AI Diffusion Report H2 2025: Global Adoption at 16.3%

Global AI Adoption Reaches New Heights in H2 2025

The Microsoft AI Economy Institute has released its second-half 2025 AI Diffusion Report, revealing that global generative AI adoption has climbed to 16.3% of the world’s working-age population. This represents a meaningful increase from the 15.1% recorded in the first half of 2025, translating to roughly one in six people worldwide now regularly using generative AI tools. The report marks the continuation of Microsoft’s population-normalized methodology for tracking AI usage, drawing on aggregated and anonymized telemetry data adjusted for OS market share, internet penetration, and country-level demographics.

For organizations and policymakers watching the AI revolution unfold, this second-half update provides crucial context. While the headline number reflects remarkable progress for a technology that entered mainstream use only recently, the underlying data tells a more nuanced story about who benefits from AI—and who risks being left behind. Understanding these dynamics is essential for anyone making strategic decisions about AI investment, workforce development, or digital infrastructure. The first-half Microsoft AI Diffusion Report established the baseline; this H2 update reveals how quickly the landscape is shifting.

The Widening Digital Divide Between North and South

Perhaps the most consequential finding in the Microsoft AI Diffusion Report H2 2025 is the growing gap between the Global North and Global South. Adoption in high-income economies reached 24.7%, up from 22.9% in H1—a gain of 1.8 percentage points. Meanwhile, the Global South advanced from 13.1% to just 14.1%, a gain of only 1.0 percentage points. The adoption gap has widened from 9.8 to 10.6 percentage points in a single reporting period.

This divergence carries profound implications. As the report states, “AI’s benefits are expanding, but they are not expanding equally.” Of the ten countries with the largest absolute increases in AI adoption share during H2 2025, all are high-income economies. The pattern suggests that existing advantages in digital infrastructure, education systems, and economic resources are compounding through AI adoption, potentially creating a self-reinforcing cycle. Countries with higher internet penetration, more widespread device ownership, and stronger educational foundations find it easier to integrate AI into daily work and life.

The lowest-adoption countries paint a stark picture: Cambodia at 5.1%, Afghanistan at 5.6%, Rwanda at 6.3%, and Ethiopia at 6.8%. These numbers represent not just a technology gap but a potential productivity and economic competitiveness gap that could define the next decade of global development. Research on AI’s impact on firm productivity and employment suggests that organizations leveraging AI early gain compounding advantages over those that delay adoption.

Top 30 Countries by AI Diffusion Rate

The H2 2025 rankings reveal both stability at the top and significant movement in the middle tiers. The United Arab Emirates maintains its commanding lead at 64.0% adoption (up from 59.4%), followed by Singapore at 60.9%. Norway holds third place at 46.4%, with Ireland (44.6%) and France (44.0%) rounding out the top five. European nations dominate the top 30, with 16 countries represented, followed by Asia-Pacific with 7, and the Middle East with 4.

Notable ranking shifts include the Netherlands passing the United Kingdom (both at 38.9%, but the Netherlands edging ahead on decimal precision), Belgium moving ahead of Canada (36.0% vs. 35.0%), and most dramatically, South Korea climbing seven positions from 25th to 18th place. The United States dropped one spot to 24th at 28.3%, reinforcing the paradox of innovation leadership without proportional population-level adoption.

Among the fastest-growing countries, South Korea led with a 4.8 percentage point increase, followed by the UAE at 4.5 and France at 3.1. The concentration of growth among wealthy nations underscores the structural advantages that drive adoption: robust broadband infrastructure, high smartphone penetration, strong educational systems, and economic environments where AI tools translate directly into productivity gains. Meanwhile, some countries showed flat or even negative growth—Jamaica and Ukraine both declined by 0.1 percentage points.

South Korea’s Remarkable AI Surge

South Korea’s ascent from 25th to 18th place represents the most dramatic country-level movement in the H2 2025 report. AI usage grew from approximately 26% to over 30% of the population, with total growth since October 2024 exceeding 80%—far outpacing both the global average of 35% and the U.S. rate of 25%. South Korea has also become the world’s second-largest ChatGPT subscriber market, behind only the United States.

The Microsoft report identifies three converging drivers behind this surge. First, aggressive national AI policies accelerated integration across sectors. The Korean government reconstituted its national AI coordinating body as the National AI Strategy Committee, a cross-ministerial decision-making group with real authority. The country enacted the AI Basic Act to balance innovation with governance, announced plans to expand AI-focused science high schools in regional areas, and committed $1.2 billion to teach AI skills from elementary school through the workplace.

