A case for universal access to artificial intelligence

As the real cost of generative artificial intelligence tools begins to appear for consumers, post initial launch, one thing is clear: artificial intelligence is expensive. Basic subscriptions remain relatively affordable, but the more useful these systems become, the easier it is to exceed what a basic plan provides. Businesses increasingly pay for API access, coding tools, higher usage limits, image and video generation, automated workflows and specialized models. For a large company, those costs are another software expense. For a freelancer, independent operator or very small business, they can become a meaningful barrier. That matters because artificial intelligence is increasingly capable of doing work that small businesses have historically needed employees, contractors or specialized firms to perform. A single person can now use AI to research a market, build a website, draft a business plan, analyze a spreadsheet, translate materials, create marketing assets, write basic software or organize customer information. The systems remain imperfect and often require professional review, but they can sharply reduce the cost of getting started.

There is a reasonable public-policy question here: if access to AI can increase the productive capacity of small businesses, should the federal government help small businesses pay for it? The United States has more than 36 million small businesses, according to the Small Business Administration. They employ more than 62 million people and account for an estimated 43.5 percent of U.S. gross domestic product. Even relatively small productivity improvements across that population could have significant economic effects. There is already evidence that AI can produce those improvements. In one large field study involving more than 5,000 customer-service workers, access to a generative AI assistant increased productivity by roughly 15 percent on average, with the largest gains among less experienced workers. That result is particularly relevant to small firms, where individual employees and owners often handle several functions at once and cannot simply hire a specialist every time they encounter a skill gap.

AI adoption, however, remains uneven. Surveys produce different figures depending on how use is defined. A 2025 U.S. Chamber of Commerce survey found that 58 percent of small businesses reported using generative AI, while more conservative Census Bureau measurements of AI use within actual business functions have produced figures closer to one in five employer businesses. Larger companies are also generally more likely to report meaningful adoption than smaller ones. The difference between those figures is useful. Many people have experimented with ChatGPT or another AI product; far fewer appear to have incorporated AI deeply into the operation of a business. Cost is not the only reason. Training, technical knowledge, usable data and time all matter. But cost is one variable that public policy could address relatively directly.

The federal government could create an AI Access Credit for small businesses below a defined revenue or employee threshold. Eligible businesses might receive a modest monthly credit redeemable with approved AI providers. The benefit should be denominated in dollars rather than tokens because tokens are a technical measurement whose price and usefulness vary considerably between models. A portable dollar credit would also encourage competition. Businesses could use it with a general-purpose AI company, a specialized accounting platform, a coding service, a design product or another qualified provider. The government would establish basic standards for security, privacy, data portability, transparent pricing and reliability, while allowing the customer to decide where the credit is spent. This distinction is important. A poorly designed program could easily become a large subsidy to a small number of dominant technology companies. A better one would subsidize the user and require providers to compete for that spending.

There are already precedents for this kind of approach. The federal National Artificial Intelligence Research Resource is building shared access to computing resources, models and other AI infrastructure for researchers, educators and innovators. Governments have also used technology vouchers, research grants, broadband subsidies and Small Business Development Centers to reduce the cost of technologies or expertise believed to have wider economic benefits. An AI credit would extend that logic to small-business operations. It should begin as a pilot rather than a permanent entitlement. Congress could, for example, authorize a three-year program for businesses below a certain revenue threshold, with participants receiving perhaps $25 or $50 per month. Researchers could then measure revenue, business survival, employment, administrative hours saved, new products launched and other outcomes. The important question would not be how much AI people consumed. It would be whether subsidized access produced measurable economic value.

There are also substantial arguments against the proposal. AI is becoming cheaper very quickly, and it is entirely possible that market competition will reduce the cost of useful access enough to make a subsidy unnecessary. If that happens, the program should shrink or end. Any serious proposal should therefore include automatic review and sunset provisions. Providers could also raise prices to capture the value of the subsidy, which would require pricing transparency and rules preventing participating companies from charging subsidized users more than comparable customers. Misuse is another concern. Generative AI lowers the cost of producing useful business material, but it also lowers the cost of spam, fake reviews and fraudulent content. Public money should not subsidize those activities. There are broader costs as well. AI systems require significant computing infrastructure and electricity, and wider adoption has physical consequences even when the software appears inexpensive. Labor effects also cannot be ignored, particularly for entry-level work involving research, drafting and administrative tasks that AI can increasingly perform.

I recently built Mental Health Walk Club, a functioning web app, using AI as a major part of the development process. I am not a traditional software engineer, and building something similar would once have required hiring a developer or finding a technical co-founder. AI helped translate the idea into code, troubleshoot problems and iterate on the product. That is one of the clearest economic arguments for wider AI access: it lowers the cost of building software and gives more people the ability to create apps, test ideas and start businesses.

None of those objections eliminates the underlying economic question. Large corporations are already buying enterprise AI systems. Venture-backed startups routinely receive cloud and computing credits. Employees inside well-funded organizations increasingly have access to powerful AI tools paid for by someone else. The people most likely to face a meaningful access barrier are those operating outside those institutions: freelancers, very small businesses and people attempting to start something with limited capital. A person with money can already hire a developer to test an idea, an accountant to build a model, a designer to create a prototype or a consultant to research a market. AI does not fully replace those professionals, but it can reduce the cost of an initial attempt. That is the strongest argument for universal basic AI access. The government already spends significant amounts encouraging small-business formation, technology adoption and workforce development. A limited, competitive and temporary AI Access Credit would be a relatively inexpensive way to test whether wider access to artificial intelligence produces enough additional productivity and entrepreneurship to justify the cost.