The AI race has run into an outlet: there is no more money for electricity

The artificial intelligence race in 2026 has faced a new barrier: even the most ambitious projects do not rely on computing power, but on the cost of infrastructure and access to electricity. Over the past two weeks, two events have occurred that clearly show that the AI boom is entering a phase where money and physical resources are becoming the main constraint. First, Nvidia revised the financial guarantees for the OpenAI data center project in Ohio, reducing them by more than half — from $250 billion to $105 billion. Then Nvidia's largest customers received notifications about an increase in prices for AI servers by more than 15% due to a sharp increase in the cost of memory. All the details are in the Izvestia article.

Reduction of guarantees from Nvidia

In 2025, Nvidia and OpenAI announced a grand partnership. The chipmaker has committed to invest up to $100 billion in OpenAI and supply chips for data centers. It was about the construction of 10 gigawatts of computing power, a project with a cost comparable to the GDP of entire countries. However, by August 2026, Nvidia's enthusiasm had noticeably waned. Instead of an initial financial guarantee of $250 billion, the company agreed to $105 billion. The official agreement, signed on August 17, set the ceiling of warranty obligations at this level.

An important caveat: Nvidia does not invest this money directly. She acts as a financial guarantor, covering OpenAI's lease and construction obligations to creditors. But the fact that the warranty has been reduced by more than half — despite the fact that Nvidia's market capitalization has already exceeded $5 trillion — speaks volumes. Even the market leader began to doubt the scale of the project. The guarantee covers only the land, electricity supply and the construction of basic infrastructure for an initial capacity of 4.25 GW. The cost of the Nvidia chips themselves, which will be required for this data center, may require separate financing — potentially up to $350 billion.

Trust, but verify: is AI capable of working with the human psyche

Izvestia tested ChatGPT and Claude after a study on the risk of increased psychological vulnerabilities by chatbots

Rising prices for AI servers and an energy infrastructure crisis

But even if companies have money for chips and servers, they face another problem: electricity. According to analysts, the construction of a 1 GW data center based on the Nvidia Vera Rubin architecture costs about $47 billion. And the annual electricity bill for such a facility is about $1.3 billion. The planned OpenAI facility in Ohio with a capacity of 10 GW will consume about the same amount of electricity as the entire state of Connecticut.

According to Goldman Sachs, AI infrastructure may require about $765 billion in investments by 2026. At the same time, the cost of building AI data centers is estimated at $ 15-20 million per megawatt - almost twice as expensive as traditional facilities. According to Gartner forecasts, global data center electricity consumption will reach 565 TWh in 2026, an increase of 26% compared to 2025. At the same time, AI-optimized servers will consume 31% of the total energy of data centers.

Investors' doubts

The revision of Nvidia's guarantees and the rise in server prices are signals that even the most optimistic forecasts for the AI infrastructure are beginning to be adjusted. Investors who have invested hundreds of billions in AI companies are starting to ask questions about the return on investment. Nvidia CEO Jensen Huang said in March 2026 that investments in OpenAI "may be the last." At the time, he explained this by saying that OpenAI would probably go public, and direct investment would no longer be needed. But now, looking at the reduction in guarantees and rising prices, it can be assumed that Nvidia had other reasons for caution.

The paradox is that AI is becoming both more affordable (OpenAI has already reduced prices for its models this year) and more expensive in terms of infrastructure. Building data centers, purchasing chips, and paying for electricity require investments that only the largest players can afford. The AI race is turning into an infrastructure race. And, as the events of recent weeks show, even the leaders of this race are beginning to doubt whether they can keep up with its pace.