Google and Amazon's self-developed chip businesses alone are already quite large, especially Amazon's chip business, which accounts for one-fifth of Nvidia's annual revenue by volume. 1. Google TPU shipped 2.5 million units in 25 years, mainly for Google Cloud's own use. But if we look at it separately, it is still a business volume with an annual output value of one or two billion US dollars. Half of Google Cloud's $460 billion unfulfilled orders in the first quarter are estimated to be reserved for TPU computing power. So the TPU industry chain is worth paying attention to, and Fiona's analysis of the TPU industry chain is very comprehensive. 2. And Amazon's self-developed chip business has a larger scale Amazon's self-developed chip business (including Graviton CPUs, Trainium/Inferentia AI chips, Nitro, etc.) currently has an annualized revenue run rate of over 20 billion yuan, with a quarter on quarter growth of nearly 40% and a triple digit year-on-year growth. If the internal usage is also included, the annual revenue scale reaches 50 billion: Amazon CEO Jassy explicitly stated that if this business is treated as an independent company and the chips produced this year are sold to AWS itself and other third-party customers according to chip company practices, the annualized run rate will be close to $50 billion. A volume of $50 billion is already one-fifth of Nvidia's annual revenue for the past four quarters. Did you not expect it At present, the internal use of Amazon chips (consumed by AWS itself) accounts for a relatively large proportion, so the revenue reflected in external sales has been "masked". 1) Trainium's committed order amount reaches 225 billion yuan: Trainium (mainly Trainium 2/3/4) has over 225 billion yuan in revenue commitments from major AI labs and enterprise clients such as OpenAI, Anthropic (a large-scale multi-year contract at the Dogawa level), Uber, etc. 2) Trainium2 is almost sold out, Trainium3 is also close to full order, and Trainium4 has a large number of reservations. This year's first quarter financial report is Amazon's first systematic disclosure of the scale of its self-developed chip business, highlighting its status as one of the top three players in the global data center chip field (tied with Nvidia and others). The overall annualized run rate of AWS has reached around 150 billion, and the chip business is an important growth point, especially in the AI training/inference field, helping to reduce reliance on external GPUs.