5 GW and 2 million accelerators: details of the OpenAI kilometer AI data center project have been published

In September 2024, the head of OpenAI, Sam Altman, approached the American authorities with an initiative that involved the construction of an entire 5-GW data center network in the country. Altman argued that this would help the US maintain its leadership in AI on a global scale and was a matter of national security. Datacenter Dynamics was able to obtain a report from OpenAI addressed to the National Telecommunications and Information Administration (NTIA), which revealed some details of the company’s ambitious plan.

According to OpenAI calculations, data centers with a capacity of 5 GW will occupy an area of ​​about 2.78 km2 (278 hectares) and use up to 2 million accelerators. The project could provide the United States with absolute global supremacy in the field of AI. For comparison, in 2023, AWS operated a data center with a total area of ​​3.55 km2 (355 hectares, including office space), and Microsoft’s data center capacity at the beginning of 2024 was “only” 5 GW, although the company expects to develop 2 more by the end of the year .5 GW. A number of companies are working on gigawatt data centers, but intend to build them in stages over several years.

Image source: Microsoft

OpenAI claims that a 5 GW data center will cost $100 billion (in expected 2028 prices), but will generate $40 billion each year. For every 100 MW of capacity, 80 employees will be needed, i.e. a total of 4 thousand permanent jobs will appear. The construction of data centers will require about 14 thousand people. The company also provided calculations of GDP growth in individual states and locations where the construction of new data centers is expected – from $16 to $20 billion. Jobs and costs for the creation of new energy infrastructure and semiconductor production were not taken into account.

According to OpenAI estimates, global funds are ready to invest about $175 billion in infrastructure. The company threatens that this money may be spent not in the United States, but in unfriendly countries. At the same time, OpenAI emphasizes that the development of new AI data centers will be of strategic importance, taking into account the fact that AI platforms are developing in parallel, for example, in China. The company noted that delay threatens to lag behind China and other countries that do not share the values ​​of the United States and its allies.

The company initially planned to implement the 5-GW Stargate project together with Microsoft, the largest sponsor and provider of AI infrastructure for OpenAI. However, recently relations between the companies have cooled somewhat due to competition between AI divisions and disagreements related to the speed of construction of Microsoft data centers. Along the way, OpenAI is seeking funding and legal support from US authorities and holding meetings with potential investors.

However, it is not enough to build data centers; they still need to be powered. In some states with a large concentration of data centers, the capacity of power grids will have to double in the next 10 years. It is estimated that $50 billion of investment in the energy system will be required by 2030. By this point, North American power grids are expected to have grown by approximately 200 GW, with 100 GW of new AI projects required. However, due to the change of the head of the United States, even current projects to increase energy production and delivery may be canceled, and Elon Musk even stated that he intends to reduce government spending by $2 trillion.