A recent breakthrough in technology has been made by Chinese startup DeepSeek with the creation of its new AI-based model, “AR1,” which has revolutionized the tech world. Like the top models from Google and OpenAI, AR1 demonstrates the same level of functionality. However, it was developed with relatively less reliance on hardware. If this technology becomes more widespread, it could reduce the massive amounts of electricity and computational power required for AI data centers. As a result, tech giants might scale back their investments in nuclear energy, leading to concerns among experts and investors about the potential slowdown in the nuclear energy renaissance.
DeepSeek claims that the AR1 model was trained for two months using just 2,000 of Nvidia’s H800 GPUs, a significantly smaller amount compared to what OpenAI uses. Following this news, Nvidia’s stock price dropped by 16%.
On the other hand, various companies and startups had invested a significant amount of money in building nuclear power and natural gas-based new power plants for AI. However, DeepSeeks success may cause concern for them, as the need for so many data centers for AI might no longer be necessary.
In recent years, nuclear energy has been on the verge of a renaissance due to innovations in fuel and reactor design, promising to make new generation power plants safer and more affordable. However, there has not been a strong push to advance nuclear energy so far. Nuclear power remains more expensive than wind, solar, and natural gas. Additionally, next-generation nuclear technologies have not yet been commercially tested.
The demand for electricity to power AI technologies has increased significantly. Data centers are projected to consume up to 12% of the U.S. energy supply in the future, which is three times more than in 2023. By 2027, there will be a shortfall in the energy needed to run AI data centers. To address this, technology companies are making significant investments. Google has promised to purchase 500 megawatts of power from the nuclear startup ‘Kairos,’ Amazon has invested $500 million in ‘X Energy,’ and Microsoft, in partnership with Constellation Energy, is working on the renovation of the ‘Three Mile Island’ nuclear reactor with an investment of $1.6 billion.
There is no guarantee that the development of AI technology will solely rely on increased computing power. While this approach worked well for some time, recently, even with the use of more computing power, no significant differences have been observed.
AI researchers are actively exploring various avenues to find a solution, and it seems that DeepSeek may have found a potential solution with their AR1 model.
Atif Malik, an analyst at Citigroup, said, “While DeepSeek achievement could be groundbreaking, we question the notion that its success has been achieved without using advanced GPUs.”
Although DeepSeek may be withholding some important details, it is likely that another AI will find a way to make this technology cheaper and more efficient. Researchers will find it easier and quicker to develop new types of models than to build new nuclear reactors.
The likelihood of new nuclear reactors being operational before 2030 is very low. New fuel-based power plants won’t be available until the end of the next decade. Technology companies’ investments in the nuclear sector are primarily a safety measure, as they will turn to this alternative if software projects fail.
However, if technology companies had to choose between nuclear power plants or software, in most cases, they would opt for the second alternative.
The question now is—what impact will this new technology have on nuclear startups and electricity supply companies? Some companies might be able to produce electricity at such a low cost that even if the energy demand for AI decreases, it will not pose a problem for them. Moreover, even before the popularity of AI technology, there were concerns about increasing electricity demand.
However, if the demand for electricity in AI technology decreases, the cost pressure is likely to increase further. On the other hand, wind, solar energy, and batteries are becoming increasingly cheaper. These can be created more easily and in smaller sizes, with faster production processes.
Additionally, developers can implement new renewable energy projects in phases. Even before the entire project is completed, they can supply electricity (and revenue). However, this is not the case for nuclear power plants or gas turbines. Technology companies are investing in safe renewable energy sources to supply power to data centers, knowing this.
Liberty News’ tech correspondent reports that technology analysts have been unable to make predictions about AI technology. It is also uncertain how the tech world will evolve in the next five years. Therefore, companies are likely to invest in proven technologies. Currently, renewable energy is meeting this demand.
LND/SAKIB
