Back in September, OpenAI's Sam Altman shared a document with the White House pitching the economic benefit of a 5GW data center in a number of different states.

In a letter to the National Telecommunications and Information Administration (NTIA), OpenAI attached the previously unpublished report, revealing details about what it believes is needed to maintain US' AI edge.

The report is low on specifics, but does provide a few new data points: OpenAI's internal analysis estimates that a 5GW data center would span some 30 million square feet (2,787,100 sqm), and feature some two million GPUs.

Such a project would be the largest data center in the world by a significant margin.

In its latest 10-K filing, Amazon Web Services revealed that it operated a total of 38.2 million sq ft (3,549,000 sqm) of data center and office space in 2023. Rival Microsoft, meanwhile, had a global combined data center capacity of 5GW at the beginning of this year, but aims to add 2.5GW by the first half of next year.

Several companies, including hyperscalers, are working on 1GW data centers but expect to reach that figure in phases over multiple years.

The 5GW data center could cost $100 billion, OpenAI said, but bring in $40 billion annually. Dollar figures are based on the estimated value of a dollar in 2028, "to allow for time for construction."

At 80 employees per 100MW of power, the site could employ 4,000 people - as well as 14,000 for the construction - on average.

The generative AI company then looked at more specific job figures and GDP growth in 12 states, none of which included the jobs required for any new energy infrastructure or semiconductor manufacturing.

The states are: Arizona, California, Georgia, Michigan, Nevada, New York, North Carolina, Ohio, Pennsylvania, Texas, West Virginia, and Wisconsin.

Texas would lead to the most new jobs, at 48,279, while West Virginia was the lowest, at 31,671. Texas also had the largest economic impact, at $20,396,059,524, while West Virginia was the lowest with $16,186,312,913.

Against the backdrop of this potential project, OpenAI detailed "China's infrastructure surge at-a-glance," including new nuclear reactors, data centers, and the Digital Silk Road.

"Who will control the future of AI is the urgent question of our time," the report states.

"The rapid progress being made means that we face a strategic choice about what kind of world we are going to live in – one in which the US and allied nations advance a global AI that spreads the technology’s benefits and opens access to it, or one in which China and other nations that don’t share our values use AI to cement and expand their power?"

The report adds: "The infrastructure build needed to sustain the US edge on AI development is the kind of visionary undertaking for which the country is uniquely known and equipped to execute – a massive effort shaped by democratic values and designed to broadly distribute economic benefits. In contrast, China’s top-down, centralized AI infrastructure strategy presents a real and competitive alternative shaped by autocratic values that would deploy the technology and dole out the benefits in ways that cement its own influence."

In its separate letter to the NTIA, OpenAI said that it calculated that there was an estimated $175 billion in global infrastructure funds waiting to be committed. "The question isn’t whether that money will be spent, but where. If we don't channel it into infrastructure projects that will support democratic AI ecosystems around the world, the funds will flow to our global competitors."

An OpenAI source told DCD in September that the 5GW figure was "illustrative of the level of scale we and many believe is needed," but noted that it was not definitive. The entire company is worth $157 billion as of its latest fundraise, and is believed to be losing money.

The 5GW 'Stargate' project was initially reported by The Information in March as a potential partnership with Microsoft, the company's largest financial backer and primary cloud provider.

But the two firms have drifted apart, both as their AI divisions increasingly compete and due to friction over the speed of Microsoft's data center build-out.

OpenAI is now taking a more hands-on approach to finding data center capacity and will use a 1GW data center in Texas that is being built by Crusoe, used by Oracle, and then leased by Microsoft.

Alongside that, the company has also become more involved with the US government, seeking funding and regulatory support for ever-larger training clusters.

"Even if overall grid capacity grows by two percent annually, utilities still need to add well in excess of 100GW of peak capacity to a system that currently handles around 800GW at peak capacity. The increase in power demand will also likely be hyper-localized, potentially requiring a doubling of grid capacity over the next decade in areas with a concentration of data centers," OpenAI's head of infrastructure partnerships and policy Benjamin Schwartz said in the NTIA letter.

"Approximately $50 billion of investment through 2030, or roughly $7 billion annually, is needed to facilitate the new power generation alone. North America as a whole is expected to add ~200GW to the grid by 2030. The economically transformative AI models that benefit the United States and the rest of the world will use tens of GW of this additional power in coming years and may require 100GW of that power by 2030."

Schwartz previously worked on the CHIPS Program, which aims to fund a renaissance in US semiconductor manufacturing.

Earlier this year, the Biden Administration launched a Task Force on AI Datacenter Infrastructure while the Department of Energy stepped forward to help the sector find power.

But with the re-election of President Donald Trump, it is unclear what will happen with such initiatives. Trump has criticized CHIPS, instead threatening tariffs, and backer Elon Musk has said he will cut $2 trillion from government spending.