Nvidia May Back OpenAI’s $500 Billion AI Data Centre. What Are We Building Here?

A half-trillion-dollar data centre sounds like finance news until you look at what is being built. One side sees thousands of jobs, new power stations and the next American industrial boom. Another sees an enormous gas-powered server farm built to run technology that may remove jobs faster than the construction site creates them. Then you have the people worrying about electricity bills, water, OpenAI’s losses and, somewhere at the far end of the conversation, the Terminator crowd preparing canned food for the robot apocalypse.

That is why this story matters. Almost every serious argument about AI has somehow ended up inside one proposed campus in southern Ohio.

Quick Answer

Nvidia has not agreed to back OpenAI’s data centre yet. The companies are reportedly discussing a $250 billion guarantee, alongside separate financing for up to $350 billion in Nvidia chips. The project could create thousands of jobs and power ChatGPT’s 900 million weekly users, but it is also an enormous energy and financial bet on a company that still does not make a profit.

What Nvidia and OpenAI Are Discussing

According to Reuters, Nvidia is discussing roughly $250 billion in financing guarantees connected to a proposed 10-gigawatt data-centre campus in southern Ohio.

The project would be developed by SB Energy, the energy arm of SoftBank, with OpenAI reportedly in talks to become its main tenant. Nvidia’s guarantee would support the lease and debt financing, giving lenders more confidence that somebody with very deep pockets stands behind the project.

The chips are separate. Nvidia is also reportedly discussing financing up to $350 billion of OpenAI’s chip purchases, which is convenient when Nvidia happens to make the chips OpenAI wants to buy.

None of this has been signed or confirmed by the companies. Reuters could not independently verify the original report, while Nvidia, OpenAI and the US Commerce Department had not commented when its article was published

These figures overlap, so please do not add them together and announce that OpenAI is spending $1.1 trillion in Ohio. AI companies are perfectly capable of producing ridiculous numbers without our assistance.

Why Does OpenAI Need Something This Big?

Because ChatGPT is huge, even if Anthropic has gained ground in coding and enterprise work.

OpenAI says ChatGPT has more than 900 million weekly active users and over 50 million consumer subscribers. That gives it one of the largest active user bases of any consumer product in the world. Anthropic is doing very well, particularly with developers and companies, but ordinary consumer usage is still a different sport. Sorry, Anthropic.

Every question, image, voice conversation, coding task and agent action requires computing power. Better models usually demand more of it, while new products encourage people to use AI for longer and more complicated jobs.

OpenAI also wants greater control over that infrastructure. It currently rents much of its computing capacity through Microsoft, Amazon and Oracle. Becoming the main tenant of a dedicated campus would give the company more direct control over the hardware, power and pace of expansion.

That part makes sense. The question is whether the money coming in can ever catch up with the amount going out.

OpenAI Has the Users. It Still Does Not Have the Profit

Treating OpenAI like some empty startup with a chatbot nobody uses would be stupid. The demand exists, the user base is enormous and the company says it is now generating around $2 billion in revenue each month.

The financial problem is that OpenAI can earn absurd amounts of money and still spend even more.

The Financial Times reported that OpenAI generated around $13 billion in revenue during 2025 while spending approximately $34 billion. Its reported net loss was much larger because of a major non-cash charge connected to its old investor structure, but even after removing that and other non-cash expenses, operational losses were estimated at roughly $8 billion.

OpenAI has also missed some internal revenue and user targets. In April, Reuters reported that CFO Sarah Friar had raised concerns about whether the company could meet future computing commitments if revenue did not grow quickly enough. OpenAI publicly rejected the idea that its leadership was divided over spending.

This does not mean OpenAI is failing. Many important technology companies spent heavily before becoming profitable. Amazon spent years being treated as an online shop that enjoyed losing money, and that worked out reasonably well.

The difference is the size of the bet. OpenAI is not asking investors to fund a few more offices and servers. It is discussing infrastructure spending large enough to reshape power grids, energy policy and Nvidia’s future sales.

The Job Argument Is More Complicated Than Either Side Admits

Supporters will point towards the construction work first, and they have a point. A campus of this size needs builders, electricians, engineers, pipefitters, security staff, network specialists, maintenance workers, power infrastructure and a large chain of local suppliers.

The US Department of Energy says the wider southern Ohio project should create thousands of jobs. New transmission lines, power generation and the cleanup of former federal industrial land would also bring work that exists outside the data-centre buildings.

Data centres are often criticised for creating many temporary construction jobs and relatively few permanent positions once the servers are running. That criticism is fair for some projects, but newer research suggests the local effect can be larger when several hyperscale facilities create a wider technology cluster.

A 2026 Brookings study found that counties receiving their first large data centre saw private employment and wages rise over the following years. Construction gained immediately, while larger hyperscale clusters also attracted IT services, telecommunications companies and contractors.

Then comes the uncomfortable part. The same campus creating those jobs will power AI systems designed to automate work elsewhere.

The International Labour Organization estimates that roughly one in four jobs worldwide could be changed by generative AI. Changed does not automatically mean deleted, but some companies are already cutting junior, administrative, support and content roles while promising that AI will handle more of the workload.

