Potential forecasting scenarios to guide policy proposals

A great deal of existing work forecasts the effects of AI’s impact on the U.S. economy and U.S. workers. This RFI is not seeking forecasts, but instead soliciting policies for how governments should respond to such potential future conditions.

Many policy experts hesitate to engage on policy design in this space due to the uncertainty of AI’s impacts. To encourage these experts to contribute, we have provided broad scenarios to which policymakers may have to respond, and we invite applicants to develop policies in response to one or more of these scenarios.

The Center does not endorse any of these scenarios, nor do we think any of these scenarios are inevitable. The right policy choices, made early enough, can bend the curve toward something better.

Scenario 1: The floor falls quickly and AI is broadly bad for all workers

2030: Real GDP ~$32–34 trillion · Unemployment ~8–10%

GDP. GDP growth does not collapse, but it is merely disappointing relative to the AI investment made, and deeply misleading as a welfare indicator. Displacement-driven unemployment produces a significant drag on consumer demand, partially offsetting the productivity gains captured by capital. By 2030, real GDP reaches only $32–34 trillion, as hysteria and demand destruction set in earlier than expected. A prolonged bout of higher unemployment and persistent structural mismatch between workers’ skills and those needed to use AI effectively leads to greater structural unemployment, which functions as a permanent tax on growth potential.

Unemployment. Displacement in exposed sectors — administrative, clerical, legal support, customer service, mid-level financial analysis, content production — arrives faster than new sector formation can absorb it. By 2030, headline unemployment reaches 8–10%, with structural unemployment concentrated in former knowledge-work occupations that have no clear successor category. The long-term unemployment population multiplies several times over as displacement accelerates. Productivity surges, but the gains flow to capital. Workers across the income distribution find that their output per hour rises while their hourly wages do not — or fall, as labor surplus depresses bargaining power economy-wide.

Exposure to AI automation. For roughly 40% of employment, at least half of all tasks could be replaced by AI in the future, and in this scenario that substitution happens on a compressed timeline driven by competitive pressure rather than measured adoption. Firms that hesitate lose market share to those that do not. Even “safe” occupations see wage erosion. The general surplus of labor, produced by displacement in exposed sectors, depresses wages in adjacent ones. Tradespeople, healthcare workers, teachers, and service workers who are nominally “AI-proof” find their wages squeezed by an economy in which large numbers of displaced knowledge workers are competing for any available work. The floor drops for everyone.

Scenario 2: The Great Divergence — AI splits the workforce; some are better off, some worse off

2030: Real GDP ~$35–38 trillion · Unemployment ~5–6%, but 10–14% in highly exposed occupations

GDP. In this scenario, real GDP reaches $35–38 trillion by 2030 as early AI productivity gains concentrate in software, professional services, financial analysis, and legal work. Rapid adoption of generative AI could add up to $2.84 trillion to U.S. GDP by 2030 above baseline, with gains concentrated in firms and sectors that successfully reorganize around AI-augmented workflows.

Unemployment. The headline unemployment rate by 2030 sits at a moderate 5–6%. But beneath that headline, occupation-level unemployment tells a radically different story. Exposed professional categories — paralegals, junior analysts, graphic designers, customer service managers, medical coders — face unemployment rates of 10–14%.

Exposure to AI automation. Exposure to AI automation rises with earnings until the 80th–90th percentile of earners — programmers, engineers, and professionals — where around half of the work could be performed by generative AI. Exposure is sharply lower for the highest earners, such as executives, athletes, and medical specialists, and also among the lowest for building and grounds cleaning (2.6%), construction (9%), and farming (10%). This creates an inverted-U exposure profile: the workers with the most to lose from AI are not the poorest or the richest, but the broad professional middle class.

Occupation-level winners. Workers in AI-adjacent technical roles see wages and opportunities surge. Global demand for AI-related roles is surging, with AI-related job postings in some regions offering wages roughly 30% higher than comparable white-collar jobs.

Occupation-level losers. The clearest losers are mid-career, mid-credential knowledge workers whose jobs are heavily task-exposed and who lack either the seniority to provide protection through managerial and relational functions, or the technical capacity to pivot.

Scenario 1: The platform giants win

The compute requirements for frontier AI are so enormous that only a handful of companies can afford to compete (e.g., OpenAI, Anthropic, Google DeepMind, and Meta, all resting on infrastructure owned by Microsoft, Amazon, and Google). The market ends up resembling mobile: two or three operating systems control everything, and building on top of them is the only viable option for almost everyone else. Startups either get acquired or become dependent on the giants’ APIs. The consolidation wave currently underway — where enterprises are spending more on fewer vendors — accelerates this outcome.

Scenario 2: The cloud-stack model

A few giants own the foundation — models and infrastructure — but thousands of companies thrive in the application layer, just as AWS dominates cloud but thousands of SaaS companies run on top of it. Vertical AI companies that own proprietary data in specific industries (legal, medical, finance, logistics) carve out durable niches the giants can’t easily replicate. Open-source models, particularly from Meta and Chinese labs, prevent any single company from monopolizing the foundation entirely, keeping the ecosystem healthier than the pure consolidation scenario. Enterprises end up with a small number of model vendors but a rich ecosystem of applications.

Scenario 3: Fragmentation and open-source wins

Open-source models close the capability gap with proprietary ones fast enough that the compute moat never fully forms. Companies run their own models internally, much the way they run their own databases today (no one pays Oracle for every query). The AI market ends up looking more like enterprise software in general: hundreds of vendors, no single dominant platform, lots of customization per industry.