Wall Street is recalibrating its view of Google's in-house TPU (tensor processing unit) franchise as a material competitive lever against NVIDIA's dominance in AI infrastructure. Google Cloud revenue is projected to surge roughly 64% this year to $96 billion, with analyst growth exceeding 50% expected in 2027, driven heavily by TPU availability and adoption by major customers including Anthropic. Google has also launched a new AI compute venture with Blackstone, signaling confidence in the TPU's value proposition.
TPUs belong to a class of application-specific integrated circuits (ASICs) optimized specifically for machine learning training and inference, not general computing. Unlike NVIDIA GPUs—which offer flexibility and benefit from years of developer investment in CUDA—ASICs like TPUs consume 20–40% less energy than NVIDIA processors and enable Google to charge 20–30% less for excess compute capacity. This cost advantage has proven compelling enough to attract AI unicorns, including Anthropic, to lease or purchase access through Google Cloud and enterprise services.
Brad Gastwirth, global head of market research at Circular Technology, noted Google is "probably the most underappreciated competitor of NVIDIA." For architects evaluating multi-cloud or on-premise options: TPU's specialization and energy footprint—combined with Google Cloud's software and AI services—may deliver better total-cost-of-ownership for training and inference workloads, especially at scale.