Google is actively exploring new strategies to expand the market for its AI chips, known as Tensor Processing Units (TPUs), and challenge Nvidia's dominance in the AI chip sector. The company is using its financial might to build a broader AI ecosystem, acknowledging challenges like manufacturing bottlenecks and limited interest from traditional cloud computing rivals.
To achieve this, Google is increasing financial support to a network of data-center partners that can offer computing power to a wider array of customers. For instance, Google is in discussions to invest approximately $100 million in the cloud-computing startup Fluidstack, with a deal valuing the company at around $7.5 billion. Fluidstack is one of several "neocloud" companies providing computing services to AI firms and aims to amplify its growth by encouraging more providers to use Google's TPUs. Google has also reportedly discussed expanding financial commitments to other data-center partners that could drive additional TPU demand.
Google's new strategy includes directly selling TPU chips to external customers, moving beyond its traditional approach of offering access through its cloud services. This effort reflects a push to expand the potential market for its chips, which have been praised by AI customers, including Anthropic, for their effectiveness in training some models and for inference tasks. Google has also unveiled its eighth-generation TPUs, splitting them into two specialized chips: the 8t for training and the 8i for inference, to optimize for different AI workloads. The company's Gemini 3 Pro model was entirely trained on TPUs, showcasing their capability for frontier models.
