AI Reshapes the Workforce: Banks See Mild Job Gains, Tech Hardware and Software Suffer Largest Net Losses

Key Conclusions

AI adoption is driving a structural restructuring of the global workforce rather than massive direct replacement. Based on two surveys covering 808 companies across the United States, United Kingdom, Germany, Japan, and Australia, AI has led to an average net job loss of 5% (second-wave industries) and 4% (first-wave industries) over the past 12 months, while delivering approximately 10% productivity improvement. Sector divergence is pronounced: U.S. banking added 5% net jobs, tech hardware lost 19%, and software lost 7%; Japan saw the heaviest net loss of 10%, while Germany gained 1%. Younger employees (2-10 years of experience) and outsourced/contract workers bore the brunt, while small firms actually achieved net job gains. Productivity gains concentrated in IT/software development, customer service, and finance functions; companies expect an additional ~2 percentage points of productivity improvement over the next 12 months. European AI adopters still trade at a 23% P/E discount versus their U.S. peers, and the earnings growth gap is widening—potentially a structural divergence not yet fully priced by markets.

Sector and Regional Differences: Banks Relatively Benefited, Tech Hardware Worst Hit

U.S. data clearly shows sector divergence: banking net added 5% jobs, professional services net lost 1%, semiconductors 6%, software 7%, and tech hardware 19%. In the second-wave cross-country survey, banking net lost 3% (lowest), software 7%, semiconductors 8%. Tech hardware and software are the most obvious areas for AI substitution effects, because their core tasks (code writing, testing, hardware design automation) can be directly replaced by large language models and generative AI. Banking benefits from AI-assisted customer service, risk control, and process automation, but front-office relationship-based roles are hard to replace, and U.S. banks already achieved net job gains after the first survey round, suggesting the industry is using AI to expand rather than shrink.

Asia-Pacific dispersion is higher: Japan net lost 10% (highest), with banking net losing 13%, software 9%, professional services 8%; Australia net lost 4%, but professional services lost 9%, software 7%, while banking actually gained 2% net. Within Europe: the UK net loss was around the average; Germany gained 1% net, with banking net losing 4%. These differences may relate to labor market rigidities, industry structures, and AI adoption stages across countries.

Firm Size and Employee Experience: Small Firms More Agile, Junior Workers Hardest Hit

Small firms (<50 employees) net added 4% jobs, while large firms (1,001-10,000 employees) net lost 9%. Small firms retained 61% of employees, large firms only 48%. This reflects that small firms deploy AI more lightly, have flatter employee structures, and face lower retraining costs, enabling them to quickly convert AI into incremental output rather than cutting positions.

By employee experience: workers with less than 2 years of experience had a net loss rate of 18%, 2-5 years 17%, and university graduates without experience about 13%; conversely, those with 21-30 years of experience lost 4% net, and those with 30+ years only 3%. Younger employees show the highest combined rates of reduction and non-replacement (16-19%), but they are also the focus of retraining and new hiring: 72% of companies plan to recruit employees with 6-10 years of experience in the next 12 months, and 65% plan to recruit those with 2-5 years. This means AI is reshaping labor demand structure: low-experience execution roles are being compressed, but demand for mid-level experience (with AI collaboration potential) is rising.

Productivity Gains: IT Functions Lead, Expected Acceleration but with Ceilings

The second-wave survey showed an average net productivity improvement of 9.6% (first wave: 11.5%), with software highest at 10.4%, banking 10.0%, and semiconductors lowest at 8.2%. By function, 70% of companies cited IT/software development as having the largest productivity gains, far ahead of customer service, finance, and operations. Companies expect an additional ~2 percentage points of productivity improvement over the next 12 months, but this expectation may be constrained by data readiness (the top barrier), AI skills shortages, and legacy system integration. Among U.S. listed companies, the share mentioning quantifiable AI benefits rose from 13% a year ago to 25%, with tech and financial sectors leading.