Second, frontier language model improvements proved transformative for Korean-speaking users. The report provides striking benchmark data: GPT-3.5 scored just 16 on the Korean college entrance exam (CSAT), placing it below the 5th percentile in Korean proficiency. GPT-4o, released in April 2025, scored 75—roughly the 70th percentile. GPT-5, released in August 2025, achieved a perfect 100, placing above the 95th percentile. As the report notes, “For Korean-speaking users, GPT-4o was the turning point, making everyday tasks practical and reliable with large language models for the first time.”

Third, a viral cultural moment in April 2025 served as a mass onboarding event. Ghibli-style AI-generated images created with ChatGPT-4o swept Korean social platforms, requiring no technical skills and producing instantly shareable results. The phenomenon introduced millions of first-time users to AI capabilities, and engagement data suggests many continued exploring additional features after the trend faded—turning a viral moment into lasting adoption.

UAE’s Deliberate Path to AI Leadership

The United Arab Emirates’ continued dominance at 64.0% adoption is no accident. The UAE’s AI journey began in October 2017, when the country appointed the world’s first Minister of State for Artificial Intelligence—a full five years before ChatGPT launched. That same year, the UAE published a national AI strategy covering nine priority sectors, establishing a comprehensive roadmap that has been systematically executed ever since.

The report highlights several structural factors behind the UAE’s success. The country created regulatory sandbox environments that allowed controlled AI experimentation without the bureaucratic paralysis that slows adoption elsewhere. Special visa programs, including the Golden Visa initiative, were deployed to attract global AI talent. Most importantly, the UAE adopted principles-based regulatory guidelines that provided clear direction without creating the compliance burden that discourages innovation. As the PwC Global CEO Survey 2025 underscores, regulatory clarity is one of the strongest predictors of enterprise AI adoption.

Perhaps the most revealing datapoint is the trust differential. The 2025 Edelman Trust Barometer shows UAE residents trust AI at approximately 67%, compared to just 32% in the United States—a staggering 35 percentage point gap. Western European nations average similarly low trust scores around 32%. The report suggests this trust was earned through demonstrated outcomes: “AI that actually works in daily transactions.” The lesson for other nations is clear—trust follows function, and function follows strategic investment.

The United States Paradox in AI Adoption

One of the report’s most thought-provoking findings is the United States’ relatively modest position at 24th globally, despite being the unquestioned leader in AI infrastructure and frontier model development. The U.S. grew from 26.3% to 28.3% adoption in H2 2025, a respectable 2.0 percentage point increase, but the country actually slipped one rank as others grew faster. Since October 2024, U.S. AI adoption has grown by only 25%—less than the global average of 35% and far behind South Korea’s 80%+ growth rate.

The report is direct about the implications: “Leadership in innovation and infrastructure, while critical, does not by themselves lead to broad AI adoption.” The United States builds the most powerful AI models, hosts the largest AI companies, and attracts the most AI investment globally. Yet it lags far behind smaller, more digitized economies in actual population-level usage. This paradox echoes findings from other technology adoption cycles, where the countries that create breakthrough technologies are not always the ones that most effectively distribute their benefits across the population.

Several factors may explain the gap. The U.S. lacks a coordinated national AI adoption strategy comparable to the UAE’s or South Korea’s. High income inequality means that AI’s productivity benefits may be concentrated among knowledge workers in major metropolitan areas while bypassing large segments of the population. Additionally, lower trust in AI technology—the same 32% figure as in the Edelman survey—suggests cultural resistance that policy alone cannot overcome. For U.S. policymakers, the data presents an uncomfortable question: can you maintain AI leadership if your population increasingly falls behind in AI literacy?

DeepSeek’s Open-Source Disruption

The emergence of DeepSeek as a major force in global AI adoption is described as “one of 2025’s most unexpected developments.” The Chinese company released its model weights under an MIT License and launched a completely free chatbot across web and mobile platforms—no subscription fees, no payment requirements, no credit card needed. In September 2025, DeepSeek took the unusual step of inviting independent researchers to rigorously test its model performance, targeting publication in Nature.

The market share data reveals a striking geographic pattern. DeepSeek captured 89% of the AI market in China, 56% in Belarus, 49% in Cuba, and 43% in Russia. Across Africa, DeepSeek’s adoption was estimated at 2 to 4 times higher than in other regions. Chinese tech companies, including DeepSeek and Huawei, actively promoted the platform in African markets through partnerships, outreach, and integration with local telecom services. This adoption pattern has significant geopolitical implications.

