Project Braid Encounters the Physical Limits of AI Growth

Alphabet and Blackstone’s ambitious cloud-computing venture is confronting a problem increasingly affecting the broader artificial-intelligence industry: securing the physical infrastructure required to turn massive AI investment into usable computing capacity. The initiative, internally known as Project Braid, was launched with $5 billion from Blackstone and aims to rent Google’s AI processors to customers beginning in 2027.

The delays highlight a widening gap between demand for AI computing and the ability to construct data centers quickly enough to accommodate it. In Wyoming, Google reportedly abandoned a planned development arrangement with Crusoe, taking control of the project and reapplying for permits. Local officials have indicated that the facility will be smaller than originally planned. Another prospective site reportedly lacked the appropriate transformers, an increasingly important constraint as developers compete for specialized electrical equipment.

Infrastructure Shortages Are Becoming a Strategic Risk

The transformer issue illustrates why AI infrastructure is no longer simply a question of securing chips and capital. Data-center construction depends on electricity connections, specialized equipment, skilled labor, permitting and local acceptance. According to a McKinsey report cited in the source material, transformer wait times can approach a year, adding significant uncertainty to development schedules.

Texas has introduced another obstacle. Governor Greg Abbott ordered a freeze on new data-center projects while the state examines how developers and technology companies pay for electricity. The move reflects growing concern that rapidly expanding data centers could place pressure on power systems and potentially shift infrastructure costs toward other electricity users.

These developments are forcing investors to reconsider assumptions about the speed of AI infrastructure deployment. JPMorgan estimated in May that more than 60% of data-center capacity targeted for completion in 2027 had not yet entered construction.

Delivery Expectations Are Being Reset

The changing environment is reflected in Project Braid’s internal planning. Executives involved with the venture now reportedly assume that individual data-center projects have roughly a 50% probability of meeting delivery dates, compared with approximately 90% three years ago. That dramatic shift illustrates how execution risk has become a central variable in the AI investment equation.

Project Braid CEO Benjamin Treynor Sloss said the venture remains on track to deliver its targeted 500 megawatts next year, emphasizing that different sites can progress at different speeds and that the company maintains a diversified pipeline. The venture has identified 29 potential locations and is actively expanding its options as demand for computing capacity exceeds earlier expectations.

Google and Blackstone Have More at Stake Than One Venture

The partnership has strategic implications for both companies. Google wants to expand access to its tensor processing units, which power its Gemini AI models, while reducing the direct infrastructure burden associated with its enormous AI capital spending. The company projects as much as $205 billion in capital expenditures this year, making partnerships that distribute costs and infrastructure risks increasingly important.

For Blackstone, Project Braid represents a test of its ambition to become the world’s largest financier of digital infrastructure. Its success could strengthen the case for alternative capital in AI infrastructure, but continued delays could demonstrate that financing alone cannot eliminate supply-chain, permitting and power constraints.

What Investors Should Watch Next

The critical question is whether Project Braid can convert its extensive site pipeline into operational capacity without allowing delays to materially increase costs. Demand for AI computing remains strong, but the economics of the sector increasingly depend on access to electricity, equipment and suitable locations.

For investors in the U.S. and Israel, the broader lesson is that AI infrastructure is becoming an execution story as much as a technology story. Google and Blackstone’s ability to deliver the targeted 500 megawatts in 2027, while navigating regulatory and supply constraints, could provide an important signal about how quickly the next phase of the AI buildout can realistically proceed.

Highlights