The joint venture, known internally as “Project Braid”, was established with $5 billion from Blackstone and is expected to begin renting Google’s artificial intelligence processors to customers in 2027.
The business is part of a rapidly expanding group of so-called “neoclouds” — companies that lease specialised AI chips and computing capacity to businesses seeking access to the infrastructure required to develop and run AI applications.
However, the venture has encountered difficulties at several proposed locations, raising questions about how quickly the massive amounts of infrastructure required to support the AI boom can actually be built.
One of the early setbacks involved a proposed data centre development in Cheyenne, Wyoming, where Google abandoned a plan for upstart data centre developer Crusoe to construct facilities for the project, according to people familiar with the matter who asked not to be identified because they were discussing confidential information.
Google subsequently took over the project and is reapplying for permits after losing confidence in Crusoe’s ability to deliver the site, the people said.
Local officials have also told residents that the project will be scaled down.
Another potential location was deemed unsuitable because it lacked the appropriate transformers, one of the people said. The shortage of such equipment has emerged as a major constraint on the global data centre industry, with developers facing waiting periods of nearly a year for some transformers, according to a McKinsey report.
The situation has been further complicated in Texas, where a political backlash against the rapid expansion of data centres has disrupted development plans.
Some plans were thrown into uncertainty after Texas Governor Greg Abbott ordered a freeze on new data centre projects while the state gathers more information about how developers and technology companies contribute towards electricity costs.
The developments underscore the growing difficulties facing companies seeking to translate soaring demand for AI computing into physical infrastructure.
JV executives remain confident
Despite the setbacks, executives overseeing Project Braid have sought to reassure investors and potential customers that the venture remains on schedule.
Benjamin Treynor Sloss, the venture’s chief executive officer, said the differing pace of individual projects was normal in the data centre industry.
“It is very common for individual potential sites to progress at different rates,” Sloss said in an emailed statement.
He said the venture had anticipated such challenges in its planning and had developed a “diversified set of options” to reduce the impact of delays at individual locations.
According to Sloss, Project Braid is making “strong progress” and remains on track to deliver its targeted 500 megawatts of computing capacity in 2027.
Google declined to comment on the development, while Blackstone referred questions to the joint venture’s statement.
Crusoe, which is known for its involvement in OpenAI and Oracle’s flagship Stargate facility in Texas, did not respond to requests for comment.
AI infrastructure faces mounting constraints
The difficulties confronting Project Braid reflect a much broader challenge facing the technology industry: building physical infrastructure fast enough to satisfy the extraordinary demand generated by artificial intelligence.
Data centre developers are increasingly confronting shortages of electricity, transformers, switchgear, skilled labour and suitable sites. At the same time, communities and policymakers are scrutinising projects more closely because of concerns about power consumption, land use, water requirements and the effect of large facilities on local electricity prices.
Political opposition has also intensified, turning data centres into an increasingly prominent campaign issue ahead of the 2026 US midterm elections.
A May report by JPMorgan Chase estimated that more than 60 per cent of the data centre capacity planned for completion in 2027 had yet to begin construction, underscoring the potential for further delays across the sector.
People familiar with Project Braid said executives involved in the venture now estimate that data centre projects have roughly a 50 per cent chance of meeting their scheduled delivery dates. That compares with an estimated 90 per cent probability three years ago.
The changing outlook could complicate Blackstone’s ambition to establish itself as the world’s largest financier of digital infrastructure, particularly as demand for AI-related facilities continues to surge.
Google seeks wider market for AI chips
For Google, the success of Project Braid carries strategic importance beyond the development of data centres.
The technology giant is seeking to make the processors used to power its Gemini artificial intelligence models available to a broader range of customers while building an ecosystem around its own hardware that can compete more effectively with Nvidia, the dominant supplier of AI chips.
Many of the major neocloud companies, including CoreWeave, rely heavily on Nvidia’s processors.
Google therefore needs to expand its data centre footprint, both through its Google Cloud business and through partners capable of hosting its tensor processing units (TPUs), the specialised chips developed by the company for AI workloads.
Alphabet is forecasting capital expenditure of as much as $205 billion in 2026 as it accelerates spending on artificial intelligence.
Its partnership with Blackstone allows Google to pursue the expansion while limiting the amount of infrastructure spending and associated risks carried directly on its balance sheet.
Under the arrangement, some of the financial burden and development risk associated with building AI infrastructure can be shifted to the joint venture and its financial backers.
Search for new sites intensifies
With some proposed locations encountering delays, Project Braid’s backers have intensified efforts to identify alternative sites.
In recent weeks, Google has approached executives at neocloud companies with which it already has relationships, seeking assistance in locating potential data centre sites, according to one person familiar with the matter.
Blackstone, meanwhile, has been using consultants and its wider network of business partners to identify additional locations, other people familiar with the effort said.
The search will provide an early test for BXN1, Blackstone’s newly created division focused on artificial intelligence investments. The unit emerged from a business previously focused on growth-equity investments.
Blackstone is also diversifying its exposure to the rapidly expanding AI infrastructure market. The alternative asset manager recently joined other investment firms in pledging to work with Nvidia on billions of dollars in AI infrastructure spending.
Despite the setbacks, Project Braid has already identified 29 potential data centre locations, according to one person familiar with the venture.
The joint venture said the search was expanding in response to strong demand for its planned computing capacity.
“Given higher-than-expected demand for our compute capacity, we are actively expanding our pipeline to evaluate a broader pool of sites to support our long-term scaling plans,” the emailed statement said.
For Google and Blackstone, however, the challenge now is not simply finding customers for AI computing power, but securing the land, electricity, equipment, permits and construction capacity needed to deliver that computing power on time.