Google is actively exploring strategies to increase the adoption of its AI chips, known as Tensor Processing Units (TPUs), aiming to establish a broader AI ecosystem that can effectively compete with market leader Nvidia. The company is using its financial strength to support data-center partners, with discussions ongoing for a potential investment of approximately $100 million in cloud-computing startup Fluidstack, which would value the company at around $7.5 billion. Fluidstack is one of several "neocloud" companies that provide computing services for AI businesses, similar to CoreWeave, which predominantly uses Nvidia GPUs.
This initiative includes broadening financial commitments to other data-center partners to stimulate further demand for TPUs. Google has previously backstopped financing for projects involving Hut 8, Cipher Mining, and TeraWulf. Google began selling access to TPUs through its cloud services in 2018 and is now also directly selling TPU chips to external customers, according to industry research group SemiAnalysis.
Google's TPUs are gaining recognition for their effectiveness in training AI models and performing inference tasks. The company recently unveiled its eighth-generation TPUs, splitting them into two specialized chips: the 8t for training and the 8i for inference. The latest Gemini 3 Pro model was entirely trained on TPUs, demonstrating their capability for frontier models. Morgan Stanley projects that selling 500,000 TPU chips could generate approximately $13 billion in revenue for Google by 2027.
