Alphabet and Blackstone's $5 billion joint venture to build data centers for Google's AI chips has encountered multiple setbacks across planned sites, underscoring how physical constraints are now slowing the rollout of AI infrastructure. The venture, internally called Project Braid, aims to deliver 500 megawatts of capacity in 2027 and has identified 29 potential locations. Yet executives at the venture now estimate data center projects have a 50% chance of meeting delivery dates, down from 90% three years ago. That sharp decline in confidence signals a broader shift: capital and demand are no longer the primary bottlenecks; transformers, switchgear, skilled labor, permits, and state policy are.
The material facts point to a convergence of supply chain and regulatory friction. Google pulled a project in Cheyenne, Wyoming from developer Crusoe and took over permitting after losing confidence in the firm's ability to deliver. Another site lacked necessary electrical transformers, equipment now facing wait times of nearly a year. Supply shortages for switchgear and skilled labor add to delays. Separately, Texas Governor Greg Abbott's freeze on new data center projects while the state studies electricity cost allocation has also disrupted plans. JPMorgan reported in May that more than 60% of data center capacity planned for 2027 completion has not yet broken ground, reinforcing the gap between announced capacity and physical execution.
The evidence comes from a single Propmodo report published on September 9, 2026, which was read in full. The source describes the venture's structure and setbacks but does not provide corroborating documents, earnings disclosures, or named executive quotes. Key figures include the $5 billion joint venture amount, Google's projected $205 billion capital expenditure, the 500 megawatt 2027 delivery target, the 29 potential locations, the 50% delivery probability estimate, and the JPMorgan finding that more than 60% of planned 2027 capacity has not yet broken ground. The report also notes that Google structured the venture with Blackstone to move some of its projected capital expenditure off its balance sheet while expanding access to its tensor processing units beyond its own cloud.
The sector implications are significant for data center operators, equipment suppliers, and hyperscalers. The venture joins a wave of neocloud entities that lease computing capacity, though most run Nvidia chips rather than Google's processors. Blackstone is testing its new BXN1 division for AI investments, which evolved from its growth equity business. If delivery timelines continue to slip, the economics of AI infrastructure could shift toward existing capacity owners and suppliers with secured equipment. The Texas freeze also highlights how state-level electricity cost allocation debates can become a new source of project risk, particularly in markets with rapid data center growth.
Several limitations temper the analysis. The dossier relies on one secondary source, and the material claims are not corroborated by primary filings or company statements. The report does not specify which sites are affected beyond Cheyenne, Wyoming and Texas, nor does it detail the financial impact of delays on the venture's returns. It is also unclear whether the 50% delivery probability estimate reflects all 29 locations or a subset. What to watch: whether Google and Blackstone disclose project-level timelines, whether transformer and switchgear lead times improve, and whether Texas completes its electricity cost allocation study without imposing permanent restrictions on new data center projects.