<div id="readability-page-1"><div> <p>DeepSeek, the Chinese AI startup that caught global attention with its affordable models last year, has just given a sneak peek at its newest AI system, the V4. This time around, the model is designed to run seamlessly on Huawei's Ascend chip technology, marking a shift from its previous reliance on Nvidia processors.</p><p>The collaboration with Huawei isn't just surface-level. Huawei claims its entire Ascend supernode lineup now supports DeepSeek's V4 series, underscoring a strategic move to bolster domestic semiconductor capabilities amidst ongoing U.S.-China tech restrictions.</p><p>Performance-wise, DeepSeek's V4 reportedly surpasses all open-source competitors on world knowledge benchmarks, with the only exception being Google's closed-source Gemini-Pro-3.1. There's also a flash version of V4, focused on cost efficiency. Early previews mean tweaks are still underway, and the company hasn't pinned down an official release date yet.</p><p>DeepSeek's progress unfolds against a tense backdrop. Just a day before the preview, the U.S. government accused China of mass-scale AI intellectual property theft, complicating international tech diplomacy ahead of an upcoming summit. DeepSeek itself has faced allegations regarding misuse of Nvidia chips under U.S. export controls and claims of copying proprietary AI tech from companies like Anthropic and OpenAI.</p><p>While the startup confirms its use of Nvidia hardware for training previous models, it remains tight-lipped about specifics related to export bans. It also asserts its data sources for earlier models avoided synthetic data from OpenAI, relying instead on natural and web-crawled data.</p><p>The Chinese Embassy in Washington dismissed the U.S. claims as unfounded, reiterating Beijing's commitment to intellectual property protections. Yet, the broader push by China towards tech self-reliance is evident, with Huawei's AI chip advancements playing a pivotal role.</p><p>This V4 rollout hasn't been without market ripple effects. Shares of some Chinese AI competitors felt the pressure - for example, Zhipu AI dropped 9%, while MiniMax lost 7% in response to V4's launch. DeepSeek, backed by High-Flyer Capital Management, is reportedly eyeing a valuation north of $20 billion as it gears up for another fundraising push, with tech behemoths Alibaba and Tencent in talks to invest.</p><p>The fusion of a leading AI startup with Huawei's chip expertise paints a picture of China's determined march towards AI independence, even as the geopolitical chess game with the U.S. intensifies. How this will shape the competitive dynamics in AI and semiconductor markets remains an unfolding story.</p> </div></div>

TCEHY

TCTZF

<p>DeepSeek, a Chinese AI startup, has unveiled its new AI model, V4, optimized for Huawei's Ascend chips, moving away from Nvidia processors. This collaboration aims to enhance domestic semiconductor capabilities amid U.S.-China tech tensions. The V4 model reportedly outperforms all open-source competitors except Google's Gemini-Pro-3.1. The launch has impacted the market, causing declines in shares of rival companies. DeepSeek is also seeking a valuation over $20 billion with potential investments from Alibaba and Tencent, highlighting China's push for AI independence amidst geopolitical challenges.</p>

- DeepSeek, a Chinese AI startup, has previewed its new V4 AI system, which operates on Huawei's Ascend chips instead of Nvidia processors.  
- This collaboration reflects China's strategic push for semiconductor self-reliance amid U.S. tech restrictions and highlights the V4's performance, which reportedly outperforms all open-source rivals except Google's Gemini-Pro-3.1.  
- As DeepSeek navigates claims of intellectual property theft and faces a competitive market, it seeks a valuation over $20 billion while ramifications for Chinese AI competitors are already evident.

DeepSeek Unveils Huawei-Optimized AI Model Amid U.S.-China Tech Tensions