OpenAI's Fate: Why Token Selling Is a Broken Business

Why David Shapiro Is Bearish on OpenAI’s Business Model

David Shapiro offers a contrarian take on OpenAI’s $500B+ valuation, arguing that the company’s business model is fundamentally broken. His thesis: OpenAI is essentially a utility company selling commoditized tokens, but valued like a tech monopoly. The analysis is bearish but well-structured around four “failing pillars.”

On the AGI lottery ticket: “Investors aren’t buying a business. They are buying a chance at a singular outcome: the invention of the digital god. A valuation of over $500 billion implies that OpenAI has already won the future.” The critique: the valuation prices in winning a race that hasn’t been run.

On the missing moat: “Everyone from Google to Microsoft to DeepSeek has proven there’s no secret sauce to AI. Gemini 3 has surpassed ChatGPT across most benchmarks. There are better video generators and image generators out there. OpenAI doesn’t have a technological moat.” The competitive reality: model parity arrived faster than valuations assumed.

On being a utility: “They are essentially an electric utility. They spend billions on power plants—data centers—to sell electricity—tokens. History shows that utilities are low margin capital intensive businesses, not high margin tech monopolies.” The business model critique: utilities don’t justify venture valuations.

On “too cheap to meter”: “When Sam Altman said ‘intelligence too cheap to meter,’ the metering is the only thing that gets them revenue. When marginal cost of your product approaches zero, you cannot service a trillion dollars in debt.” The nuclear power analogy: cheap electricity can’t pay back expensive plants.

On who actually wins: “Value moves from the power plant to the grid and the appliances. The winners are hardware (Nvidia, Apple), cloud (AWS, Azure), electricians (Accenture, Palantir doing the integration work), and appliance makers (vertical AI companies using free tokens).” The reframe: token generators lose, token users win.

6 Insights on AI Token Economics and OpenAI’s Future

• No moat, no ecosystem - OpenAI sells tokens while competitors have operating systems, smartphones, and business software to integrate AI into
• Model parity is real - Gemini 3 surpassed GPT, Claude beat it on coding, and enterprises are building on Llama/Mistral 70B they can own
• Utility economics don’t work - Spending $100B+ annually on infrastructure to sell commoditized tokens isn’t a venture-scale business
• Open source is the solar model - The “nuclear age” of centralized data centers gives way to edge AI on phones, laptops, and cars
• The value chain shifts - Winners are hardware makers, cloud landlords, integration consultants, and vertical AI companies—not token generators
• Three possible fates - IP strip-mine by Microsoft, WeWork-style implosion, or IPO exit before unit economics are exposed

What This Means for Building on AI Infrastructure

Shapiro’s analysis inverts the standard AI narrative: the problem isn’t capability but business model. If tokens become commoditized (as they demonstrably have), and if open source models run locally (as they increasingly do), then the value shifts from the “power plant” (OpenAI) to the “appliances” (vertical AI companies using cheap tokens). This doesn’t mean AI is a bubble—he explicitly argues the opposite—but that the supply side structure is broken. For organizations, the implication is to build on commoditized AI rather than bet on any single provider winning.