The Gates Foundation is investing at least one billion dollars in better access to artificial intelligence for poor countries over the next two years. Four out of ten dollars are going into health projects, four more into education, and the rest into agriculture and multilingual AI data.
Commitment consolidates four funding lines
The commitment is part of the tenth Goalkeepers Report of the foundation, which was published on September 15, 2026, under the title “Make This Matter: AI, Equity, and the Choice We Can’t Delay.” Four out of ten dollars are flowing into health projects, such as AI-supported ultrasound devices that are intended to enable rural doctors to make hospital-level diagnoses, as GeekWire reports.
The same amount of money is going into education, specifically into software that measures the math learning levels of schoolchildren in the USA and other countries. The remaining two-tenths are distributed between agriculture and the development of multilingual AI datasets.
Around 100 million dollars are earmarked for soil analysis and crop forecasts for small farmers in poorer regions, as well as the same amount for training data in currently underrepresented languages. Foundation CEO Mark Suzman emphasizes that it is not about technology for its own sake, but about concrete applications with ongoing human oversight. According to their own statements, the money is primarily intended for practical implementation on the ground, rather than for the development of new AI models.
Study shows significant language gap in AI systems
The background of the commitment is an analysis published in the same report regarding the linguistic imbalance of today’s AI systems. More than ninety percent of the training data for earlier large language models comes from English-language sources.
For English-language speech recognition, the error rate is below six percent. In the West African language Yoruba, which is spoken by tens of millions of people in Nigeria, the error rate, on the other hand, is over sixty percent. This figure comes from the foundation itself and has not been independently verified. The foundation sees similar gaps in numerous African and South Asian languages. The report warns that without countermeasures, AI is more likely to deepen existing wealth disparities rather than reduce them.
The analysis follows the same logic as an earlier essay by Bill Gates on the AI disruption in the labor market. He already warned in August that wealthy countries could utilize the technology first and best if no one intervened. In the new report, he sharpens the thought: people with the greatest need often have the least influence over where innovation goes.
Child mortality rises for the first time in decades
The foundation deliberately places the AI commitment alongside a second, grim statistic. According to data from the research institute IHME, the number of children dying before their fifth birthday is expected to rise from 4.6 to 4.8 million by 2025. This would be according to Hoodline the first increase in this century. The Seattle-based institute has been monitoring global health data for decades and is considered one of the most important sources for worldwide mortality statistics.
For fifty years, child mortality had almost continuously declined, most recently by 51 percent since the year 2000. The report cites decreasing international health aid as the cause: the USA alone cut its foreign aid by about 27 percent in 2025, from 49 to 36 billion dollars.
The dissolution of the development agency USAID led to the shutdown of more than 1,700 HIV treatment centers worldwide. Programs against tuberculosis and malaria lost 1.4 billion dollars in donor aid. The billion for AI is not intended to close this gap, but according to the foundation, at least to replace part of the lost impact.
It remains open whether a private foundation can even remotely compensate for the effects of shrinking state development aid – one billion dollars seems small compared to the billions in aid funds that have been cut. Therefore, it will be crucial whether the announced language datasets actually reach the local organizations that the foundation names as partners, without naming them specifically. The next Goalkeepers Report in a year is likely to show whether the Yoruba error rate has even changed.
