The AI bubble theory has been rampant recently, but data revealed by *The Information* suggests that the so-called AI bubble phase is not an indisputable fact.
Internal data shows that OpenAI's computational profit margin for paid users has increased to 70%, providing strong empirical evidence to refute the "AI bubble theory."
Computational profit margin refers to the revenue share earned by running AI models for paid users. For reference, the computational profit margin of general software businesses is generally over 90%. The reason is simple: providing an additional software service does not require much additional computational cost, while large AI model companies need to incur significant computing power costs to meet the model usage demands of paid users.
This leap in growth means that, through model optimization and adjustments to subscription tiers, top AI vendors have proven that their core businesses possess profitability potential similar to that of the traditional software industry.
The anxiety about cost reduction and efficiency improvement initially stemmed from the impact of low-cost competitors like DeepSeek in early 2025, forcing OpenAI to shift its focus from simply pursuing parameter scale to extreme inference cost control.
Meanwhile, Google's inherent operational efficiency advantage, stemming from its self-developed TPU chips, has put OpenAI, reliant on expensive Nvidia chips, at a slight disadvantage in this cost tug-of-war.
This demonstrates that the AI industry is at a turning point, transitioning from a "technology illusion" phase to an "efficiency delivery" phase. The second half of this race is not just about whose model is smarter, but also about who can be the first to solve the problem of "computing power consuming money."
Whether the bubble theory will completely dissipate depends on whether these companies can extend their current profit margin advantage from a niche market of paid subscriptions to a broader market of hundreds of millions of free users.
However, the recent valuation decline in US stocks, driven by AI bubble theories, is expected to ease in the near future. The next step depends on whether OpenAI can raise enough money in the Middle East. Successful fundraising means fulfilling the large orders previously placed for data centers; otherwise, the entire AI industry chain may need another round of bubble deflation.
