AI与人类：解码60国AI使用情况

AI vs Human: Decoding AI usage across 60 countries The AI debate has largely centred on models and infrastructure, but we believe the moreimportant question is adoption. Ultimately, the extent and quality of AI usage will determinewho captures the resulting economic gains. Our proprietary LLMs vs. Coders study, whichanalysed AI usage across global tech companies, offered limited evidence that adoption hasyet translated into meaningful productivity gains. In this report, based on LLM usage stats,we examine how AI is being adopted globally. The results uncover notable divergences inadoption patterns and raise a broader question: can these early trends help identify thefuture winners and losers in the AI era? Venugopal Garre+65 6326 7643venugopal.garre@bernsteinsg.com Nikhil Arela+91 226 842 1482nikhil.arela@bernsteinsg.com Rising adoption, misleading comparisons:AI adoption has accelerated rapidly sinceChatGPT's launch in late 2022. By March 2026, nearly 18% of the global working-agepopulation was using AI, up from 15% just nine months earlier. The prevailing narrativeis that developed markets are pulling ahead while emerging markets risk falling behind,threatening the labor-arbitrage models on which many economies were built. At firstglance, the data appears to support this view. But per-capita usage is a poor measure.Emerging markets account for a disproportionate share of the world's population andemploy far more workers in agriculture and low-end manufacturing, sectors that areinherently less AI-intensive. When viewed through the lens of absolute usage within theinformation economy, the gap narrows dramatically. In fact, several emerging markets rivalor surpass many developed nations. The great usage divide:If the debate on adoption is largely settled, the more importantquestion is how AI is being used. Here, a striking divide emerges. Emerging marketsconcentrate AI usage in software development, writing, editing and education, whiledeveloped markets lead in sales, finance and healthcare applications. Middle-incomeeconomies also show a greater concentration of usage in a handful of domains, whereasdeveloped markets display a broader and more diversified adoption profile. The disproportionate gains from AI:This is where the findings become most surprising.Emerging markets have historically lagged on productivity, yet they appear to extractgreater value from AI. Users in EMs report average time savings of 4.6 hours per taskversus 3.8 hours in developed markets. More importantly, 16% of tasks delegated to AI arebeyond what users could have performed themselves, compared with 12% in high-incomeeconomies. The data points to two powerful trends: diminishing productivity returns as AIadoption matures, and the unintended consequences of excessive human oversight, whereworkers gradually lose the ability to effectively review and challenge AI-generated outputs. The hidden, uncomfortable truth?Developed markets use AI predominantly to enhancequality, while EMs devote more than half of AI activity to automation. The implication isuncomfortable: many of the routine, process-driven jobs most vulnerable to disruptionare already being automated, often within the very EMs that became global hubs for suchwork. The effects are beginning to show elsewhere too. Traditional workweek boundariesare blurring, with weekend AI activity among top-skilled workers rising by 8%. As softwareengineers shift from weekday debugging to weekend experimentation with increasinglyagentic systems, the industry's value pool may already be moving quietly but decisively upthe stack. DETAILS SUMMARY IMPLICATIONS: THE FUTURE OF AI USE Based on our analysis of the data and trends across over 60 countries (Ex China, as LLMs data are based on US LLMs), weexpect the following trends to play out in the medium to long term, as AI adoption scales up and peaks: 1. IT services industry will become less and less labor-intensive in the future, and much of their competition will emerge fromsoftware engineers within the same countries, and consulting firms who have positioned better towards tech and AI. Thehigh usage in the field of computers, and within that a disproportionate share in automation tasks, is a testament to that 2. Global AI adoption at 18% is not way off and might peak at ~55-60% in the long run. It actually may hit a structural ceilingwell before that in some large, emerging economies - given a big chunk of workforce in professions unexposed to AI, localidiosyncracies, plus language constraints. This starts making a strong case for local LLMs trained in vernacular languages,laws and customs specific to geographies and vernacular languages and aligned with local professions and norms. Weexpect a further sovereign push to AI, which will be instrumental in hitting the usage cap 3. Given a particularly heavy AI use in media industry already emerging, it could very well be among the next bastions to bedisrupted by AI (along with healthcar