Google TPU Cloud and Blackstone Investment to Deliver 500MW AI Capacity

Source Domain: techgenyz.com

Google and Blackstone have announced a serious partnership that is really about expanding artificial intelligence cloud infrastructure, in a straight-up, practical way. They’re launching a TPU cloud venture that is meant to offer large-scale AI computing capacity powered by Google’s custom Tensor Processing Units, also known as TPUs.

Google TPU cloud investment with Blackstone isn’t happening in a vacuum, because the demand for AI infrastructure keeps climbing. Companies everywhere are trying to develop and deploy increasingly strong artificial intelligence systems faster than before. Blackstone will initially invest $5 billion into the project. Meanwhile, the overall value of the infrastructure expansion could end up closer to $25 billion, depending on additional financing that comes later.

The Google TPU cloud venture will concentrate on AI-first cloud infrastructure, designed for advanced AI workloads, model training, and inference systems in production. At the same time, this partnership also points to Google pushing harder to commercialize its AI hardware ecosystem beyond its own internal uses. In other words, Google seems intent on going more aggressively after the global AI cloud market, where competition is not exactly light.

Image source: blog.google

What The Google TPU Cloud Venture Will Provide

This venture is set up so businesses can access Google’s TPU hardware through a cloud-based “compute-as-a-service” model. Instead of companies building expensive AI infrastructure on their own, they can rent big computing capacity directly from the platform.

The service will reportedly include TPUs for AI-compute powered workloads, and likely a setup that makes scaling less of a headache.

Currently, most AI cloud providers lean pretty hard on NVIDIA GPUs to power AI training and inference systems. Companies like CoreWeave and quite a few other cloud infrastructure startups have effectively built full businesses around renting out NVIDIA-powered computing…

Source