to growth, Refiant AI said it is already in discussions with multinational technology firms to explore how its technology can reduce compute costs without compromising performance.

“AI’s biggest constraint isn’t demand – it’s energy,” said Joseph Goodman, Managing Partner, VoLo Earth. “What’s been missing is a fundamentally more efficient way to compute. Refiant’s architecture replaces brute-force scaling with a far more efficient, nature-inspired approach that lowers energy use while increasing capability.”