OpenAI and Anthropic were established by AI developers concerned about the unchecked advancement of corporate AI, particularly fearing that companies like Google and Meta might guide the technology towards harmful and potentially disastrous consequences for society. The founders asserted that their organizations would be devoted to developing AI in a manner that prioritizes human welfare. However, both companies have gradually succumbed to similar market pressures, evolving into substantial corporations focused more on maximizing shareholder value than serving the public good.
Recently, in June, both OpenAI and Anthropic announced their intentions to go public, sparking discussions about potential trillion-dollar valuations. This excitement has raised alarms regarding the potential for these companies to exacerbate global wealth inequality. In response, some analysts have suggested that the government should acquire a portion of these companies’ stocks to establish a sovereign wealth fund or redistribute their profits to benefit taxpayers directly.
Currently, the narrative has shifted to public dissent against AI data centers, alongside a decline in the stock value of AI chip manufacturer Nvidia. Additionally, SpaceX’s stock has plummeted shortly after its initial public offering. Questions are now emerging regarding the long-term profitability of leading AI laboratories. The creators of ChatGPT and Claude are facing significant challenges in achieving the substantial equity growth that once seemed guaranteed.
Evidence indicates that the market may be reevaluating the financial viability of these enterprises. If they fail to prove their worth on the financial stage, there could be an opportunity to redirect them back to their foundational missions. Should these AI firms falter in the marketplace, nationalization could be considered, transforming them into government-operated laboratories focused on serving public interests under democratic governance.
The economics surrounding major AI laboratories do not inherently ensure high returns on investment. Training advanced AI models is costly, and they quickly lose value when newer models are introduced. This narrow window for profitability is further complicated as enterprise clients become more adept at reducing their AI token usage. Moreover, since many AI models have started to behave similarly, competition has driven prices down. Additionally, open-source and Chinese alternatives, which closely follow the leading labs in performance, are often available for free, undermining the market position of companies like Anthropic and OpenAI.
Even when disregarding the expenses associated with model training, the current economic framework for AI raises doubts about long-term profitability. Numerous free and open-source models can be operated on local machines, from high-end servers to personal laptops and even smartphones, challenging the justification for the substantial capital investments these companies have made in data centers.
It is essential to recognize that OpenAI and Anthropic are not without value. Both organizations boast impressive teams of AI researchers and engineers who are consistently generating innovations that fuel a global demand for their products. While these labs may struggle to achieve profitability, their offerings are making significant positive impacts. Regardless of personal opinions on their technologies, the rapid growth in user engagement indicates that many would be disheartened if these companies were to disappear.
The core issue lies not with the individuals or the technologies but with the existing system. Under the current structure, OpenAI and Anthropic might not be seen as viable market investments. If the market determines they cannot deliver increasing financial returns for shareholders, their survival is at risk.
It may be that a for-profit model is not suitable for the development of AI. A potential solution could involve reverting OpenAI to its original non-profit status, a legacy that its founders sought to change, or even restructuring both organizations as university research centers, reinstating the high-caliber researchers they have attracted.
A more beneficial approach for society would be to establish public ownership and management of their productive capacities. Converting OpenAI and Anthropic into government agencies could facilitate the development of AI as a public good.
This transition would necessitate some reorganization. The companies could be divided into two distinct functions: product innovation and computational operations. The innovation side could be managed publicly, similar to national laboratories, with Congress providing stricter oversight than the unrestricted venture capital they currently enjoy. The U.S. has a proven track record of these types of institutions, which have led to groundbreaking advancements in various fields, including space exploration and telecommunications. Notably, Congress oversees a $200 billion research and development budget, highlighting a potential gap in frontier AI efforts.
On the other hand, AI operations could be treated as a public utility, akin to electricity or water services—locally or regionally owned, with national distribution and stringent regulations to balance fee collection with investment in infrastructure. While AI data centers differ from traditional utilities, the U.S. has a rich history of managing supercomputing centers at various levels.
Countries such as Switzerland, Spain, and Singapore have already established public AI laboratories. They also possess national supercomputing centers that offer public access to run AI models, similar to initiatives in Germany and Australia.
The advantages of public ownership are evident. With democratic oversight, critical AI models could become open, transparent, and responsive to societal needs, ultimately benefiting the wider community.
