In the endless arms race of artificial intelligence, OpenAI has once again pushed the stakes forward. According to Bloomberg, the company plans to invest hundreds of billions of dollars to build a new ultra-large-scale data center near Savannah, Georgia, in the United States.

This massive facility will be located within a data center park in Effingham County. OpenAI has already signed a contract with Georgia Power to secure 3.2 gigawatts of energy supply—this level of power is enough to make it one of OpenAI's largest projects to date. According to the plan, hundreds of megawatts of electricity will be put into use starting in 2028, while the remaining part will continue to be built until 2032. This is not a one-time completion project, but a continuous construction that spans several years.

In order to qualify for a local incentive policy, OpenAI has committed to an investment of 20 billion US dollars (approximately 135.534 billion RMB at the current exchange rate), and this policy will soon go to a voting session. Sachin Katti, Vice President of Computational Strategy at OpenAI, calculated a larger figure: if this data center reaches its full capacity of 3.2 gigawatts, the total cost is likely to exceed 30 billion US dollars. In other words, the 20 billion currently committed is just a part of the final bill for this campus.

The data center in Georgia is just the tip of the iceberg. According to sources cited by the Wall Street Journal, OpenAI has raised its projected computing power expenditure through 2030 to nearly 750 billion US dollars (approximately 5.08 trillion RMB). Just a few months ago, this number was around 600 billion US dollars. Within a few months, the projection has increased by about 150 billion US dollars—this reflects both a re-evaluation of computing power demand and a public gamble.

For competitors also laying down AI infrastructure, such figures mean they have to recalculate their costs. When OpenAI bets on computing power for the next five years at the level of 750 billion US dollars, the entire industry's cost benchmarks may have to shift accordingly.