NEW YORK — The Gates Foundation has pledged at least $1 billion over the next two years to expand access to artificial intelligence in lower-income and underserved communities, betting that the technology can improve health care, education and agriculture if it is built around local needs rather than concentrated in the world’s richest countries.
The commitment, announced alongside the foundation’s annual Goalkeepers report, will fund AI-related work in education, health care, agriculture and the digital foundations required for systems to work in local languages and contexts. The effort comes with a warning from Microsoft co-founder Bill Gates: artificial intelligence could either widen access to knowledge and opportunity or deepen existing inequality.
“AI, if shaped well, could expand who has access to solutions, opportunities, and knowledge that has too often been out of reach,” Gates wrote in the report. “The decisions made in the next 12 to 18 months, about how AI is built, funded, and deployed, will determine whether this technology primarily benefits the people who already have the most or reaches those who have the least.”
The foundation’s spending plan directs approximately $400 million to education, $400 million to health care, $100 million to agriculture and $100 million to the underlying data and language infrastructure needed to make AI more useful for communities often excluded from the technology’s development.
The pledge is large by philanthropic standards, but small relative to the sums being spent by major technology companies on commercial AI systems. Gates Foundation Chief Executive Mark Suzman acknowledged that the organization’s investment is “a tiny proportion” of the billions being deployed by technology companies but argued that targeted funding can help shape who benefits from AI and how.
The central idea is not to give every person access to the same general-purpose chatbot. It is to support tools designed for specific, high-impact problems: helping health workers spot overlooked symptoms, assisting teachers with lesson plans, providing farmers with weather and crop advice, and ensuring AI systems understand languages spoken by communities that have little representation in existing data sets.
An AI-access strategy, not simply an AI-funding strategy
The Gates Foundation’s announcement is framed around equitable access rather than pure technological advancement.
That distinction matters because many of the best-known AI systems are trained primarily on data from wealthier countries and are optimized for widely used languages, well-connected users and commercial markets. A model that performs well in English for an office worker in New York or London may be far less useful for a rural health worker, teacher or farmer who speaks a language underrepresented in the data used to train it.
The foundation has set aside about $100 million to develop data sets in languages spoken by underserved communities and to make AI tools reflect local knowledge and conditions.
This is not only a translation problem.
A useful health-care tool needs to understand local disease patterns, medical terminology, treatment protocols and the realities of limited staffing or medical supplies. A useful agricultural tool must account for crop varieties, soil conditions, weather patterns, pests, access to fertilizer and local market conditions. A useful education tool must align with a country’s curriculum, classroom size, teacher workload and student language.
Without that local adaptation, AI can reproduce a familiar digital divide: the people and institutions with the most data, computing power and engineering talent receive the best systems, while others are given generic tools that do not reflect their needs.
Suzman told The Associated Press that AI must work in all local languages, not only in English and, to a degree, Mandarin, if it is to accelerate progress for people who have been left out of previous technology booms.
The foundation is partnering with Google.org and Microsoft’s AI for Good Lab on an open funding call for projects focused on strengthening AI frameworks for thousands of underrepresented languages across Africa.
Where the $1 billion will go
The Gates Foundation’s planned allocation shows that it is concentrating on service delivery rather than only on research or software development.
| Area | Planned funding | Intended use |
|---|---|---|
| Education | About $400 million | AI-enabled tools to support teachers and personalize learning |
| Health care | About $400 million | Clinical support, health-worker assistance, drug and vaccine development |
| Agriculture | About $100 million | Localized advice for smallholder farmers on weather, pests and fertilizer |
| Data and language infrastructure | About $100 million | Datasets and AI systems for underserved languages and local contexts |
The largest education investment is aimed at tools that help teachers understand where students are struggling and adapt lessons accordingly. One example cited by the foundation is a system that helps a teacher revise a lesson plan around concepts least understood by a class.
In health care, the foundation envisions AI systems that can review clinicians’ reports and flag symptoms or patterns that may have been overlooked. It also sees potential for AI to support frontline health workers, speed research into drugs and vaccines, and improve the flow of information in under-resourced clinics.
In agriculture, the funding could support applications that offer small farmers real-time advice on weather conditions, pest control, fertilizer use and crop management. These tools could be especially useful in regions where agricultural extension services, the networks that provide farmers with practical advice, are limited or understaffed.
