ICTworks

African AI startups captured barely 1 to 1.5% of global AI spending in 2025, even as AI pulled in close to half of all venture dollars worldwide. At the earliest stages the gap is wider still: in the second quarter of 2025, startups across the continent raised just $14 million in AI deals, about 0.02% of the $47.3 billion invested in AI globally. The constraint is not talent. It is capital, and the pathways to turn technical skill into a fundable company.

The development community is obsessing over the wrong climate threat. While everyone debates if data centers are causing rising temperatures, a groundbreaking study from Bangladesh reveals the real killer: heat doesn’t just slow down individuals. High temperatures destroy the collaborative advantage that makes teams twice as productive as solo workers. There are serious implications for ICT4D solutions.

On June 9, an American company shipped its most capable artificial intelligence model. Three days later, a letter from the US Commerce Department forced it offline for every user on the planet, with no notice and no appeal. The model was Anthropic’s Fable 5. The mechanism was an export control directive that barred access by any foreign national, so the company disabled it worldwide to comply. Other models kept running. This one did not.

Implementing case management systems in humanitarian and development contexts requires balancing individual service delivery with complex data governance requirements and donor expectations. Whether it’s tracking a child protection case, managing Mental Health and Psychosocial Support (MHPSS) services, or overseeing a refugee intake process, the challenge is making that data flow securely and efficiently. This challenge is amplified by a shifting macro environment.

Last week, Nivi and MSD announced a global collaboration to expand HPV disease awareness and cervical cancer prevention beginning in India, Indonesia, and Thailand, delivered through Nivi’s goal-oriented, AI-routing chat platform. The arrangement may look like a traditional development-sector initiative. The product, our askNivi chat platform, is free on WhatsApp and other popular messaging channels, and the money comes from an organizational payer, not individual end users.

Nonprofits in the Global South are now 35% more capable of digitalizing than their Global North peers, according to NetHope’s 2024 Digital Nonprofit Ability assessment. The catch: capability does not equal capacity. Most mission-driven organizations still treat technology as overhead rather than infrastructure, and 58% of nonprofit leaders cite staffing and operational capacity, not funding alone, as their biggest barrier to scaling impact.

Part 3 on the sovereign AI trap. Part 1: Nvidia sales channel. Part 2: Small models and edge inference. The case for small language models running on local devices is technically sound and economically honest. It is also politically inert without an institutional layer to enforce it. A community health worker in Rwanda running a fine-tuned Phi-4 model on a $40 smartphone is the right architecture.

Most digital health implementers I know can recite these headline statistics in their sleep. A projected global shortfall of 10 million health workers by 2030. An estimated 8.6 million deaths a year in 137 LMICs from inadequate access to quality care The pitch decks practically write themselves: AI is the force multiplier that saves global health from austerity. I am increasingly worried that we are mistaking the pitch deck for a plan.

Globally, 273 million children are out of school, and the gap is widening fastest in the lowest-income countries, where 33 percent of school-aged children are out of school, compared to 3 percent in high-income countries. School meals remain one of the most reliable interventions for boosting enrollment, attendance, and literacy, especially for girls.

Part 2 of a three-part series. Part 1: Sovereign AI is a Nvidia sales channel. The trap is real. The escape route is also real, and most strategy documents have not caught up to it. For most of the past three years, the implicit assumption in sovereign AI debates has been that meaningful AI capability requires hyperscaler-class compute. That assumption was correct in 2023. It is wrong now.