TLDR: The Penn Wharton Budget Model projects that generative AI will substantially boost global productivity and GDP, with a 1.5% increase by 2035, nearly 3% by 2055, and 3.7% by 2075. The strongest annual productivity growth contribution is expected in the early 2030s, peaking at 0.2 percentage points in 2032, before gradually fading. The report highlights that 40% of current GDP could be affected by AI, with mid-to-high earning occupations being most exposed. However, caution is advised due to limited current data.
A new analysis from the Penn Wharton Budget Model (PWBM) forecasts a significant and lasting impact of generative artificial intelligence (AI) on global productivity and Gross Domestic Product (GDP) over the coming decades. The model projects that AI will lead to a 1.5% increase in productivity and GDP by 2035, nearly 3% by 2055, and a substantial 3.7% by 2075.
The report, published on September 8, 2025, indicates that the most pronounced boost to annual productivity growth from AI is anticipated in the early 2030s, with a peak contribution of 0.2 percentage points in 2032. Following this peak, the annual contribution is expected to gradually diminish, settling into a permanent effect of less than 0.04 percentage points due to shifts across economic sectors.
Key findings from the PWBM study reveal that approximately 40% of current GDP could be substantially influenced by generative AI. The analysis suggests that occupations around the 80th percentile of earnings are the most susceptible, with roughly half of their work activities potentially automated by AI. Conversely, the highest-earning and lowest-earning occupations are projected to be less exposed to AI automation.
Generative AI technologies, such as large language models (LLMs), are increasingly being integrated into tasks requiring digital tools and information processing. The PWBM’s methodology combines a task-based framework with a projected adoption timeline, drawing parallels from the historical diffusion of technologies like the commercial web and cloud computing services.
In terms of economic activity exposed to AI, the study defines a job as exposed if at least 50% of its activities could be automated by generative AI. Based on detailed classifications, around 42% of current jobs are identified as potentially exposed. While only about 1% of jobs are deemed ‘completely exposed’ (performable entirely by AI with minimal human oversight), over a quarter of U.S. employment could see AI perform between 90% and 99% of the required work.
The relationship between wages and AI exposure is notable: occupations at the lower end of the wage spectrum, often involving manual labor or personal services, are the least exposed. Exposure generally increases with earnings, peaking in the 80th-90th percentiles, which include professionals like programmers and engineers. In these roles, about half of the work could be performed by generative AI. This proportion then declines sharply for the highest-earning occupations, such as business executives and medical specialists.
Combining employment, wages, and exposure data, the PWBM estimates that 40% of current labor income is potentially exposed to generative AI automation. This figure is nearly double previous estimates, a difference attributed to varying interpretations of automation exposure metrics. The report assumes that the share of GDP exposed to AI mirrors that of labor income and that 23% of exposed tasks will eventually be profitably automated. This translates to just under 10% of current GDP being impacted over time, a share projected to grow to around 15% in the next two decades due to faster growth in AI-exposed sectors.
Regarding cost savings, the model assumes an average labor cost saving of approximately 25% from adopting current AI tools, based on real-world applications. This is projected to increase to 40% in the coming decades. Examples of observed gains include a 14% increase in task completion rates for customer service with AI assistants, a 17% increase in job starts and 18% in retention rates for job interviews with AI voice agents, and a 40% increase in speed for basic professional writing with ChatGPT-3.5.
The timeline for AI adoption is crucial. Data from Bick et al. (2025) indicates that 26.4% of workers used generative AI at work in the latter half of 2024, with 33.7% of adults using it outside of work. These early adoption patterns are similar to those of personal computers in the early 1980s. The PWBM projects that generative AI will be almost fully integrated into production within the next 15 years, with gradual, diminishing adoption thereafter.
Suggestive evidence from the labor market indicates that AI adoption is already having an effect. Since 2021, job growth has stagnated in occupations with the highest AI automation potential. For jobs entirely performable by AI, employment fell by 0.75% in 2024 compared to 2021. Occupations with 90-99% AI exposure have seen a significant slowdown in employment growth since 2022.
Also Read:
- Generative AI Reshapes Content Creation: Opportunities, Challenges, and Market Growth in 2025
- Capgemini Report: Generative AI Adoption Soars, Reshaping Enterprises by 2025 Amidst Governance Challenges
Despite these projections, the PWBM emphasizes the need for caution due to the early stage of generative AI development. The current analysis does not account for potential changes in product quality, the emergence of new products and labor tasks, or AI’s impact on innovation. These estimates will be updated as more data and information become available.