The approach reflects a broader shift in philanthropy and development policy: technology is increasingly viewed not as a separate sector but as a tool embedded in public health, education, farming and government services.
The challenge will be to ensure that the tools are genuinely useful, affordable and accountable to the communities expected to use them.
Partnerships with major AI companies
The Gates Foundation is not attempting to build all of the technology itself.
It is working with major companies that have become central to the AI race, including OpenAI, Anthropic, Google and Microsoft. The partnerships are intended to pair the companies’ technical resources with the foundation’s global-health, development and education networks.
OpenAI has committed $50 million alongside the foundation for a pilot program to train health workers in Rwandan clinics. Suzman said the initiative is expected to expand to other African countries.
Anthropic is working with the foundation on efforts related to vaccine development and on improving local-crop data used by its chatbot and related systems.
Google.org and Microsoft’s AI for Good Lab are supporting projects aimed at improving AI tools for underrepresented African languages.
These partnerships could accelerate deployment. The companies bring model-development expertise, cloud capacity, engineering talent and existing AI platforms. The foundation brings relationships with local governments, clinics, schools, nonprofits and research institutions.
But the arrangement also raises questions that the foundation will need to answer over time: Who owns the resulting data? Who controls the tools after pilots end? Can local institutions maintain them? How are privacy, consent, bias and safety handled? And are communities participating in design, or merely receiving products developed elsewhere?
The success of the pledge will depend on more than how much money is spent. It will depend on governance.
Gates warns governments are behind
The funding announcement comes as Gates has adopted a more urgent public tone about AI’s risks.
In a Reuters interview, he said no government in the world is adequately prepared for the social changes AI could bring. He cited possible job disruption, cyberattacks and increasing reliance on AI “companions” among the concerns he believes policymakers must confront.
“I don’t think any government is nearly as deep on this as they have to be,” Gates said. “Governments are way behind on this one.”
Gates’ comments arrive during an intense global debate over whether AI companies should slow development of increasingly capable models. Anthropic CEO Dario Amodei has urged companies to pace frontier AI progress while safety methods catch up. OpenAI CEO Sam Altman has expressed support for greater caution, and regulators in several countries are considering how to audit, test and govern powerful systems.
The Gates Foundation is not presenting its $1 billion commitment as a direct solution to frontier-AI safety. Instead, it is focused on a different but related question: whether the social and economic gains from AI will reach people outside the wealthy countries and corporations that dominate development today.
That emphasis is particularly important at a time when international aid budgets are under pressure. The Gates Foundation’s Goalkeepers report points to setbacks in global development, including concern that 2025 could be the first year this century in which child deaths rise. It also notes disruption to global HIV/AIDS prevention and treatment efforts following changes to U.S. foreign-assistance programs.
In that setting, the foundation sees AI as a possible way to expand the reach of constrained systems. A clinical decision-support tool cannot replace doctors or nurses, but it may help a frontline worker manage a larger workload. A teacher-assistance tool cannot solve overcrowded classrooms, but it may give an educator better insight into which students need help. A farming app cannot end drought, but it may improve decisions about planting, water, pests or fertilizer.
The difference between a useful tool and a harmful shortcut will often come down to implementation.
The risks of “AI for good”
AI projects in health, education and agriculture carry real risks, especially when deployed in communities with less power to challenge errors or demand accountability.
A health tool that produces inaccurate recommendations can cause harm if clinicians trust it too much. An education system that incorrectly labels a student’s ability can reinforce disadvantage. An agricultural model trained on incomplete data can offer bad advice that costs a farmer part of a harvest.
Privacy is another central concern. AI systems often require substantial amounts of data. Health records, student information, language data, agricultural practices and location data can all be sensitive. Communities need meaningful information about how their data will be used, stored, shared and protected.
Bias also remains a major issue. Gates has warned that AI can become “the greatest equalizer” or the “worst source of injustice,” depending on how it is developed and deployed.
A system trained primarily on data from high-income countries can misinterpret accents, local vocabulary, symptoms, crops or social conditions elsewhere. A system designed without input from women, rural communities, people with disabilities or minority-language speakers can amplify existing inequities.
The Gates Foundation’s language-data investment is one response to that problem, but it is only a starting point. Creating a data set is not the same as creating a fair, accurate and trusted system. The work requires local researchers, community organizations, practitioners and governments to have meaningful decision-making roles.
Why local languages matter
The foundation’s focus on language may be one of the most significant parts of the announcement.
