Why AI Frontier Labs Sell Tokens Instead of Building Their Own Companies

The Strategic Rationale for Selling Infrastructure

Frontier AI labs like OpenAI and Anthropic prioritize selling model access (tokens) over building their own product companies because they are effectively operating as platform providers in a gold rush. By selling the "shovels"—the underlying intelligence—they capture revenue from a broad ecosystem without assuming the specific operational, regulatory, and market risks associated with running individual businesses. This approach allows them to maintain a focus on their core mission of model development while gathering high-value data from diverse use cases.

The Operational Bottleneck of Business Execution

While AI models excel at specific, verifiable tasks like coding or text generation, they currently lack the holistic agency required to run a business from end to end. Successful companies require more than just software; they demand domain expertise, human-in-the-loop management, regulatory compliance, and complex relationship building. As one observer noted, "Most products and businesses fail because they build the wrong thing, have the wrong requirements, or have poor market fit." AI models can generate code, but they do not yet possess the judgment to navigate the multifaceted challenges of market entry and operational sustainability.

Platform Economics and Risk Mitigation

Selling access to a platform is fundamentally lower-risk than competing in downstream markets. By acting as a utility, these companies avoid the "cannibalization" problem—competing directly with their own customers. Furthermore, the token-based business model provides a steady stream of revenue and, crucially, a massive influx of semantic data. This data serves as a feedback loop, helping labs refine their models based on how developers and enterprises actually use them in the wild.

"Selling tokens is risk-free money since they get a cut for every app without taking any business risk," notes one industry observer.

The Future of Vertical Integration

While the current strategy is to provide infrastructure, there are signs that frontier labs are moving up the value chain. Some labs have already begun to integrate more deeply into specific industries, such as drug discovery or enterprise security, effectively acting as specialized consultants. This shift mirrors the evolution of other tech giants, which often start as platforms before absorbing successful features or verticals into their core offerings.

However, there are significant counter-arguments to this expansion:

• Diseconomies of Scale: A single organization can only manage a finite number of disparate business lines before operational inefficiencies degrade performance.
• Regulatory Scrutiny: Aggressively entering multiple markets could trigger antitrust concerns and alienate the very partners who provide the data and capital necessary for the labs' continued growth.
• The 'Jagged' Nature of AI: AI capabilities are inconsistent; they may be superhuman in one domain while failing in another. Relying on these models to run critical business infrastructure remains a high-stakes gamble.

Ultimately, the current strategy reflects a pragmatic realization: in a period of rapid technological flux, the most valuable position is to own the underlying intelligence layer, allowing the rest of the market to experiment, fail, and provide the data needed to build the next generation of models.