Sam Altman: Computing Power Shortage May Become a Structural Problem
OpenAI CEO Sam Altman stated during a lecture in Stanford's CS 153 course that as model capabilities improve and costs decrease, the global demand for inference computing power may approach an infinite state. The computing power shortage is likely to evolve into a long-term rather than cyclical issue, especially after the widespread adoption of personal intelligent agents, where dozens to hundreds of concurrent agents will continuously occupy massive computing resources.
Altman views computing power as one of the most important public infrastructures of the future, emphasizing that once the price of computing power spirals out of control due to severe supply-demand imbalance, "how to fairly allocate computing power" will become a real societal conflict rather than a theoretical issue. He believes that there is a need to redesign the supply investment and allocation mechanisms for computing power at the institutional level, rather than simply relying on market self-regulation.
Regarding the distribution of social wealth and control, Altman estimates an 80% probability that the path will lead to widespread technology distribution and decentralized power democratization. However, he also warns that forces operating under the guise of "safety" and "stability" will push AI capabilities and capital to be highly concentrated among a few giants. He clearly stated that even if OpenAI itself is part of this power structure, the world should not tolerate a few companies monopolizing the vast wealth generated by AI.
Altman expressed that he prefers the model of a "citizen wealth fund" that allows the entire society to hold capital shares over direct cash distribution through universal basic income (UBI). He believes that in the context of social leverage shifting from labor to capital, allowing the public to collectively hold corporate equity and share long-term capital returns is a more sustainable wealth distribution path, helping to embed productivity dividends into the balance sheets of all citizens in the AI era.
Source: Public Information
ABAB AI Insight
Historical behavior shows that Sam Altman has been experimenting on two dimensions since his early role as president of Y Combinator, promoting projects like Worldcoin and OpenAI: one end is "extremely scalable technology platforms," and the other is the institutional design of "redistribution and social contracts"—he has publicly supported UBI experiments while also promoting new distribution frameworks closer to "digital identity + assets" through Worldcoin. His recent emphasis on "citizen wealth funds" essentially acknowledges that if capital returns are the dominant form of dividends in the AI era, then directly making the public asset holders is preferable to simply distributing cash.
From a capital perspective, defining computing power as a "public utility" rather than "internal corporate resources" is paving the way for two routes:
One is the construction of ultra-large-scale computing and energy infrastructure involving state and sovereign funds, treating GPUs, dedicated chips, data centers, and supporting energy as foundational assets for investment and regulation, similar to power grids and highways;
The other is embedding the equity of these capital-intensive assets into residents' asset sides through "citizen wealth funds," allowing society to share productivity not just through wages but by obtaining compound returns through long-term holdings of computing power and AI platform equity. Altman's statements lay the theoretical groundwork for a deep binding of "AI–computing power–capital markets–sovereign wealth."
Historically, in the past century, oil, electricity, and communication networks have entered a structure of "quasi-public utilities + oligopoly + state regulation":
In the oil era, pricing power was concentrated in the Seven Sisters and oil-producing countries, later restructured through OPEC and national oil companies;
In the telecommunications and internet era, it was dominated by tech and network giants like AT&T, Cisco, and Google, with continuous regulatory and antitrust pulls.
By placing AI computing power on the same level, Altman indicates that the future control of large models and data centers will not be limited to "service-selling companies" but will also include structural powers similar to resource-rich countries and grid operators. This explains why he repeatedly emphasizes that "the world should not tolerate a few companies monopolizing AI wealth," as that would equate to handing over future "digital oil" and "smart grids" to a few shareholders.
In terms of structural judgment, the key difference between "citizen wealth funds" and UBI is that UBI is "cash flow redistribution," while citizen wealth funds are about "rewriting the asset side." As social leverage shifts from labor to capital, if the public only receives wages + UBI without holding production means (AI platforms, computing infrastructure, data, and model equity), wealth gaps will continue to widen on the asset side; allowing the public to hold a basket of AI and computing assets as citizens or global citizens equates to transforming the "excess capital returns" of the AI era into a widely shared dividend mechanism. This logic is similar to Norway's sovereign fund or Singapore's Temasek, only the assets shift from oil and real estate to large models and data centers.
On a more micro level, Altman's concerns about "computing power allocation" directly foreshadow several inevitable conflict points in the future:
Between enterprises: Leading companies will compete for cheap, high-priority inference and training computing power; whoever can lock in lower-cost computing power long-term will suppress competitors in product and price wars.
Between nations: Countries with energy, land, and capital will attempt to treat computing power as a new type of "export" and "geopolitical leverage," similar to liquefied natural gas and rare earths.
Between individuals and institutions: If personal intelligent agents require long-term concurrent operation and computing power prices soar, ordinary users may be priced out, only able to use "weak agents" with strictly controlled quotas by platforms, while true "high-IQ agents" will only serve a small number of institutions with large-scale capital and computing contracts.
The combination of "citizen wealth funds + computing power public utilities" proposed by Altman attempts to design an acceptable social solution for the impending conflicts, but its implementation will inevitably clash deeply with existing capital structures and national interests.