Even that claim needs some suspicion. The head of Adecco recently told Reuters that companies are sometimes blaming AI for layoffs caused by restructuring, weak performance or ordinary cost-cutting. Apparently, “AI did it” has become a more exciting way of saying the numbers were bad.

So yes, the project can create jobs while the technology inside it takes other jobs. Both statements can be true, which is less satisfying than choosing one team and screaming at the other, but closer to reality.

The Power Has to Come From Somewhere

Ten gigawatts is an extraordinary amount of power. OpenAI previously described 30 gigawatts as enough electricity for roughly 25 million American homes, which puts the proposed Ohio campus in the range of several million homes at full capacity.

The plan is tied to 10 gigawatts of new power generation, including at least 9.2 gigawatts from natural gas. According to the US Department of Energy, Japanese funding would support the gas plants, while SB Energy has committed to paying for $4.2 billion of new transmission infrastructure.

Supporters argue that the company will pay for the energy system it needs, make unused power available to the grid and stop ordinary Ohio residents from carrying the cost.

That promise matters because data centres have already become a major source of new electricity demand. The Department of Energy estimated that they consumed around 4.4% of all US electricity in 2023 and could reach between 6.7% and 12% by 2028.

Ohio’s consumer advocate has also warned that large data centres can push up electricity prices when the cost of new substations and transmission lines is shared with households. Ohio has introduced special rules requiring large data-centre customers to pay for most of the power capacity they reserve, even when they later use less.

Water is another concern, although we do not yet know which cooling system the proposed campus would use. Some hyperscale facilities consume millions of gallons of water each day, while waterless cooling can increase electricity use. There is no magic version where hundreds of thousands of hot chips cool themselves because the company added a green leaf to its sustainability page.

The environmental case is also more complicated than “AI will destroy the planet.” The International Energy Agency expects data-centre emissions to grow quickly, but still remain below 1.5% of total energy-sector emissions through 2035. AI may also reduce energy use in other industries through better planning and efficiency.

The project will have a real environmental cost. The debate is whether the economic and technological value produced inside the buildings will be worth it.

Nvidia Is Selling the Chips and Supporting the Buyer

Nvidia’s position is easy to understand. If OpenAI secures the campus, the buildings will need an almost offensive number of Nvidia chips, networking products and software systems.

By supporting the financing, Nvidia can make lenders more comfortable, help OpenAI obtain the computing power it wants and protect years of future hardware demand. This is business, not charity.

It also creates a strange financial circle.

Nvidia invests in OpenAI. OpenAI uses part of its funding to buy Nvidia products. Nvidia may guarantee the financing behind OpenAI’s data centre and could support the purchase of more Nvidia products inside it.

The user demand behind ChatGPT is real, but the financial structure still deserves attention. If Nvidia helps create the conditions for its own future sales, investors need to separate genuine customer demand from demand made possible through supplier-backed financing.

When everything grows, the arrangement looks brilliant. If OpenAI’s revenue disappoints, several companies may discover that they were passing the same risk around the table.

What About the Terminator Argument?

The proposed campus does not mean a robot army is about to leave Ohio and start asking people for their clothes, boots and motorcycle.

The serious safety argument is less cinematic. More computing power allows companies to train larger systems, run more agents and give AI access to increasingly important work. The models may become more useful, but failures also become more expensive when AI is connected to software development, finance, infrastructure, weapons, healthcare or government systems.

You do not need to believe that artificial intelligence will wipe out humanity to care about who controls this much computing power, which safety rules apply and whether commercial pressure is moving faster than oversight.

At the same time, calling every data centre the beginning of Skynet does not tell us much. The likely problems will probably arrive in boring forms first: bad automated decisions, security failures, convincing scams, disappearing junior jobs, higher power demand and companies trusting unreliable systems because replacing people looked cheaper on a spreadsheet.

That is less exciting than Terminator, but it is already close enough to real life to deserve attention.

The TGK Take

The $250 billion headline is easy to misunderstand. Nvidia has not transferred the money, no deal has been announced and the reported guarantee would cover financing connected to the lease and debt, rather than the Nvidia chips inside the buildings.

The larger question is whether a project costing more than $500 billion makes sense when its likely main tenant is still losing money.

OpenAI has the strongest answer any unprofitable company could offer: an enormous user base, fast-growing revenue and a product that has already changed how people work, search, write, code and waste time. If even a larger share of those users starts paying, or if AI becomes a normal part of every company’s software, the demand for computing power could be huge for decades.

The risk is that popularity and profitability are different things. ChatGPT can become one of the most-used products in the world while the cost of running and improving it remains higher than the money it earns.

Nvidia sees the upside because Nvidia gets paid during the building stage. OpenAI gets the computing power, SoftBank gets a huge tenant, Ohio gets investment and the lenders get Nvidia standing behind part of the deal. Everybody looks clever while demand keeps rising.

If growth slows, the same arrangement looks far less clever.

I would not call this a scam, and I am not preparing for Terminator. I would call it one of the largest bets ever made on AI becoming a profitable part of ordinary life.

The usage is already real. The profits are not, the environmental cost will not be imaginary, and nobody can yet give an honest answer about how many jobs the technology eventually creates or removes.

That is why we are talking about it.

Information current as of 27 July 2026. The financing discussions remain unconfirmed and could change or collapse.