
Chinese AI start-up DeepSeek has made headlines by crediting Tencent Holdings for a major technical breakthrough that significantly improved its DeepEP code library-a core component for inter-chip communications in AI model training. DeepSeek AI infrastructure engineer, Zhao Chenggang, publicly acknowledged that Tencent’s network platform department provided a solution that not only resolved a long-standing system bottleneck but also delivered a “huge speed boost,” enhancing both the robustness and communication performance of DeepEP.
DeepEP, which supports high-throughput, low-latency GPU operations and low-precision computing, is crucial for training large-scale AI models efficiently. Tencent’s Starlink Networking team identified and addressed two main technical issues: underutilized dual-port NIC bandwidth and CPU control latency. Their optimizations resulted in a 100% performance improvement on RoCE (RDMA over Converged Ethernet) networks and a 30% gain on InfiniBand, two leading technologies for high-speed data transfer in AI data centers. These enhancements have already been deployed in Tencent’s own Hunyuan AI model training and are now available as open-source for the broader community.
This collaboration is emblematic of a broader trend in China’s tech industry: a growing embrace of open-source AI development. Rather than competing solely through proprietary technologies, major players like Tencent, Alibaba, and Baidu are increasingly sharing their innovations, fostering a culture of transparency and collective advancement. DeepSeek’s DeepEP is just one of eight open-source projects the company unveiled during a week-long initiative in February, which also showcased the technical underpinnings of its cost-effective, high-performance foundation and reasoning AI models.
Alibaba, for example, has become a global leader in open-source AI through its Qwen model series, which powers one of the world’s largest open-source AI ecosystems and has inspired over 100,000 derivative models. The company’s open-source strategy aims to democratize AI, making advanced technology accessible to small businesses, researchers, and developers, and accelerating scientific innovation. Baidu, too, has responded to the open-source movement by releasing its Ernie 4.5 and Ernie X1 models for free and pledging to open-source the Ernie 4.5 series by June. These moves are seen as strategic responses to the disruptive impact of DeepSeek’s R1 model, which is distributed under the permissive MIT License.
The open-source approach is also transforming the competitive landscape. DeepSeek’s R1 and Alibaba’s Qwen 3 now top global open-source AI model rankings, challenging the dominance of Western giants like OpenAI and Meta. OpenAI, for its part, has announced plans to release an open-source model, though it will lag behind its most advanced offerings to avoid fueling competition from China.
Beyond technical collaboration, Tencent’s partnership with DeepSeek extends to product integration. In early 2025, Tencent began beta testing an “AI Search” feature within Weixin (WeChat), powered by DeepSeek’s R1 model. This move allowed WeChat’s massive user base to access advanced AI-powered search and question-answering capabilities directly within the app, further validating DeepSeek’s technology and accelerating its adoption across China’s digital ecosystem.
Other articles you may find interesting
Summary
- DeepSeek’s R1 model achieved viral success, briefly surpassing OpenAI’s ChatGPT as the most-downloaded free app on Apple’s App Store in China, highlighting the rapid domestic adoption of homegrown AI solutions.
- The open-source AI movement in China is often compared to the “Android moment” in mobile technology, marking a shift from closed, proprietary systems to open, collaborative platforms that can drive faster innovation and broader adoption.
- Open-source AI models like DeepSeek’s R1 and Alibaba’s Qwen are not only free to use and modify but also offer commercial APIs at lower costs than Western competitors, making them attractive for businesses seeking to integrate advanced AI at scale.
- The Chinese government and tech industry see open-source collaboration as a strategic lever to close the AI gap with global leaders, leveraging collective expertise and resources to accelerate progress.
In summary, DeepSeek’s partnership with Tencent is a microcosm of China’s rapidly evolving AI ecosystem, where open-source collaboration, technical innovation, and strategic integration are reshaping the global AI landscape and challenging the dominance of established Western players.