Recently,Google with a renowned artificial intelligence company Anthropic Formally announced a new phase of strategic cooperation in cloud infrastructure. According to official statements released by both parties and multiple media reports, the value of this agreement is as high as tens of billions of dollars, and Anthropic will significantly expand in the coming years in Google Cloud the scale of usage is expected to involve over 1 million units developed by Google. TPU(Tensor Processing Unit, TPU) to train and deploy future AI models. It is expected that by 2026, the total computing power Google Cloud provides to Anthropic will surpass the 1 gigawatt (GW) level, which also overturns the established impression that NVIDIA is the only choice for traditional large-scale AI computing.
Google and Anthropic will use a million TPUs to build ultra-gigawatt-scale AI computing power.
A gigawatt (GW) is a unit of power equal to one billion watts (1,000,000,000 watts), commonly used to describe the capacity of large power generation facilities such as nuclear power plants, wind farms, and solar power systems. For example, a 1-gigawatt power plant could ideally supply electricity to approximately 876,000 households for a year, which gives a sense of the scale involved.
According to the latest agreement between Anthropic and Google, this represents not just an expansion of cloud rental capacity, but a strategic reshaping of the AI computing foundation. Google stated that the partnership will cover the deployment of the latest-generation TPU v7 “Ironwood” chip, which is known for its high efficiency and cost-effectiveness in AI model training, even OpenAI is also using。
Google Cloud CEO Thomas Kurian noted: “Anthropic’s decision to significantly expand its use of TPUs is based on the outstanding cost-performance and energy efficiency we’ve experienced in real-world applications over the years. We continue to deliver through our mature portfolio of AI accelerators, including the latest seventh-generation TPU.”Ironwood』, pushing the limits of computing efficiency.」
This collaboration enables Anthropic to scale its computing capacity at remarkable speed, keeping pace with the surging global demand for its flagship model, Claude. According to the company’s latest data, Anthropic now serves over 300,000 enterprise customers, and the number of high-end clients with annual recurring revenue exceeding $100,000 (approximately NT$3 million) has grown nearly sevenfold over the past year.
Anthropic Chief Financial Officer Krishna Rao said in an official statement that this collaboration is a natural extension of the two sides’ multi-year partnership and a necessary investment for Anthropic to continue expanding its AI capabilities.
He emphasized: “Anthropic’s partnership with Google has spanned many years, and this latest expansion will enable us to continuously enhance the computing power needed to define the frontier of AI. From Fortune 500 companies to AI-native startups, customers rely on Claude for their most critical tasks. This collaboration ensures that we can maintain our models’ leading position even as demand surges rapidly.”
Three Major Chip Platform Strategy: Anthropic’s Diversified Computing Layout
Unlike a single-supplier dependency strategy, Anthropic has taken a more flexible multi-platform approach. The company currently utilizes Google’s TPU, Amazon’s Trainium, and NVIDIA’s GPU simultaneously, forming a diversified high-performance computing architecture.
Anthropic says this strategy not only ensures it maintains negotiating and technical flexibility across different cloud providers, but also allows the company to select the most cost-effective and performance-efficient chip platform for specific workload needs. For example, during large-scale training phases, it may lean on the flexibility and ecosystem advantages of GPUs, while for inference or multimodal applications, it can combine the high concurrency and energy efficiency of TPUs.
For Google, this multi-billion-dollar partnership not only strengthens its position in the cloud infrastructure market, but also cements its role as a key pillar of the AI ecosystem. Over the past few years, Google Cloud has gradually evolved from a distant third behind AWS and Microsoft Azure into a key player in the generative AI era, with its dedicated TPU architecture and integrated platform now becoming the solutions that major AI companies are competing to adopt.
The rapid development of the AI industry has made the demand for computing resources the core of a new round of competition in technology infrastructure. From the GPU alliance between Microsoft and OpenAI to cloud partnerships involving Google, Amazon, and Anthropic, nearly all of the world’s top five cloud providers are making unprecedented capital expenditures on AI infrastructure. According to market forecasts, within the next three years, the electricity demand from generative AI data centers alone could account for more than 3% of global total electricity consumption.
Google emphasized that the latest generation of TPUs has significantly improved energy efficiency in its design, and it continues to power its data centers with renewable energy to mitigate the environmental impact of AI computing. Anthropic also noted that part of the reason for choosing Google is its technological leadership in sustainable computing. In the next phase of AI, as analysts put it: “In the era of generative AI, whoever has energy and computing power holds the future.”
Source: KOCPC Chinese