Google names new AI infrastructure chief

Google has named Amin Vahdat as its new chief technologist for AI infrastructure, according to an internal memo reported by Semafor.

Vahdat is a longtime executive at the company.

The appointment comes as Google increases investment in data centers and hardware to support AI workloads, with its capital expenditures expected to exceed US$90 billion by the end of 2025.

Google Cloud CEO Thomas Kurian said in the memo that AI infrastructure would be a key focus for the company.

Google designs its own tensor processing units (TPUs) for AI and is focusing on scaling its infrastructure as competition in the sector intensifies.

The company reported a US$155 billion cloud backlog, reflecting ongoing demand for cloud and AI services.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Google plans $90B capex, TPU edge unclear due to sparse public roadmap

Vahdat’s appointment signals a push to offer TPUs beyond internal use 1. Anthropic committed to up to 1 million TPUs, while prospects include Meta and xAI, with SSI plus OpenAI also evaluating 1. Sparse info on TPU v7 Ironwood timing and regions plus third‑party tests clouds comparisons with Nvidia’s Blackwell 1.
Google says TPU v7 nears Nvidia GB200 on peak FLOPs and offers about 44% lower total cost of ownership versus comparable GB200 systems 1. Inference tests lag, and XLA:TPU remains closed source 1. XLA is Google’s Accelerated Linear Algebra compiler backend for TPU workloads, which adds friction for teams outside Google as it rolls out “native” PyTorch support 1.

Vendors can land near‑term orders by following Google’s AI data center buildout and energy timelines

Suppliers in power gear or liquid cooling can win work by tracking new sites and bid windows. Google operates at least two Texas data centers and announced three more worth $40 billion 2 3. The Texas projects create thousands of jobs with spending through 2027, and one pairs with solar plus battery storage 2 3.
OEMs plus service providers that align with the Water Efficiency request for information (RFI) for 100+ million gallons per year by 2028 can win orders as Google scales TPU pods, which are pre‑wired clusters, to 9,216 chips per cluster 4 5 1.