Large language models are most capable in languages with extensive digital text, high-quality training data and substantial commercial demand. English dominates that landscape. Other widely used languages have more representation than many African, Indigenous and regional languages, but performance can still vary sharply.
For someone who does not speak a language well supported by AI systems, the technology may be inaccessible, inaccurate or culturally irrelevant.
Local-language AI could have broad effects:
- A health worker could consult a tool in the language used in the clinic.
- A farmer could receive advice through voice messages rather than text-heavy interfaces.
- A teacher could generate materials suited to students’ first language.
- A small business owner could use digital tools without needing English-language fluency.
- Public services could communicate with citizens more clearly.
But language support cannot be a purely technical exercise. It requires careful work on dialects, cultural meanings, local terminology and safeguards against using language data in ways that exploit or misrepresent communities.
The foundation’s $100 million commitment to language and data infrastructure recognizes that without this work, AI access will remain uneven by design.
Measuring results, not announcements
The Gates Foundation’s pledge is substantial, but the most important outcomes will not be measured by dollars committed or pilot projects announced.
They will be measured by questions such as:
- Do health workers make more accurate or timely decisions?
- Do teachers save time and improve student learning?
- Do farmers earn more or reduce crop losses?
- Do AI tools work reliably in local languages?
- Are women, rural communities and people with disabilities able to use them?
- Are data rights and privacy protected?
- Can governments and local institutions sustain programs after philanthropic funding ends?
- Do communities have the ability to challenge errors or opt out?
Those questions matter because many “AI for good” projects have struggled to move from demonstration to durable public benefit. Technology can be impressive in a pilot but difficult to maintain once the original grant, outside engineers or cloud credits disappear.
The foundation’s two-year commitment should therefore be viewed as an opening phase, not a completed solution.
Gates Foundation CEO Suzman has described the $1 billion as a “down payment,” suggesting that the organization expects the effort to grow beyond the initial period.
A high-stakes test for equitable AI
The Gates Foundation’s pledge arrives at a moment when the AI debate has become polarized.
Some see AI as a productivity engine capable of transforming science, education and public services. Others see an industry moving too quickly, with insufficient safeguards against job loss, misinformation, surveillance, cyberattacks and loss of human control.
The foundation is trying to hold both views at once.
It accepts that AI carries serious risks. Gates has openly warned about them. But it also argues that leaving low-income communities outside the technology’s development and benefits would create another form of harm.
The $1 billion commitment is a bet that AI can serve as public-interest infrastructure rather than merely a commercial product. It is also a challenge to governments, companies and development institutions: access must mean more than an internet connection or a free chatbot. It must include useful local tools, safe systems, language inclusion, strong public institutions and people who have a say in how the technology affects their lives.
Whether the pledge achieves that standard will depend on the choices made after the announcement, in clinics, classrooms, farms, public agencies and the communities the funding is meant to serve.
FAQs
How much is the Gates Foundation pledging for AI access?
The Gates Foundation is committing at least $1 billion over the next two years to expand equitable access to AI and AI-enabled tools.
Where will the $1 billion go?
The foundation said about $400 million will go to education, $400 million to health care, $100 million to agriculture and $100 million to data and language infrastructure supporting underserved communities.
Which countries will receive the funding?
The foundation has not published a complete country-by-country list. It said the funding will support partners working globally, with particular emphasis on underserved communities and local-language AI systems. A health-worker pilot in Rwanda is expected to expand to other African countries.
What does “equitable AI access” mean?
In this context, it means more than giving people access to a chatbot. The aim is to create and deploy tools that work in local languages, reflect local conditions, protect users and address practical needs in health care, education and agriculture.
Which companies are involved?
The foundation is working with OpenAI, Anthropic, Google.org and Microsoft’s AI for Good Lab on different parts of the effort, including health-worker training, vaccine work, crop data and support for underrepresented languages.
What are the main risks of AI in lower-income communities?
Risks include inaccurate recommendations, bias, privacy violations, exclusion of local languages, dependence on outside technology providers and a lack of accountability when systems make mistakes. These risks can be greater where users have limited ability to contest automated decisions or access alternative services.
Why is local-language data important for AI?
AI tools are often strongest in languages with large digital data sets, particularly English. Local-language data can help systems understand the words, contexts and needs of communities that are otherwise poorly served, making tools more useful for teachers, health workers, farmers and public services.