For years, the AI story has been about breakthrough models and eye‑popping demos. Now the more interesting story is who actually gets to use those capabilities in a meaningful way—and who risks getting left behind. OpenAI’s “for Countries” push is basically its answer to that problem: a bid to turn advanced AI from something a few tech‑savvy users exploit into core infrastructure for education systems, public services, and national economies.

At the heart of this is a phrase you’ll hear a lot from OpenAI executives: “capability overhang.” In plain terms, today’s frontier models can already do far more than most people, companies, or governments are using them for. Internal data shows a huge gap between casual users who fire off simple prompts and “power users” who lean on AI for multi‑step, complex work—those power users are tapping into roughly seven times more advanced “thinking capabilities” than the typical user. Zoom out to the country level and the divide gets sharper: among more than 70 countries with heavy ChatGPT usage, some are using about three times more of these advanced capabilities per person than others, regardless of GDP. Vietnam and Pakistan, for example, show unusually high use of agentic tools like data analysis and code assistants, even though they’re not among the richest nations.

That asymmetry is what worries policymakers gathering around events like the World Economic Forum in Davos. If a small group of nations manages to integrate AI deeply into their schools, businesses, and bureaucracies while others stick to surface‑level use, the productivity and growth gap could become very hard to close later. In OpenAI’s own framing, some countries are already using AI to solve harder problems, freeing up workers for higher‑value tasks, speeding up new products and services, and generally moving faster than their peers.

That’s the context for “OpenAI for Countries,” a program the company formally introduced in 2025 as part of its broader Stargate strategy. The pitch to governments is straightforward: instead of treating ChatGPT as just another consumer app, treat OpenAI as a strategic infrastructure partner. The company says it wants to support nations that “prefer to build on democratic AI rails,” positioning its platform as an alternative to more centralized, surveillance‑heavy AI models associated with authoritarian regimes. In practice, that offer bundles three big promises: help build in‑country data center capacity so data stays sovereign, customized versions of ChatGPT for citizens and public institutions, and support in using AI to modernize services like healthcare and education.

The latest evolution of that strategy is an expansion announced in January 2026, alongside OpenAI’s new report, “Ending the Capability Overhang.” In that report, the company lays out the data on uneven AI adoption and argues that closing the gap is now as much a policy challenge as a technical one. On stage, OpenAI’s Head of “for Countries,” George Osborne, framed the 2026 expansion as a move from pilots to broader national programs spanning education, health, AI skills training and certifications, disaster response, cybersecurity, and startup accelerators. The goal is to give governments a menu of options: they can plug AI into classrooms, hospitals, emergency response systems, or entrepreneurship programs, depending on where they see the biggest upside.

Education is clearly the first big testbed. Under the banner “Education for Countries,” OpenAI is partnering with governments and university consortia to embed AI into national education systems, not just as a gadget but as part of the core teaching and assessment stack. The initial cohort reads like a mini‑map of AI‑curious states: Estonia, Greece, Italy’s CRUI (a major university rectors’ conference), Jordan, Kazakhstan, Slovakia, Trinidad & Tobago, and the United Arab Emirates. For these governments, OpenAI is offering a mix of tailored access to tools like ChatGPT Edu and newer GPT‑5‑class models, large‑scale research on learning outcomes, national‑level training programs for teachers and students, and a network to swap best practices with peers.

Some of this is already happening on the ground. Estonia, long seen as a digital‑first government, has rolled out AI tools across public universities and secondary schools, reaching more than 30,000 students, educators, and researchers in the first year. A longitudinal research project with the University of Tartu and Stanford is tracking how AI affects learning outcomes for around 20,000 students over time—exactly the kind of data ministries will want before scaling these tools further. Kazakhstan, meanwhile, has joined the same education cohort and plans nationwide training programs with OpenAI specialists to focus on practical classroom use starting in 2026.

Behind the scenes, this educational push doubles as workforce policy. Governments are starting to treat AI skills almost like basic digital literacy, something as fundamental as using email or spreadsheets. OpenAI’s offer includes certifications via an “OpenAI Academy” model and ChatGPT‑based assessments designed to align with local workforce needs, so graduates can prove they know how to use AI tools effectively on the job. For teachers, the promise is reduced administrative overhead and new ways to personalize learning, with the caveat that they’ll need training and clear guidelines so AI augments rather than replaces their judgment.

The company says the rest of its “for Countries” portfolio will follow the same pattern: broad themes, but flexible implementation. In healthcare, that might mean leveraging OpenAI research like the “Horizon 1000” initiative on primary care, pairing large language models with clinical workflows to help triage patients or draft documentation—always with local regulation and safety oversight in the loop. In disaster response, AI models could help forecast risks, synthesize incoming reports, or route resources more efficiently during floods or wildfires. Cybersecurity partnerships might focus on both sides of the arms race: helping governments defend critical infrastructure while putting guardrails in place against AI‑enabled attacks. And startup accelerators backed by OpenAI tooling are meant to ensure local founders, not just Silicon Valley players, can build AI‑native businesses on top of these models.

Of course, none of this is purely altruistic. Strategically, OpenAI is trying to do two things at once: grow the global market for its services and shape the governance norms that will define how AI is used in public institutions. By working directly with ministries and university consortia, the company positions itself as a thought partner on issues like safety, privacy, and responsible deployment, not just a vendor selling API credits. That gives it a front‑row seat as countries write rules around everything from data localization to classroom AI policies.

The “democratic AI rails” language is doing a lot of work here. It taps into a geopolitical undercurrent: many governments are wary of both US‑centric tech dominance and Chinese‑style AI models tied closely to state surveillance. OpenAI’s pitch is that its platform can offer strong privacy controls, in‑country data centers, and alignment with liberal democratic norms, while still giving countries enough flexibility to tailor systems to local languages and cultural contexts. Whether that balance holds as more nations sign up—and as regulations tighten—will be one of the big storylines to watch.

For governments, the opportunity and the risk are both obvious. Move too slowly and you risk a permanent capability gap: your students, startups, and civil servants are stuck a generation behind peers who’ve normalized AI‑augmented work. Move too fast without robust guardrails and you invite problems around bias, privacy, over‑reliance on opaque systems, or simple overhype, especially in sensitive sectors like education and healthcare. That’s why OpenAI keeps emphasizing “iterative deployment”—rolling out tools in stages, studying real‑world impacts, and adjusting along the way—something it originally developed as a safety principle for model launches but is now applying to national deployments.

In the end, “OpenAI for Countries” is less about a single flagship product and more about a long negotiation between a powerful AI vendor and states trying to future‑proof themselves. The capability overhang that OpenAI is so keen to talk about is real: the models are racing ahead of the ways we use them. The open question, as these partnerships scale through 2026 and beyond, is which countries will manage to turn that unused capacity into tangible gains for their citizens—and on whose terms.