DeepSeek succeeded by eliminating the cost barriers that prevented millions of potential users in lower-income countries from accessing AI tools. Where Western platforms require credit cards, paid subscriptions, or premium upgrades for full functionality, DeepSeek offered comparable capabilities for free. As the report observes, “Global AI adoption is shaped as much by access and availability as by model quality. Users gravitate toward platforms that fit their economic, linguistic, and political context.” This insight should concern Western AI leaders—the next billion AI users may not be using their products. The implications for SaaS business models in 2026 are substantial, as free open-source alternatives reshape competitive dynamics.

Language Models as Adoption Catalysts

South Korea’s adoption story provides a compelling case study for how language model improvements drive usage in non-English-speaking markets. The progression from GPT-3.5’s near-unusable Korean performance (below the 5th percentile on the CSAT) to GPT-5’s near-perfect score (above the 95th percentile) happened in less than two years. Each improvement corresponded to measurable adoption gains, with GPT-4o identified as the specific inflection point that made everyday Korean-language tasks—conversation, drafting, translation, and analysis—practical for the first time.

This finding carries important implications for the hundreds of languages worldwide that remain underserved by current AI models. The report frames it optimistically: “As models expand and deepen linguistic coverage, usage is likely to follow.” However, the data also reveals an interim challenge: in countries where local languages aren’t well served by AI, users sometimes switch to English at disproportionate rates, potentially exacerbating existing linguistic inequalities. Research on large language model capabilities and limitations consistently shows that performance varies significantly across languages, with resource-rich languages receiving disproportionate attention in training data.

For AI developers, the message is clear: investing in multilingual capabilities is not just a matter of inclusivity—it is a growth strategy. The South Korean example demonstrates that when language barriers fall, adoption can accelerate dramatically. For the OECD and international development organizations, the data suggests that supporting multilingual AI development should be a priority for bridging the digital divide.

What the AI Diffusion Data Means for 2026

Looking ahead, several trends from the H2 2025 data are likely to intensify. The digital divide between the Global North and Global South may continue widening unless deliberate intervention occurs. Countries with strong national AI strategies—the UAE model of long-term institutional commitment—will likely continue outperforming those without coordinated approaches. The DeepSeek phenomenon suggests that open-source AI models will play an increasingly important role in extending access to populations previously excluded by cost barriers.

The report warns that “the rapid spread of an open model also raises questions about standards and safety, as open-source systems can propagate widely with limited oversight.” This tension between accessibility and governance will be one of the defining challenges of 2026. As AI adoption penetrates deeper into daily life across more countries, the need for coordinated international frameworks becomes more urgent—yet the geopolitical dynamics revealed by DeepSeek’s success make such coordination more difficult.

For business leaders, the data demands a reassessment of global AI strategies. Markets with high adoption rates—the UAE, Singapore, the Nordics—offer environments where AI-enhanced products and services find ready audiences. Emerging markets, particularly in Southeast Asia and Latin America where growth is accelerating (Vietnam at 23.5%, Brazil at 17.1%), represent the next wave of opportunity. Organizations examining deep research systems and survey methodologies can use these diffusion metrics to inform market-entry timing and localization investments.

Key Takeaways for Business Leaders and Policymakers

The Microsoft AI Diffusion Report H2 2025 delivers a clear message: AI adoption is accelerating globally, but the benefits are distributed unevenly. For business leaders, the most actionable insights center on three themes. First, national context matters enormously—the same AI product will be adopted at vastly different rates depending on regulatory environment, language support, digital infrastructure, and cultural trust. Second, pricing strategy is a competitive weapon, as DeepSeek demonstrated that eliminating cost barriers can capture entire markets. Third, language capabilities are adoption multipliers, with the South Korean example showing that improved local-language performance can trigger exponential growth.

For policymakers, the report makes an implicit but powerful case for proactive AI strategy. The UAE and South Korea both demonstrate that deliberate government action—regulatory clarity, education investment, talent attraction, and institutional coordination—directly accelerates adoption. The U.S. paradox shows that infrastructure and innovation leadership alone are insufficient. The widening digital divide suggests that without intervention, AI could deepen global inequalities rather than reduce them.

The report closes with a challenge that resonates beyond the data: “The challenge ahead is ensuring that this next wave of innovation benefits more people, in more places, and not fewer.” Whether that challenge is met will depend on decisions made in 2026 by the governments, companies, and organizations now reading these adoption numbers. The methodology behind this report—population-normalized, telemetry-based measurement—provides the tracking tool. The strategic response remains up to the leaders who use it.