There is a quiet inefficiency sitting at the heart of every AI operation, from the world's biggest labs to a mid-sized company using cloud tools. The powerful chips that run AI, which can cost several dollars per hour to rent, spend a surprising amount of time doing nothing. They are waiting for data to be loaded from storage. Studies from Google and Microsoft show this idle time can consume up to 70% of a training job. For a company running a large AI workload, that is not a rounding error. That is the majority of the bill producing no output.
This is why all three major cloud providers, Google, Amazon, and Microsoft, are now competing aggressively on storage speed. Google just launched Cloud Storage Rapid. Amazon has had a comparable product called S3 Express One Zone since late 2023. Azure offers its own premium-speed tier. The fact that all three are investing heavily here signals that the problem is big and the money at stake is significant.
What Google is specifically claiming with its new product is notable. It says AI teams using the new storage saw 50% less idle chip time and checkpoints, which are the regular saves made during a long training run so work is not lost if something crashes, restored five times faster. That last number matters more than it sounds. When a training job crashes without a fast restore, you lose hours of compute time and pay again to redo it. Faster saves and restores are essentially insurance against very expensive accidents.
For businesses that are not running their own AI training, this still matters. If you are a company that uses AI services built by others, the cost and quality of that underlying infrastructure gets passed on to you in pricing and in response speeds. A provider running inefficient infrastructure is either losing money or charging you more. Neither is sustainable.
The deeper issue is one of lock-in. When a company moves its AI workloads and data into Google's high-speed storage system, or Amazon's, or Microsoft's, the practical cost of moving it somewhere else grows considerably over time. Data egress fees alone, which are what cloud providers charge to move your data out, can exceed the original storage cost at large enough volumes. Add to that the engineering time needed to reconfigure pipelines, and you understand why businesses that choose a cloud provider for AI today are likely to be with that provider for five or more years.
This is the strategic calculation Google is making. It is not just selling faster storage. It is building the infrastructure that makes teams want to consolidate everything on Google's cloud. Anthropic, one of the leading AI companies, is already named as a customer, and Anthropic has a deep partnership with Google as an investor. The pattern of an AI lab running on the infrastructure of its cloud backer is not accidental.
For non-technical business leaders evaluating cloud strategies, the practical question is not which storage product has the most impressive specification. The question is: what does choosing a cloud provider today commit you to over the next decade? The performance race being run by Google, Amazon, and Microsoft is real and it will improve AI results. But each improvement also deepens the dependency. The companies that think about this now, while their AI footprint is still small, will have far more negotiating power than those who only notice the lock-in after the data is already there and moving it would cost a fortune.