Valuation Discount for European AI Adopters: Earnings Gap Widens, Pricing Efficiency in Question

The gap in next-twelve-month EPS growth between European AI adopters and their own industries widened from about 1 percentage point at end-2024 to about 3.2 percentage points at end-2025, yet they still trade at a 23% P/E discount versus U.S. peers. Meanwhile, companies in Europe that reduced headcount posted a Sharpe ratio of 2.07 over the past two years, far exceeding the long-term average of 0.56, indicating the market's preference for AI-driven cost optimization. However, if European AI adopters cannot close the earnings growth gap, the discount may persist or even widen. Investors need to distinguish which companies have replicable, scalable AI monetization paths.

Japan and Australia: Distinct Paths of Technology-Driven vs. Compliance-First

Japan's AI adoption rate rose from 34% to 39%, with productivity gains of about 9-10%, but scaling bottlenecks and high costs rank as major barriers. The human role is shifting from direct supervision to model learning and optimization (supervisory role dropped from 49% to 26%, learning/optimization rose from 8% to 21%), implying increased trust in AI. Australia's adoption rate jumped from 34% to 53%, but trust and security risks remain the second-largest barrier, with the supervisory role staying high at 40%, indicating a more conservative strategy. Both countries struggle with data readiness and legacy system integration, but Japan's uniquely high cost issue may dampen scalability.

Key Divergences and Risks

1. Sample bias may overstate impact: The survey covers only companies that have adopted AI for at least one year, potentially amplifying AI's downside effect on employment and not reflecting a potentially flatter path for later adopters.
2. Subjective assessment confounds: Respondents' judgments may be influenced by the macroeconomic cycle or industry-specific structural challenges (e.g., tech layoff waves), making it difficult to fully isolate AI effects.
3. "Restructuring" does not mean "harmless": Combined reduction and non-replacement rates reach 27-28%. If these positions vanish permanently, long-term structural unemployment risk is significant, especially as generational income loss for younger cohorts may depress consumption power.
4. Productivity expectations may be absorbed: Companies expect an additional 2% improvement over the next 12 months, but actual productivity gains distribution shows 58% of firms achieved only 1-10% improvement, and high-magnitude gains (>20%) accounted for only 11%. If scaling bottlenecks persist, expectations may fall short.
5. European discount may not narrow: U.S. tech and financial sectors have more mature AI monetization. If European AI adopters' earnings growth gap continues to widen, the valuation discount could widen further.
6. Japan's scaling bottlenecks and costs: High and unpredictable infrastructure and licensing costs rank as the sixth-largest barrier to AI adoption, and a lack of clear path from pilot to scale deployment may cause Japan's AI investment returns to underperform expectations.

Investment Implications

• Sector allocation: Banking is the only sector achieving net job growth (especially in the U.S.), with significant productivity gains and relatively high visibility of AI monetization; avoid execution-intensive sub-sectors in tech hardware and software, where labor reduction may outpace new revenue creation.
• Size preference: Small-cap companies may benefit more directly from AI-driven operational efficiency improvements, and their workforce structures are less vulnerable to large-firm downsizing waves, but their cash flow prospects need verification.
• European value capture: European AI adopters trade at a 23% discount to U.S. peers with a widening earnings growth gap. This spread may persist until the discount factor deteriorates or the earnings gap closes. Focus on European companies with data readiness advantages and proven AI deployment paths.
• Asia divergence: Japanese AI adopters face unique cost and scaling barriers, and productivity improvement breadth lags Australia; short-term caution warranted. Australia could capture faster adoption dividends if it resolves skills shortages and trust issues.
• Labor thematic factor: The high Sharpe ratio of European "headcount-reducing" companies suggests AI-driven cost optimization is being actively priced, but investors should watch for crowding reversal risk and distinguish between active layoffs (AI-driven) and passive contraction (demand deterioration).

Appendix Data Summary