Stargate was to be the world’s largest AI investment, a $500 billion infrastructure project to “secure U.S. leadership in AI.” Not afraid to exaggerate, its main backer, ChatGPT creator OpenAI, promised that its ability to help people “use AI to improve humanity” will “bring significant economic benefits to the entire world.”
Now, OpenAI appears to be walking away from part of the deal, an expansion of its flagship data center across a vast tract of land in Abilene, Texas, one of the most visible manifestations of its frenzied investment in the chips and power plants needed to build and run AI. Negotiations over financing the project and a timeline for when the expanded capacity will come online have broken down.
This may be fine for OpenAI. Perhaps you can find other data centers. For Oracle, the project’s OpenAI partner, this isn’t all that great. Oracle has already spent billions of dollars on hardware for the site. This is one of several cracks appearing on the capital side of the AI economy that are making investors quite nervous.
Both companies say the development will not derail their AI plans. They also said a month ago, when another $100 billion deal collapsed between OpenAI and Nvidia, the world’s largest maker of chips that train AI models and answer the billions of questions people ask every day.
The fate of such transactions is becoming increasingly important to the global economy. According to Bloomberg, future data center leases agreed to by the biggest cloud computing companies (including Amazon, Oracle and Microsoft) have increased nearly 340% in two years and are now worth more than $700 billion. If this technology does not begin to deliver on its promise of significantly increasing economic productivity, there will be significant costs. On Friday, more than three years after the AI hype was unleashed with the launch of ChatGPT, the UK reported zero GDP growth in January.
On Monday, the Guardian exposed new rifts in the AI organization. Many of the UK’s major AI deals, announced to great fanfare during a state visit last September, turned out to be different from what was described in government and company press releases, an investigation found. Key projects are delayed or unfeasible, and key “investments” are actually vague agreements, primarily between American tech companies, desperately envisioned by ministers as engines of economic growth.
If the cracks in this data center boom widen, the implications range from the UK lacking the AI infrastructure it needs to sustain itself in the global economy, to the more serious risk that the entire AI bubble will burst in a repeat of the 2001 dot-com crash, which could upend the global economy.
“There is a lot of blind optimism when it comes to building AI infrastructure,” said Andy Lawrence, executive director of research at the Uptime Institute, which inspects and evaluates data centers. “While an incredible boom is underway, with construction taking place on a scale never seen before, it has also been clear for some time that many projects will not proceed, or will take much longer to build and become operational than many claim. The high stakes and rewards of AI have attracted speculators who promise to invest but have little experience in the field.”
Most emblematically, a Guardian investigation highlighted a site in Loughton, Essex, where the government announced it would be home to “the UK’s largest sovereign AI data center” by the end of 2026. The technology secretary at the time, Peter Kyle, called it “a new start for our economy and our workers.” A year later it was still being used as a scaffolding yard and the chances of it being open at the time of the claim were almost zero. Following an investigation by the Guardian, Nscale confirmed that it had purchased the land on which to build the computers, eight months after announcing the purchase in January 2025. The company has not yet received planning permission, but said on Friday it plans to start construction by July and have the data center operational between April and July 2027.
The volatile AI deal comes amid a growing embrace between US tech companies and senior US and UK politicians. Donald Trump’s top AI advisers include David Sachs and Sriram Krishnan, both of whom have recent backgrounds as technology investors. In London, OpenAI hired former Prime Minister George Osborne. Anthropic and Microsoft have hired former chancellor Rishi Sunak. Peter Mandelson was the owner of a consultancy that worked on behalf of Palantir. And the Tony Blair Institute is funded by Oracle’s billionaire owner Larry Ellison’s foundation.
These figures are helping to shape an AI policy in which the UK essentially agrees to become a transit point for US-designed hardware rented primarily to US tech companies. The UK government says it is creating a “sovereign AI infrastructure”, the definition of which is open to debate, from British-owned hardware and data to maintain control of critical national infrastructure in a world of unstable international alliances, to AI Minister Kanishka Narayan’s more flexible definition of “strategic leverage” so that the UK can “ensure continued access to critical inputs”.
In the UK, that means relying on the US. Nvidia CEO Jensen Huang said this during President Trump’s state visit last September. “The United States must lead across the entire AI technology stack.”
Former deputy prime minister Nick Clegg was more outspoken that week, calling Britain “technically a client state.” Mr Clegg was this week appointed to the board of UK company Nscale, which is involved in the Laughton AI deal. The company’s customer is Microsoft, which plays a part in US high-tech hegemony, and he lamented its power six months ago.
On BBC Radio 4’s Today program this week, Nscale’s senior vice president Imran Shafi was asked whether the company’s Essex data center would be up and running by “Q4 2026” as promised. He replied, “The live time will be the time agreed with the customer.”
Meanwhile, Narayan defended the pace of widespread progress. “What we are saying is that we are moving forward in a coordinated manner,” the minister told CityAM. “We already have a working data center in Lanarkshire. There are spades in the ground in parts of the North East.”
Mr. Narayan might consider the example of the current Texas meltdown. After billions of dollars of construction were promised, construction began, and billions of dollars worth of equipment was purchased, OpenAI pulled out, leaving its partners in the unenviable position of having to find another AI giant to work with.
OpenAI reportedly wanted a newer chip model, and by the time construction in Texas is complete, the hardware Oracle purchased may no longer be cutting edge. It was like buying a bunch of iPhones right before a much more powerful model was released. The pace at which chips are becoming obsolete casts a further shadow on the UK government’s insistence on large-scale AI investment. It describes “investments” in cash terms, which are primarily computer chips. Tips are not money. It will probably depreciate even faster than most technology companies expect.
This means when we bring our data center in Essex or our AI hub in Lanarkshire online is critical. By the time they are ready and additional power is procured, will the leap forward in the design of AI systems mean that running the chips of 2025 will be like owning a jet-age propeller plane? Or, if the announced deals relate to future chips, will they be available? Iranian drone strikes have already affected helium supplies from Qatar that chip makers need. What would happen if China interfered with supplies from Taiwan?
“Data centers, especially large, high-density AI centers, are very complex engineering projects,” said Uptime Institute’s Lawrence. “Very few come online within two years, and usually take longer. It’s not uncommon for some projects to be delayed for several years or even indefinitely.”
The last component here is the bank. It leverages chips from Nscale and chips from other data center companies. These operators have secured billions of dollars in financing based on their graphics processing units (GPUs). At least in Nscale’s case, this debt will be used to fund expansion in the UK, but when will that debt come due? If it can’t be paid, what happens to Nscale and other financial institutions that continue to find buyers for potentially expired chips?
An Nscale spokesperson said: “We work with established financial partners and maintain disciplined governance around our funding decisions. We take a conservative approach to funding in line with our long-term infrastructure development.”
“Because chips have a limited lifespan, people and financial institutions lending money are taking on more risk,” said Forrester analyst Alvin Nguyen.
The data center investment boom represents one of the biggest infrastructure bets of our time, or of any era. Whether Laughton’s scaffolding eventually becomes a bona fide AI factory may tell us a lot about who wins and who loses.
