Goldman Sachs AI Tracker: U.S. enterprise AI adoption rises to 20.6%, with investment and productivity signals continuing to strengthen

Overview

This report is Goldman Sachs' monthly tracking of AI investment, enterprise adoption, labor-market impacts, and productivity performance. The core conclusion is that AI-related capital expenditure and the hardware chain remain in a strong expansion phase, U.S. enterprise adoption continues to rise, with information, professional services, finance, and education leading; meanwhile, AI's impact on the labor market is already visible in specific roles, but has not yet resulted in broad unemployment shocks. On productivity, areas that have deployed generative AI are showing significant efficiency gains, and official data also indicate a modest acceleration in productivity growth in high-adoption industries.

Core views

The report argues that the AI theme remains in a stage of moving from investment expansion toward diffusion into actual enterprise adoption. Semiconductor, memory, AI hardware import/export, and South Korean hardware shipment data together demonstrate strong infrastructure demand; enterprise surveys show adoption continuing to improve and beginning to extend from general generative AI tools to AI agent systems. In the labor market, AI-related job substitution is not broad-based, but concentrated in occupations with already clear use cases, while data center construction is creating offsetting job growth. Productivity evidence is more positive, but still mainly comes from limited deployment scenarios and early industry-level signals.

Analysis framework

The report cross-validates AI investment, adoption, employment impacts, and productivity changes by integrating macro statistics, enterprise surveys, industry revenue expectations, import/export and shipment data, labor-market data, corporate earnings call transcripts, academic research, and company case studies.

Methodology notes

• Enterprise adoption trackingCensus Bureau Business Trends and Outlook Survey

  U.S. enterprise AI adoption rate

  The report uses the BTOS survey to track the share of U.S. institutions using AI in routine business functions, and focuses on expected adoption over the next 6 months. Starting December 4, 2025, the wording was changed from 'producing goods and services' to 'any business function,' so the report notes that subsequent adoption rates are tracked under the new wording.

• Industry exposureGS AI automation exposure score

  The share of work content in industry subsectors affected by AI automation

  The report compares AI adoption rates with Goldman Sachs' estimates of AI automation exposure across industry subsectors, finding that adoption rates remain strongly correlated with automation exposure.

• Labor market assessmentCross-validation of employment, unemployment, hours worked, and layoff indicators

  The marginal impact of AI on employment and hours worked

  The report uses data from BLS, IPUMS, the Census Bureau, and others, compares labor outcomes with 2015-2019 averages, and examines the relationship between AI adoption rates and employment growth, hours growth, youth worker unemployment, and corporate layoff announcements.

• Productivity assessmentAcademic research, company case studies, and official productivity data

  Labor productivity gains after generative AI deployment

  The report summarizes efficiency improvement estimates from academic research and company case studies, and compares them with official U.S. industry data to assess whether high-AI-adoption industries are experiencing faster productivity growth.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Semiconductors and memory chips

  Direct beneficiaries of expanding AI investment

  Strengths

  Revenue forecasts continue to be revised upward, with global semiconductor revenue expected to reach an annualized $826bn by the end of 2026, and memory manufacturers clearly benefiting from AI demand.

  Weaknesses

  Valuations and capex expectations may already largely reflect strong demand, and cyclical and supply expansion risks still need to be monitored.

  Comparison

  Compared with downstream software adoption, semiconductors and the hardware chain show a more direct and quantifiable investment boost in the current data.

  Risks

  A slowdown in AI capex, declining memory prices, or weakening hardware import or shipment data.

• AI hardware and data center supply chain

  Driven by AI infrastructure buildout

  Strengths

  South Korean AI hardware shipments hit new highs, U.S. net AI hardware imports remain elevated, and data center-related construction employment is up 251k versus 2022.

  Weaknesses

  Construction cycles are long, and demand realization depends on continued expansion of AI applications by cloud providers and enterprises.

  Comparison

  Data center construction is making a positive contribution to employment, partly offsetting AI-driven substitution pressure in service and technology roles.

  Risks

  Power constraints, construction delays, capex cuts, and volatility in hardware demand.

• Enterprise AI software and AI agents

  Driven by rising enterprise adoption rates and expanding use cases

  Strengths

  U.S. enterprise AI adoption has risen to 20.6%, and multiple enterprise surveys show that organizations are beginning to explore or deploy agentic AI.

  Weaknesses

  Many enterprises still face constraints such as legacy systems, unclear ROI, insufficient governance rules, and inadequate employee skills.

  Comparison

  Enterprise AI software is in the transition from pilot projects to production deployment, with maturity still below the infrastructure buildout reflected in hardware investment data.

