人工智能 vs 人类 II：生成式人工智能颠覆手册——市场忽略了什么

AI vs Human II: The Gen AI Disruption Playbook - What MarketsAre Missing Most of the global commentary on Gen AI is focused on the AI supply chain but as adoptionhits scale, its disruptive impact will be hard to miss. Will the economic fallout be felt mostacutely in developed economies, and will that in turn trigger tighter controls on AI adoption?Amid the myths and noise around these questions, we set out here our data-backed view. Venugopal Garre+65 6326 7643venugopal.garre@bernsteinsg.com Nikhil Arela+91 226 842 1482nikhil.arela@bernsteinsg.com The dominant narrative falters:The most pervasive AI storyline says knowledgeindustries will be hit hardest, entry-level roles will disappear first, and low-cost laboreconomies—India, Indonesia, Mexico—face the greatest threat because their growth isbuilt on arbitrage. Yet these claims reveal a striking disconnect. Information and knowledgesectors account for only 1–4% of employment in most middle-income economies versus6–10% in high-income ones, and the gap in GDP share is even wider: services contributeabout 50–60% of output in middle-income countries (median 55%), compared with morethan 70% in advanced economies. This is the first serious challenge to the conventionalview that lower-income economies are most exposed. Enter wages, and the narrative breaks: The second pillar of the argument is wages,where differentials make the picture starker. A typical software engineer earns around10,000 dollars a year in India, versus roughly 130,000 dollars in the US and 80,000 dollarsin the UK. The pattern repeats across accountants, paralegals, data analysts—any white-collar category where AI is advancing rapidly. Replacing a 10,000-dollar worker with AIyields only marginal savings; replacing a 100,000-dollar worker is transformative for mostorganizations. And if firms adopt human-in-the-loop systems that retain a limited pool ofdecision-makers (and risk-bearers), which humans make the most economic sense to keep? Why the inverse market reaction?Despite this, equity markets in middle-incomeeconomies such as India and Indonesia have borne the brunt of the AI trade, while the USand Korea have surged. Are markets being irrational? Not entirely: large, pure-play AI namesare scarce in most emerging markets, and the cost-competitive labor model underpinningIndia’s services boom does face genuine disruption. Broad indices in middle-incomeeconomies also have higher exposure to IT services (around 5% versus 2.9% for high-income peers). These risks are real but finite. Markets are pricing the next few quarters, notthe next few decades—and that temporal myopia may be creating the next big blind spotwhere opportunity lies. AI, the great leveller?The title may be unexpected, but the logic is straightforward.Given current service-sector employment and existing income inequality, we estimate anaverage 10% hit to national income in middle-income economies, versus about 22% inhigh-income economies. The burden will fall disproportionately on the next-richest 9–10%income cohort after the top ~1%. In that sense, AI becomes a leveller along two dimensions:narrowing income gaps between high- and middle-income countries, and flattening incomedistributions beyond the top 1% within economies. DETAILS SUMMARY IMPLICATIONS Our analysis yields that if all things are left on autopilot, we might see a case where each firm surges ahead in AI adoption tillthe point the displacement from it exceeds the new jobs it can create. At that point, a real economic pyramid transformationwill happen, with developed economies being impacted first as well as more severely. Middle income economies will see lesserimpact in absolute value terms but will also see their chances of “upgrading” up the income cohort diminish to virtually zerowith time. We see an exceptionally rich top ~1% across economies which holds a disproportionate share of income, followedby a more equal remaining 99%. This can have devastating consequences for most discretionary consumer names acrosseconomies that rely on significant volumes, with only ultra-luxury brands serving the top 1% surviving somehow. Advancedeconomies have the most to lose. Practically, we believe AI to become an industry with tight government regulations - be it minimum employment mandates orrestrictions on usage. Eventually, AI can actually end up bringing income levels for most humans closer together, rather thanbridging it apart. In those terms, AI can be the great human leveller - for the good or bad. IS THERE SOMETHING FUNDAMENTALLY WRONG IN OUR AI ASSESSMENTS? We often sing laurels of the massive productivity benefits emanating from Artificial intelligence, or sound death knells for certainknowledge industries owing to the sheer disruption it brings. What we inevitably end up missing is the bigger picture. This iswhere we will attempt to fill that void. You will find this note a rather atypical one. Our thoughts will evolve as we proceed, andour inferences