Google maps an anti‑Nvidia playbook

Google used its Cloud Next event on April 22 to do two things at once: launch new artificial-intelligence chips and put $750 million behind the partners that sell AI projects to big companies. (blog.google, googlecloudpresscorner.com) The chip side starts with Google’s eighth-generation Tensor Processing Units, or TPUs, which are custom processors for AI work inside Google Cloud. Google said the new family splits into TPU 8t for training models and TPU 8i for running them in production, and both are “coming soon.” (blog.google, cloud.google.com) That is a change from April 2025, when Google introduced Ironwood as its seventh-generation TPU and called it the first TPU designed specifically for inference, the step where a trained model answers prompts. Google’s current documentation says TPU7x is the first release in the Ironwood family and is available on Google Cloud now. (blog.google, docs.cloud.google.com) Google also said it will move to a yearly TPU cadence. The company’s technical write-up says the split reflects a market where pre-training, post-training, and real-time serving now need different kinds of hardware. (thenextweb.com, cloud.google.com) The sales side is aimed at a different bottleneck: getting AI into real corporate workflows. Google Cloud said its new $750 million fund will support consulting firms, systems integrators, software partners, and channel partners across its 120,000-member ecosystem. (googlecloudpresscorner.com, cloud.google.com) Google named Accenture, Deloitte, KPMG, PwC and NTT DATA among the firms expanding work around its agentic-AI tools. Google said Accenture has already built more than 450 AI agents on Google Cloud, while KPMG committed $100 million and PwC committed $400 million to related efforts. (cloud.google.com, thenextweb.com) This comes as Google is still the No. 3 cloud provider by market share. Synergy Research data cited by CRN put Amazon Web Services at 29%, Microsoft at 22%, and Google at 12% in the fourth quarter of 2025. (crn.com) Google is not dropping Nvidia from its cloud. At Cloud Next 2026, the company also said Google Cloud will offer Nvidia Vera Rubin graphics processing units, showing that its TPU push sits alongside outside chips rather than replacing them outright. (blog.google, cloud.google.com) The thread running through both announcements is that Google is trying to control more of the AI stack inside its own cloud: chips for model building, chips for model serving, and financing for the firms that turn demos into signed contracts. Cloud Next ran April 22 through April 24 in Las Vegas, and Google used the opening day to show that those pieces now move together. (cloud.google.com, googlecloudpresscorner.com)