DeepSeek releases largest open-source AI models and confirms Chinese hardware alternative to Nvidia chips scaling to 950 data centers by late 2025

DeepSeek released two new AI models (V4-Pro with 1.6 trillion parameters and V4-Flash with 284 billion parameters) under MIT open-source license on Hugging Face, making them freely available to any developer or company. Both are already running on the DeepSeek API. Anthropic also admitted it silently reduced the reasoning quality of Claude Sonnet 4.6 and Opus 4.6, then reversed the changes after users complained — signaling that closed AI services can be modified without user notice.

DeepSeek's models are optimized to run on Huawei Ascend chips (an alternative to Nvidia's processors that were restricted by U.S. export controls), and the company confirmed 950 data centers powered by this Chinese-made hardware will go live in the second half of 2025. This is the first large-scale, publicly confirmed alternative compute stack outside Nvidia's supply chain. OpenAI released GPT-5.5 with measurably better efficiency — it costs 5–10x less per task on certain benchmarks than Claude Opus, despite higher per-token pricing, because it requires fewer tokens to solve the same problem.

Enterprise teams evaluating AI costs now have a credible open-source option (DeepSeek V4) that runs on affordable hardware and will get cheaper as Chinese infrastructure scales up — compressing profit margins for every cloud-hosted AI service (OpenAI, Anthropic, Google). Developers building AI agents can now run capable models (Qwen 3.6 with 27 billion parameters) locally on consumer hardware instead of paying for cloud inference, eliminating API costs for agent-building projects. The trust damage from Anthropic's silent model downgrade gives competitors ammunition in sales conversations with enterprises that rely on coding tools daily.

DeepSeek V4 models are immediately available MIT-licensed (free to use and modify) on Hugging Face. OpenAI's GPT-5.5 is available in ChatGPT now but API access for developers requires a pending safety review — exact timing unknown. Watch for Anthropic's response: if the company commits to public notification before any future model changes, it could stabilize enterprise trust; silence will accelerate migration to open-source alternatives.