DeepSeek Open-Sources Ascend 950 Components as AI Dynamics Turn Toward Domestic Computing Power

Overview

Recently, the artificial intelligence field has seen a number of significant developments, spanning open-source domestic computing power, government-level AI service deployments, and progress on cutting-edge international models. DeepSeek has officially open-sourced infrastructure components for the Ascend 950, aiming to accelerate ecosystem adaptation for domestic AI computing. Meanwhile, Anthropic has launched Claude for Government, further expanding its application reach in the public sector. Additionally, reports on the capabilities assessment of GLM-5.3 and account interoperability between Cognition and ChatGPT have drawn industry attention.

DeepSeek Open-Sources Ascend 950 Components to Boost Domestic Computing Power Ecosystem

On September 30, the domestic AI company DeepSeek announced the open-sourcing of infrastructure components targeting Huawei's Ascend 950 chip. This move aims to lower the barrier for developers deploying large models on domestic hardware and to improve training and inference efficiency. Although the announcement was made at the end of the month, widespread discussion was sparked after Hong Kong media reposted the news on October 1.

This move is seen as an important step for DeepSeek in deepening its adaptation to domestic computing power. By open-sourcing the underlying support libraries, the company hopes to attract more developers to optimize and innovate models based on the Ascend platform, thereby alleviating reliance on high-end overseas GPUs. Analysts point out that this reflects the ongoing exploration and pragmatic cooperation of China's AI industry on the path toward computing power self-sufficiency.

International Giants: Government Services and Model Capability Progress

On September 30, the American AI company Anthropic officially opened its Claude for Government service, designed specifically for government agencies, with an emphasis on data security, compliance, and transparency. The service offers customized AI solutions to public sector clients, marking a new stage in the application of mainstream large language models in government affairs.

Meanwhile, a report released around September 29 indicated that Zhipu AI's GLM-5.3 model performed close to Anthropic's yet-to-be-revealed Mythos model in certain benchmark tests. Although this information was reported again on October 1, it reflects the rapid progress of domestic large models in performance competition.

Another noteworthy development is Cognition's announcement on September 29 that users can log in to its AI programming assistant, Devin, using their ChatGPT accounts. This integration aims to simplify the user experience and promote interoperability across platform toolchains, further consolidating Devin's position in the field of automated code development.