Refiant AI, a South Africa-founded startup that makes use of algorithms to compress synthetic intelligence (AI) fashions, has closed a $5 million seed spherical to construct its platform, develop its crew and assist enterprise partnerships.
VoLo Earth Ventures, a California local weather expertise fund, led the spherical, backing what it considers a extra environment friendly method to growing smarter and leaner AI fashions.
Based in 2025 by Viroshan Naicker, Siddharth Gutta, and Mathew Haswell, Refiant AI is constructing instruments that restructure and compress AI fashions by lowering computational weight and retraining them to keep up efficiency, in order that they will run effectively on smaller or native machines.
As world firms, corresponding to Meta and Microsoft, race to deploy extra environment friendly AI fashions, they’re investing closely in constructing knowledge centres outfitted with graphics processing items (GPUs) and cooling programs.
That vitality infrastructure is each costly and resource-intensive. Within the first quarter of the yr, each firms every dedicated practically $50 billion in extra knowledge centre leases to assist AI. Bloomberg reported that these pledges among the many largest cloud computing firms helped push commitments to knowledge centre leases above $700 billion.
“AI’s rising vitality footprint is among the most pressing and underappreciated challenges within the local weather house,” Gutta, one in all Refiant AI’s co-founders, mentioned. “The trade’s default reply is to construct extra knowledge centres and devour extra energy. Ours is to make the AI itself dramatically extra environment friendly.”
As an alternative of counting on constructing extra infrastructure, Refiant AI intends to make the AI fashions lighter in order that they require fewer assets to function. The corporate mentioned it had efficiently compressed a 120-billion-parameter mannequin to run on a normal laptop computer with simply 12 gigabytes (GB) of RAM, a activity that will sometimes require not less than 80GB of reminiscence. The corporate mentioned that the compressed mannequin retained between 95% and 99% of its unique efficiency, utilizing greater than 80% much less vitality.
Refiant mentioned its expertise is suited to markets the place computing infrastructure is proscribed, and cloud entry could be costly. In sectors like banking, telecoms, and authorities companies, its instruments may permit organisations to run AI programs while not having fixed entry to the cloud.
Nigeria’s Central Financial institution just lately embedded AI and machine studying into new baseline requirements that require monetary establishments to deploy automated anti-money laundering programs. Instruments like Refiant’s may allow banks to deploy these programs regionally with out counting on pricey overseas infrastructure or transferring delicate knowledge throughout borders.
The implications of compressed AI fashions differ. If highly effective AI fashions can run on on a regular basis {hardware}, African organisations might rely much less on world cloud suppliers and costly infrastructure. For the continent whose 250 knowledge centres solely account for 0.6% of the worldwide capability, the AI compression expertise may decrease the barrier to adopting superior AI fashions whereas preserving knowledge native.
On the similar time, it introduces a brand new pressure. Native knowledge centre operators throughout Africa have been increasing their capability to satisfy the rising demand for cloud companies, estimated to succeed in $25.46 billion by 2029. In July 2025, telecom big MTN accomplished the primary part of its $235 million Tiier III knowledge centre in Lagos, Nigeria. In August, Airtel introduced that its 38-megawatt knowledge centre will come stay in 2026. If extra workloads shift to smaller units, it may gradual the tempo at which these native services are wanted.
World firms are already experimenting with decreased compute capacities in AI fashions. In March, Google launched TurboQuant, a compression algorithm designed to shrink how a lot reminiscence AI fashions must run. Refiant AI’s work sits in the identical route, however goes additional by making certain compression occurs at a lot decrease compute ranges.
On its path to progress, Refiant AI mentioned it’s already in discussions with multinational expertise corporations to discover how its expertise can scale back compute prices with out compromising efficiency.
“AI’s greatest constraint isn’t demand – it’s vitality,” mentioned Joseph Goodman, Managing Associate, VoLo Earth. “What’s been lacking is a essentially extra environment friendly approach to compute. Refiant’s structure replaces brute-force scaling with a much more environment friendly, nature-inspired method that lowers vitality use whereas growing functionality.”
Elevate your perspective with NextTech Information, the place innovation meets perception.
Uncover the newest breakthroughs, get unique updates, and join with a world community of future-focused thinkers.
Unlock tomorrow’s tendencies as we speak: learn extra, subscribe to our e-newsletter, and develop into a part of the NextTech neighborhood at NextTech-news.com
