The Hidden Problem With AI Company Sales: Are They Real or Subsidized?

Robert Pozen raises a critical red flag about how major AI companies are driving adoption: they're essentially paying customers to use their products. OpenAI's new joint venture DeployCo involves the company contributing up to $1.5 billion of its own cash to get private-equity portfolio companies to adopt its tools, while guaranteeing partners a minimum 17.5% annual return. Anthropic is doing something similar with a $200 million contribution to a $1 billion joint venture, and Google Cloud created a $750 million fund to subsidize AI adoption by consulting giants like McKinsey and Accenture.

This matters because it obscures whether AI growth reflects genuine demand or financial engineering. Pozen draws a chilling parallel to the late 1990s telecom equipment bubble, when Lucent lent $7-8 billion to customers and Nortel contributed $3 billion—both companies eventually collapsed spectacularly when customers defaulted. Lucent's stock plummeted from $84 to 76 cents; Nortel paid $35 million to settle SEC fraud charges. The pattern is clear: when sellers finance their own sales and disguise incentives as revenue, investors can be fooled for years.

The transparency problem is acute. OpenAI is projected to lose nearly $14 billion in 2026 despite surging annualized revenue—a figure that gets harder to interpret when you don't know how much of that growth comes from subsidies. If the joint venture underperforms, OpenAI faces losing up to $700 million annually on its guarantee. Pozen argues investors should demand three specific data points: what percentage of enterprise revenue comes from subsidized channels, what renewal rates look like for non-subsidized customers, and whether revenue ties to actual outcomes or just traditional pricing. Without transparency, investors can't distinguish sustainable business models from schemes that collapse when incentives dry up.