{"slug": "google-tpuv7-posts-strong-agentic-inference-results-against-nvidia-gb300", "title": "Google TPUv7 posts strong agentic inference results against Nvidia GB300", "summary": "Google's TPUv7 (Ironwood) delivered 1.42 to 1.67 times better performance per dollar than Nvidia's GB300 NVL72 on the AgentX agentic inference benchmark, according to SemiAnalysis. The TPUv7 submissions ran the 397-billion-parameter Qwen3.5 model in FP8 on only four chips, while Nvidia's system ran in FP4 across full 72-GPU racks. SemiAnalysis also reported Anthropic has committed to more than one million TPU units as the largest external customer, with Oracle planning to resell TPUs through Oracle Cloud Infrastructure starting in 2027 and OpenAI reportedly preparing direct TPU purchases.", "body_md": "# Google TPUv7 posts strong agentic inference results against Nvidia GB300\n\nGoogle 2015 logo (Wikimedia Commons, public domain)\n\nSemiAnalysis data shows Ironwood beating Nvidia's flagship rack on performance per dollar while external customers keep signing up\n\n[Google](https://cryptobriefing.com/markets/alphabet/)’s TPUv7, the chip also known as Ironwood, has delivered early benchmark results that put it ahead of [Nvidia](https://cryptobriefing.com/markets/nvidia/)’s GB300 NVL72 on cost efficiency for agentic AI workloads, according to SemiAnalysis.\n\nOn the AgentX benchmark, TPUv7 submissions posted performance per dollar between 1.42 and 1.67 times better than Nvidia’s rack-scale system.\n\nThe timing matters because the AI industry’s spending focus is shifting. Training giant models grabbed the headlines for years. Running those models for millions of users, and increasingly for autonomous agents, is where the recurring bills now pile up.\n\n## What the benchmark actually measured\n\nAgentX tests inference for agentic workloads. Inference is the stage where a trained model answers questions or carries out tasks, and agentic workloads involve AI systems working through multi-step jobs instead of single replies.\n\nThose jobs tend to involve long context and multi-turn exchanges, meaning the model has to remember a lot and talk back and forth repeatedly. SemiAnalysis’s research notes that TPU architecture now supports exactly that kind of long-context, multi-turn inference.\n\nThe test model was Qwen3.5 with 397 billion parameters, an external model and not one built by Google.\n\nResults were checked across several interactivity levels. Interactivity, roughly speaking, describes how fast a single user gets responses versus how many users a system can serve at once.\n\nThen there is the precision question. Google’s submissions ran in FP8, while Nvidia’s ran in FP4. These are number formats that determine how many bits a chip uses to represent each value in the model.\n\n### AI, tech, and the markets they move—in one daily briefing.\n\nDaily. Free. Join 34,000+ readers across crypto, finance, and policy.\n\nFewer bits generally means less memory and faster math, which is why FP4 is described as the more aggressive option. Google, in other words, still came out ahead on cost efficiency while playing with the less aggressive format.\n\nThe hardware footprint was also lopsided. SemiAnalysis says the TPUv7 benchmarks used only four chips, compared with full Nvidia racks of 72 GPUs.\n\n## From internal tool to external product\n\nGoogle has been building TPUs since 2013, and for most of that history they were mostly a house specialty powering Google’s own services.\n\nIn November 2025, SemiAnalysis reported that Google had begun selling TPUs to outside customers. The firm now says that call has held up, naming [Oracle](https://cryptobriefing.com/markets/oracle/), [OpenAI](https://cryptobriefing.com/markets/openai/), [Anthropic](https://cryptobriefing.com/markets/anthropic/) and [Meta](https://cryptobriefing.com/markets/meta/) as TPU customers in line with its Accelerator Model tracking.\n\nThe research outlines how some of those relationships look in practice:\n\n- **Anthropic** has committed to more than one million TPU units, making it the largest external customer.\n- **Oracle** plans to resell TPUs through Oracle Cloud Infrastructure starting in 2027 under a multi-gigawatt-scale contract. SemiAnalysis describes this as a first for a hyperscaler outside Google.\n- **OpenAI** is reportedly preparing to commit to direct TPU purchases.\n\n## The software gap is narrowing\n\nSemiAnalysis argues Google is chipping away at that advantage too. Google has partnered with upstream maintainers across the open-source AI stack, including the Meta PyTorch team, vLLM (via Inferact and RedHat), SGLang (via RadixArk), llm-d and Ray.\n\nThose names matter because they are the frameworks and serving engines developers already reach for. If TPU support lives inside the tools people use daily, switching hardware starts to look like a configuration change instead of a rewrite.\n\n## What this means for Nvidia, Google and AI buyers\n\nThe most direct stakeholder here is Nvidia. Performance per dollar is the metric that large buyers obsess over once inference bills scale up. SemiAnalysis’s research suggests results like these may put upward pressure on Nvidia to compete harder on cost efficiency and performance.\n\nThe customer roster tells its own story. When Anthropic, Oracle, Meta and reportedly OpenAI are all routing at least part of their compute toward TPUs, the market is no longer a one-vendor conversation.\n\nThe items worth tracking from here are concrete. Watch whether OpenAI’s reported TPU commitment gets formalized, how Oracle’s 2027 rollout on OCI proceeds, and whether independent benchmarks across other models confirm the AgentX results.\n\n**Disclosure:** This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).", "url": "https://wpnews.pro/news/google-tpuv7-posts-strong-agentic-inference-results-against-nvidia-gb300", "canonical_source": "https://cryptobriefing.com/google-tpuv7-agentic-inference-benchmarks-nvidia/", "published_at": "2026-11-05 19:19:21+00:00", "updated_at": "2026-11-05 19:22:40.099217+00:00", "lang": "en", "topics": ["ai-chips", "ai-infrastructure", "large-language-models", "ai-agents"], "entities": ["Google", "TPUv7", "Ironwood", "Nvidia", "GB300 NVL72", "SemiAnalysis", "Qwen3.5", "Anthropic"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/google-tpuv7-posts-strong-agentic-inference-results-against-nvidia-gb300", "markdown": "https://wpnews.pro/news/google-tpuv7-posts-strong-agentic-inference-results-against-nvidia-gb300.md", "text": "https://wpnews.pro/news/google-tpuv7-posts-strong-agentic-inference-results-against-nvidia-gb300.txt", "jsonld": "https://wpnews.pro/news/google-tpuv7-posts-strong-agentic-inference-results-against-nvidia-gb300.jsonld"}}