  Risks

  ROI falling short of expectations, insufficient data quality, and security and compliance issues limiting scale.

• Labor-intensive service roles

  Facing the dual impact of AI automation substitution and efficiency gains

  Strengths

  AI tools can improve efficiency in editing, data analysis, customer service, marketing, and professional services.

  Weaknesses

  Employment drag has already appeared in marketing, graphic design, customer service, and some tech roles, while AI-related layoff announcements have risen to 38.6k people.

  Comparison

  Compared with the overall labor market, AI's impact is more concentrated in white-collar and technical occupations with already clear use cases.

  Risks

  If AI agent systems scale up, substitution pressure could spread from localized areas to a broader range of knowledge work.

• U.S. macro productivity

  Potential medium-term growth driver

  Strengths

  Academic research shows average labor productivity gains of 23%, company case studies show efficiency gains of about 34%, and productivity growth in high-adoption industries has accelerated slightly.

  Weaknesses

  Evidence still mainly comes from limited deployment scenarios, and the overall impact in official macro data remains at an early stage.

  Comparison

  Productivity signals are more positive than employment shocks, but the degree of diffusion and persistence still require more data confirmation.

  Risks

  AI adoption failing to penetrate business processes, gains remaining confined to localized use cases, or changes in statistical definitions causing judgment bias.

Key data

• Current U.S. enterprise AI adoption rate20.6%Up 1.1 percentage points from May; expected adoption rate over the next 6 months is 23.9%.
• AI adoption rate among large institutions40.7%Institutions with more than 250 employees continue to lead; adoption among mid-sized firms with 100-250 employees has accelerated in recent months.
• Global semiconductor company revenue forecast$826bnEquity analysts expect annualized revenue to reach this level by the end of 2026.
• Upward revision to memory producer revenue forecasts$173bnThe annualized increase in consensus revenue forecasts since ChatGPT's public launch in November 2022.
• Incremental U.S. AI-related investment$425bnAn increase relative to 2022 levels, equivalent to about 1.3% of U.S. GDP.
• South Korean AI-related hardware shipments$36bn/monthRose to a record high in May 2026.
• U.S. net AI hardware imports$52bnRemained elevated in April 2026.
• Increase in data center-related construction employment+251kUp since 2022 relative to broader construction employment trends; the recent 3-month trend is about +9k per month.
• AI-related layoff announcements38.6k peopleThe number of employees attributed to AI in corporate layoff announcements continued to accelerate in May 2026.
• Share of Russell 3000 companies mentioning AI and labor-related keywords24%Share of companies mentioning AI and labor-related keywords on 2026Q1 earnings calls.
• Average productivity gain shown in academic research23%Based on limited scenarios where generative AI has already been deployed.
• Average efficiency gain shown in company case studies34%Company case studies show somewhat higher efficiency gains than academic research.

Impact & implications

The report's implications for the AI industry chain and the macroeconomy are generally positive: demand remains supported for upstream semiconductors, memory, AI hardware, data center construction, and enterprise software; on the enterprise side, use is moving from experimentation toward production and agentic applications, which may continue to create opportunities for software and service providers. At the macro level, AI has not yet caused broad labor-market loosening, but it has already created employment and hours-worked pressure in some roles. If productivity gains spread from localized use cases to a broader set of industries, AI could make a more visible contribution to medium-term growth.

Risks

• High AI investment growth may pull forward expectations; if hardware demand or capex slows, semiconductors and the data center chain face correction risk.
• Enterprise AI adoption rates are affected by changes in survey wording; the new wording after December 2025 may cause a level shift, so historical comparisons should be made cautiously.
• Although AI's negative impact on the labor market is currently narrow, the spread of AI agent systems could extend it to more knowledge-work roles.
• At the enterprise level, issues around ROI, governance, security, data quality, and legacy systems may limit AI's transition from pilot projects to scaled production.
• The report does not provide investment recommendations on individual companies, and thematic conclusions should not be directly equated with stock ratings.

What to watch

• Whether current AI adoption rates and expected adoption over the next 6 months in the U.S. BTOS survey continue to rise.
• Whether semiconductor revenue forecasts, memory producer forecast revisions, and AI hardware shipments remain strong.
• Trends in U.S. net AI hardware imports, South Korean AI-related hardware shipments, and data center construction employment.
• AI-related layoff announcements, the share of tech employment, unemployment rates for young tech workers, and changes in hours worked in related roles.
• AI agent adoption rates in enterprise surveys, the share deployed in production environments, the degree of ROI realization, and the maturity of governance rules.
• Whether productivity growth in high-AI-adoption industries in official data accelerates further.