Cloud chips and AI tiers

Amazon is testing a new idea in cloud computing: selling its own server chips more directly, while OpenAI is slicing model access into more price tiers. (theregister.com) (help.openai.com) Amazon chief executive Andy Jassy said April 9 that two large Amazon Web Services customers asked to buy all of the company’s Graviton server capacity for 2026, and he said Amazon turned them down. Jassy also said Amazon’s in-house silicon business is running at more than $20 billion annually and would be about $50 billion if it sold this year’s chips to Amazon Web Services and outside buyers “as other leading chips companies do.” (theregister.com) (bloomberg.com) Graviton is Amazon’s general-purpose processor, the main computing chip in a server, and Amazon says Graviton-based instances deliver up to 40 percent better price performance than comparable x86 instances. Amazon’s custom silicon lineup also includes Trainium for training artificial intelligence models, Inferentia for running them, and Nitro chips that handle server networking and security tasks. (aws.amazon.com 1) (aws.amazon.com 2) OpenAI made a parallel pricing move in software. Its Help Center says a new $100 ChatGPT Pro plan offers 5 times the usage of Plus, while the existing $200 Pro plan remains the top tier at 20 times Plus usage. (help.openai.com) OpenAI also tied those plans more tightly to coding work. Its Codex rate card says that, as of April 2, 2026, Codex pricing for new and existing Plus, Pro, ChatGPT Business, and new ChatGPT Enterprise plans moved from per-message charges to token-based pricing aligned with the application programming interface. (help.openai.com) The model menu changed too. OpenAI says GPT-5.3 and GPT-5.4 are now the named models in ChatGPT, and that GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 Instant and Thinking were retired from ChatGPT on February 13, 2026, though application programming interface access was unchanged. (help.openai.com) Both companies are turning internal infrastructure into products with clearer price tags. Amazon built chips first to lower its own cloud costs and reduce dependence on outside suppliers, while OpenAI is packaging model choice, coding capacity, and usage ceilings into subscription bands that teams can buy directly. (aws.amazon.com) (help.openai.com) Amazon has spent years pushing customers onto Arm-based Graviton instances inside Amazon Web Services, and OpenAI has spent months reshaping ChatGPT around newer GPT-5.x models and Codex. The latest moves put the meter in plainer view: racks for compute on one side, rate cards and usage multiples on the other. (aws.amazon.com) (openai.com)