Google-Blackstone TPU Venture Brands as Crux AI, Appoints Meta's Data Center Chief as CDO
The Google-Blackstone joint venture that has been operating under the internal codename “Project Braid” since its May 2026 disclosure now has a public name: Crux AI. The naming on September 10 was accompanied by a significant leadership hire: Alan Duong, who spent 12 years at Meta as vice president and head of data center engineering and construction, joins as chief development officer.
The announcement landed two days after Bloomberg reported that Crux AI was already facing delays at planned data center sites — making this a branding moment that is arriving with questions attached.
What Crux AI Is
Crux AI is a separately incorporated joint venture designed to offer Google’s Tensor Processing Units as a rentable cloud service outside of the standard Google Cloud platform. Blackstone contributes equity capital from its funds; Google supplies the TPU hardware plus supporting software and services. The intended customer base is enterprise and AI companies that want access to TPU compute without going through the Google Cloud console.
The structure makes Crux AI a neocloud — specifically a TPU-focused neocloud competing in a market dominated by Nvidia GPU capacity operators. Blackstone has committed $5 billion in equity capital. The first 500 megawatts of capacity are targeted for 2027.
Google has offered Cloud TPUs for years as a product line inside its broader cloud business. Crux AI is a purpose-built company designed to scale TPU capacity aggressively and market it the way GPU clouds market Nvidia capacity — as infrastructure, not a cloud upsell.
Alan Duong’s Role
Duong’s 12-year tenure at Meta included leading the redesign of Meta’s data center program for the AI era, which required scrapping and reworking facilities that had been planned before AI workloads redefined power and cooling requirements. That work gave Duong direct experience managing the gap between what hyperscaler-era data center design assumed and what AI-era workloads actually demand.
At Crux AI, as CDO, he is responsible for the physical delivery of the capacity — site selection, construction, power procurement, and the supply chain that gets TPU racks into production. Given the Bloomberg report on site delays, this hire addresses precisely the capability gap the venture is under pressure to close.
Why the Timing Matters
Several dynamics converge here. Google’s Q2 2026 earnings showed inventory rising approximately 4x to roughly $10 billion as TPU systems moved into external customer facilities. External TPU revenue — Crux AI and the Anthropic/Apollo compute deals — is now a real line item. Crux AI’s formation accelerates Google’s ability to sell TPU access at scale without cannibalizing Google Cloud’s margins or sales motion.
At the same time, the AI infrastructure market has moved toward dedicated neoclouds as the most capital-efficient way to deploy large clusters. FluidStack, which operates as an Anthropic TPU operator, demonstrated that model. Crux AI with Blackstone capital and Duong leading delivery is a significantly more institutional version of the same structure.
The delay reports are the risk. Building 500 MW of AI-grade data center capacity by 2027 requires permits, power interconnects, and construction timelines that are already stretched across the industry. A marquee CDO hire is the right response to a credibility problem on physical delivery — but the actual 2027 deadline will be the real test.