David Solomon Says Most Businesses Don't Need Frontier AI Models, as Goldman Pours $500 Million Into Small Business AI Training

David Solomon, the chief executive of Goldman Sachs, has argued that most economic activity does not actually require the kind of frontier AI models dominating headlines and investment dollars. Speaking about where cutting-edge AI development matters most, Solomon pointed to cybersecurity, defense, and health care applications like drug discovery as the areas where frontier models earn their cost, drawing a clear line between that kind of model development and the practical tools most businesses actually need.

The comments came alongside a $500 million commitment from Goldman Sachs to expand its 10,000 Small Businesses program, which has graduated nearly 20,000 entrepreneurs since it launched in 2010. The expanded effort brings in partners including Anthropic, Google, Meta, and OpenAI to help small business owners use AI tools without needing frontier-level capability to do it.

Data from the program's graduates shows adoption climbing quickly even without that higher-end technology. Eighty-six percent of graduates now use AI in their businesses, up from 56% in 2024. Most of that usage is modest: 73% spend less than $100 a month on AI tools, and 86% rely on general-purpose chatbots like ChatGPT, Claude, or Gemini rather than custom-built systems. Only 15% have fully integrated AI into their core operations, and 84% reported no change to their workforce as a result of adopting it.

Solomon has encouraged small business owners not to rush their adoption of AI, framing it as a return-on-investment question that includes real costs, like the per-token pricing of AI model usage, rather than something to chase for its own sake. His advice, in short, is that getting the application right matters more than getting it done quickly.

Two small business owners who went through the program described AI as something that has changed how they work without replacing the people doing it, using the tools to handle administrative tasks, manage inventory, and support creative work rather than cutting staff. That pattern, modest spending, general-purpose tools, and augmentation rather than automation, is largely the opposite of the frontier-model arms race playing out among the biggest AI labs and cloud providers.