Can Investors Trust AI Sales Figures? The Subsidy Problem Behind the Boom

Robert Pozen raises a critical alarm about how major AI companies are artificially inflating their growth numbers by essentially paying customers to adopt their products—a practice that echoes the catastrophic telecom equipment financing schemes of the late 1990s. OpenAI's DeployCo joint venture exemplifies the problem: the company is committing up to $1.5 billion of its own cash to push its enterprise tools into private-equity portfolio companies, while guaranteeing those PE firms a minimum 17.5% annual return. Anthropic is following suit with a $200 million contribution to a similar fund, and Google Cloud launched a $750 million subsidy program to drive adoption of its Gemini models among consulting giants like McKinsey and Accenture.

The troubling part isn't that incentive programs exist—Apple subsidizes phones through carriers, pharma companies rebate to pharmacy benefit managers—but rather the scale and opacity of these AI deals. When companies pay distributors to place their products, the line between genuine demand and manufactured growth blurs dangerously. Pozen draws a stark parallel to Lucent and Nortel, which lent billions to customers in the late 1990s to buy telecom equipment, initially booking these as legitimate sales. The strategy imploded spectacularly in 2000-2001: Lucent's stock cratered from $84 to 76 cents, and the company posted $16 billion in losses in 2001 alone. Nortel wrote down $16 billion and later paid $35 million to settle SEC fraud charges.

The stakes for AI investors are enormous. OpenAI is reportedly projected to lose $14 billion in 2026 despite surging revenue, and could lose up to $700 million annually if DeployCo underperforms. The real danger, Pozen argues, is that these subsidies mask whether AI adoption reflects true customer need or merely financial incentives that will evaporate. Private-equity portfolio companies may adopt AI tools not because they're valuable but because their PE owners mandate it. When subsidies end, so will much of the usage. Pozen calls for radical transparency: AI vendors should separately disclose what percentage of enterprise revenue comes from subsidized channels, reveal renewal rates for non-subsidized business, and clarify whether revenue is outcome-based rather than traditional licensing. Without this data, investors are essentially flying blind on whether AI companies have built sustainable business models or just elaborate financial engineering schemes.