“It would be a big boon that cuts down costs significantly,” said Botshelo Baloyi, chief executive and founder of UmKhoAI, Africa’s first offline national AI ecosystem. UmKhoAI Public allows AI access without buying data, wifi, orinternet access—a crucial feature for a continent where connectivity remains patchy. The company has built an AI model with a language processor that can speak all 12 official South African languages, trained using music, movies, TikToks, reels, and TV shows.
UmKhoAI currently operates a data centre in South Africa with MavenGlobal, but is increasingly seeking cheaper alternatives to achieve sovereignty in its efforts, Baloyi said, adding that the company is already working with the South African government on several projects, including defence.
With sovereign training mandates emerging across the Global South and pooled compute initiatives gaining traction, India’s AI infrastructure is becoming increasingly attractive to startups locked out by prohibitive Western pricing.
The interest comes as officials hint at imminent policy moves. “There are conversations around plans to announce a platform to broaden global access to AI infrastructure, including pooled compute and datasets access at this week’s summit,” S. Krishnan, secretary at Meity, told reporters at a briefing on 29 December.
Bridging connectivity gap
Others, like Pulasafe, based in Botswana, southern Africa, which works in disaster management, see India as both a potential infrastructure partner and market. “India could be a potential market for us as our solutions can be similar topographically to how India functions,” said Xoliswa Mbingo, co-founder of Pulasafe.
The company is currently looking for cheaper cloud options, “especially as cloud credits become more expensive day by day,” she said.
Robotics has also gained prominence, with the South African government deploying systems from startups.
Key Takeaways
- Indian compute is roughly 60–70% cheaper than Western commercial cloud providers, making it the ‘budget’ choice for the Global South.
- African founders are prioritizing sovereign training over the prestige of Big Tech partnerships.
- India is effectively treating AI compute as ‘Digital Public Infrastructure’, similar to UPI or Aadhaar.
- A projected 7 million GPU-hour deficit in Africa is the primary driver for this cross-continental collaboration.
- Startups like UmKhoAI are focusing on offline accessibility, a niche where India’s rugged tech stack excels.
“We are working with government schools from the Department of Basic Education and Higher Education to upskill learners with AI courses,” said Evans Parsons, founder at Optimas Robotics.
The company has raised funding from Microsoft South Africa and uses the tech giant’s cloud services.
The company is also scouting for collaborations with the Indian government and Indian AI companies for “sharing AI stack”, said Parsons.
Lifesten Health, a Rwandan healthtech company that provides credits for wellness services when users complete health goals, operates in 89 countries, including India and “would benefit from India’s AI stack solutions,” said Ogweno Stephen, founder of the firm.
But the path to affordable infrastructure remains fraught, with broader structural challenges complicating access.
Sovereign training is becoming critical as legislation mandates where AI models must be trained or hosted. “Getting access at a good price point is also a limiting factor” for training systems that present “a barrier to entry for a lot of small companies,” explained Vukosi Marivate, professor at the University of Pretoria, South Africa.
Pooled compute resources could help: “Figuring out ways to do this allows more people to play” and “allows startups to have some chances of getting whatever they’re offering out to the people that they want to actually target.”
Battle for African cloud
Africa saw its largest AI exit in 2023, when German multinational BioNTech acquired InstaDeep for $680 million. Startups like Lelapa AI (VulaVula for Southern African languages) and CDIAL (tools for 180 African languages) address critical linguistic gaps, while the continent’s 159 AI startups have raised $803 million total, with Kenya leading at $242 million across 19 companies. South Africa plans to develop 300 AI startups by 2030, according to media reports.
Yet fundamental challenges remain: only 5% of African AI innovators have reliable access to advanced compute, with 7 million GPU hours of unmet demand projected over three years, according to a World Economic Forum report. Infrastructure gaps and inconsistent electricity remain critical barriers.
Meanwhile, big tech has slowly yet steadily built its presence: Amazon’s AWS has data centres serving South Africa’s Cape Town since 2020 and Nigeria’s Lagos since 2023; Microsoft had previously announced a $1 billion investment in data centres in Kenya, and also maintains Azure regions in South Africa; and Google launched a Johannesburg cloud region in early 2024.
Local players like Africa Data Centres have partnered with Nvidia to deploy 12,000 GPUs, while MTN Nigeria is building a $240 million Tier III data centre. However, most existing facilities cannot support high-performance AI/ML workloads, with Africa’s data centre market valued at just $1.26 billion, with projections to reach $3.06 billion by 2030.
India’s AI stack, launched under the ₹10,372 crore IndiaAI Mission, offers subsidized compute access at ₹65 per hour (roughly one-third of global rates) through over 38,000 GPUs and 1,050 TPUs, alongside AIKosh—a national repository hosting 7,500+ datasets and 273 AI models. The infrastructure operates on a compute-as-a-service model, providing shared access to startups, researchers, and institutions, with up to 25% of compute costs offset through grants for developing sovereign AI models. Tensor Processing Units (TPUs) are used for training and inference of AI models.
