AI compute buyers are giving AMD GPUs and Google TPUs a closer look as Nvidia’s leadership transition raises questions about its future investment strategy, according to The Information’s November 17, 2026, report.
The report describes increased customer interest at providers including Cirrascale, Sciforium and TensorWave, alongside Google’s push to make its accelerators available beyond Google Cloud. Some prospective TPU purchases could reach hundreds of millions of dollars.
Those are consequential signals, but they describe several different things: interest in testing hardware, existing deployments, intentions to move workloads and possible future orders. The original article is paywalled; detailed public quotations are available through a Stocktwits report republished by Yahoo Finance. These selected accounts provide a view into procurement decisions, not a market-wide measurement of Nvidia losing share.
The most useful question for buyers is where alternatives are becoming practical. An earlier Cirrascale interview offers a concrete answer: inference already has room for multiple accelerator suppliers, while software and system integration still constrain broader migration.
AMD Interest Builds on Existing Inference Deployments
TensorWave’s business gives it a direct interest in AMD adoption. Its chief operating officer, Piotr Tomasik, told The Information that customers had initially been difficult to convince, according to the Stocktwits summary. He attributed a subsequent improvement in customer perceptions to outside research showing AMD hardware matching or outperforming Nvidia on some benchmarks.
That is an executive’s explanation of changing customer attitudes, not a benchmark result readers can apply to their own infrastructure. The accessible coverage does not identify the tests, model configurations or operating conditions behind that comparison.
Cirrascale provides a more concrete deployment example. In an October 3, 2025, EE Times interview, CEO Dave Driggers described more than 1,000 AMD MI300 clusters operating at the company, with some MI350 and MI355 hardware also used for inference. Cirrascale also deployed Nvidia hardware at scale.
The date matters. That account predates the November 2026 leadership story by more than a year. AMD’s opportunity was already developing through workload-specific deployments; uncertainty at Nvidia may be adding urgency to decisions that customers could already justify technically.