Two major DeepSeek developments dropped within hours of each other today, one on the open-source hardware side and one on the US regulatory side. Since a lot of headlines are glossing over the technical and legal details, here is what is actually in the GitHub repo and the Federal Register filing. First, DeepSeek released its internal production inference framework, called DeepInfer, as open-source under an Apache 2.0 license. This is the engine they use to run their V4.1-Flash model at scale. Most AI labs keep their production inference code proprietary because optimizing compute overhead is how they protect their margins. DeepSeek published theirs publicly, stating the goal is to lower compute costs and energy usage for the community. Looking at the GitHub documentation, the big hardware detail is that DeepInfer is built to run natively on both Nvidia SM100 GPUs (using CUDA 13) and Huawei Ascend 950 NPUs (using CANN 9.2), utilizing custom TileLang kernels and speculative decoding. It shows Huawei's 950 chips are now officially supporting frontier production inference alongside Nvidia. However, the repo also notes a key hardware limitation: batch-invariant inference and deterministic sampling currently only work on Nvidia GPUs, not on Huawei. Second, almost simultaneously, the Department of Homeland Security (DHS) published an official exclusion and removal order in the Federal Register (Docket DHS-2026-0628). It orders all US federal civilian agencies to stop buying DeepSeek products and remove them from their IT systems. The key detail in the filing is that DHS issued this under the Federal Acquisition Supply Chain Security Act. Because it is treated as a supply chain security rule rather than a simple workplace app ban, the restriction also covers third-party software and services built using DeepSeek models. Agencies have 60 days to request written waivers, which are rarely granted. The order cites national security risks regarding Chinese cybersecurity laws and DeepSeek's privacy policy (updated in February 2026), which states user prompts, files, and device data are stored on servers in China. This also follows last month's US accusations that several Chinese labs were using large-scale model distillation on American AI systems. Even if you aren't trading these directly, the market context is pretty clear. We are watching the global AI supply chain split in two in real time. On the tech side, open-source inference is getting much cheaper and running natively on domestic Chinese silicon alongside Nvidia. On the regulatory side, the US government just legally walled off federal procurement from any software touching Chinese open weights, effectively locking public sector contracts and government contractors into cleared US cloud and enterprise providers (like Microsoft, Amazon, and Palantir). Here are the links to the primary documents and coverage if you want to read them directly: DHS Federal Register Order (Doc 2026-21447): https://www.federalregister.gov/documents/2026/10/20/2026-21447/notice-of-order-regarding-exclusion-and-removal-of-deepseek-products-from-federal-civilian-executive DeepInfer GitHub Repo: https://github.com/deepseek-ai/DeepInfer Quartz / Bloomberg report on DeepInfer: https://qz.com/deepseek-open-source-deepinfer-inference-software-ai-102026 MarkTechPost technical breakdown: https://www.marktechpost.com/2026/10/20/deepseek-deepinfer-open-source-inference-framework/ TechRepublic on the DHS ban: https://www.techrepublic.com/article/news-dhs-deepseek-federal-systems-security-ban/
CryptoAlerta — análise de criptomoedas e mercado em tempo real