Sam Altman Proposes 5-Gigawatt AI Data Center for US National Security | OpenAI stock | OpenAI free | OpenAI Playground | Turtles AI

OpenAI CEO Sam Altman is pushing to build a massive network of AI data centers across the U.S., which would consume up to five gigawatts of energy each, in an effort to convince the Biden administration of their importance to national security.

Key Points:

• Proposal to build AI data centers with the energy consumption of five nuclear reactors. 
• Data centers could give the United States a technological edge over China. 
• Significant challenges in energy procurement and supply of necessary GPU accelerators. 
• Tech giants like Microsoft and Amazon are already adopting extreme energy solutions for their AI projects.

Sam Altman, known for his ambitious technological visions, is back in the news with a proposal that could reshape the AI ​​landscape in the United States. The OpenAI CEO has presented the Biden administration with a plan to build a network of AI data centers, each of which would require a staggering amount of power: a whopping five gigawatts, equivalent to five pressurized-water nuclear reactors. Altman’s goal is clear: to give the United States a decisive technological edge over China, while also bolstering national security.

According to a document reviewed by Bloomberg, the proposal calls for building several such data centers across the United States, with an initial pilot project to test the feasibility of such an undertaking. But building even one of these facilities presents enormous challenges. Compared to the largest power plants in the United States, such as the 6.8-gigawatt Grand Coulee Hydroelectric Plant, the plants needed to power Altman’s data centers would be among the largest in the country.

Altman’s proposal comes at a time when demand for data center power is already on the rise. A recent report from CBRE highlighted how shortages of power and the equipment needed to manage it are causing delays in new data center developments. In response to these challenges, major cloud providers are exploring innovative solutions. Microsoft, for example, signed a 20-year power purchase agreement with Constellation Energy to restart the Three Mile Island Unit 1 nuclear power plant, while Amazon invested $650 million to acquire Talen Energy’s data center facilities, giving it access to nearly a gigawatt of power.

Another significant hurdle is the availability of GPU accelerators to power these data centers. Assuming 90 percent energy efficiency, a five-gigawatt facility could support up to 35,000 Nvidia rack-scale systems, equivalent to about 2.5 million Blackwell GPUs. However, Nvidia’s current GPU production is limited: In 2023, the company is estimated to have shipped around 600,000 H100s, with that expected to grow to 2 million chips per year. TSMC’s inability to meet demand for the CoWoS packaging that’s essential to building these chips could complicate matters further.

This isn’t the first time Altman has floated grandiose ideas: Earlier this year, he floated a $7 trillion plan to create a network of chip factories. While some of this may seem far-fetched, it’s clear that the OpenAI CEO is trying to spark serious discussion about the investment needed to support AI development in the future.

Sam Altman’s proposal could be a major turning point for America’s technological future, although the energy and logistical challenges to overcome are enormous.