DeepSeek, the Chinese Bolt From the Blue, Commands Global Stage

DeepSeek has sent another shockwave through the AI industry with this week's release of its newest model, the DeepSeek V4, a follow-up to its stunning 2025 release of a chatbot lower in cost and operation than global models.

The Hangzhou-based startup's latest large language model arrives in a market already crowded with high-profile US and Chinese competitors. Its impact, however, validates Bertrand Russell's assertion that "diversity is essential to happiness" – or in this case, an essential step forward in the robust, competitive advanced technology landscape.

The release provides a powerful, reliable and affordable alternative that aligns seamlessly with local needs and regulatory frameworks. Its arrival cements China's position as the only nation outside the US to possess a full-stack AI ecosystem, encompassing chips, cloud infrastructure and independent models. By ensuring compatibility with hardware firms including Huawei, DeepSeek has carved out a structural advantage that empowers sovereign innovation and reduces reliance on foreign hardware, including Nvidia's advanced chips.

Touted as the most powerful open-source large language model currently available, V4 is free for users to download and modify. DeepSeek has highlighted significant improvements in reasoning and agent capabilities, which allow the model to manage complex, multistep tasks. While the company claims V4 matches some top-tier US products, it has noted that performance still lags behind leading closed-source models such as Anthropic's Claude 4.6 and Google's Gemini 3.1 Pro.

Cheng Meng, a research manager at International Data Corp, noted that the global market is gradually splitting into two major camps: the US model and the Chinese open-source model. This division offers global companies more options in pushing international competition into new frontiers.

DeepSeek first caused a sensation in January 2025 by releasing an OpenAI-style chatbot for a fraction of the cost of Western models. DeepSeek V4 doubles down on these core selling points: open-source accessibility and extreme cost efficiency.

Technically, V4 marks a shift toward hardware independence. While previous models primarily relied on Nvidia's Cuda framework, DeepSeek V4 has been validated on both Nvidia and Huawei Ascend processors.

Domestic chip developers such as MetaX, Cambricon and Moore Threads have announced support for the long-awaited model. Furthermore, the China Academy of Information and Communications Technology has initiated testing of the V4. Given the group's role as a think-tank for the Ministry of Industry and Information Technology, the move signals national-level support for the model.

The core advantage of open-source AI lies in the democratization of innovation. By making source code and model weights transparent, it dismantles the "black box" of proprietary systems and fosters a global collaborative ecosystem. This transparency ensures security through scrutiny, allowing developers to identify biases and vulnerabilities that closed systems might hide. For Chinese state-owned giants in banking and telecommunications, this domestic infrastructure is the natural choice in as they navigate a complex geopolitical environment.

Price remains DeepSeek's sharpest edge.

Yang Hua, a marketing official at a Shanghai gaming firm, recently began using DeepSeek V4 to manage his files. On Monday, he said he spent only 0.56 yuan (0.07 US cents) on AI tokens – less than a 10th of what he paid for a previous US model. Yang noted that the efficiency and capacity are almost identical, and he no longer "worries about the financial burden of token costs."

Industry experts believe DeepSeek has reduced training costs by 60 percent through algorithmic innovation alone. When combined with hardware optimization from domestic chip manufacturers, the cost per application programming interface call falls dramatically.

DeepSeek V4 continues its global accessibility strategy with inference costs ranging from 1/8 to 1/50 of OpenAI's GPT-5.5, enabling widespread adoption for small companies and individual users, according to International Data.

On Monday, DeepSeek further slashed interface prices, cutting input cache costs by as much as 90 percent. According to pricing updates published on April 26, DeepSeek-V4 Flash input cache costs were reduced to 0.02 yuan per million tokens. The new business-focused V4-Pro model saw even deeper promotional cuts, lowering cached input costs to 0.025 yuan per million tokens.

Looking ahead, the company's marketing and technical capabilities may soon be bolstered by tech development and by the first-ever external investment. With the mass market launch of Huawei's Ascend 950 supernodes expected in the second half of this year, the price of DeepSeek V4 Pro is expected to drop significantly.

DeepSeek was founded in 2023 by entrepreneur and Chief Executive Liang Wenfeng, funded solely by his hedge fund High Flyer. But facing mounting competition and development costs, the company is now seeking its first outside investment, which could push its valuation to US$20 million, making it one of the world's largest AI startups. Alibaba and Tencent are reportedly among those in the line seeking to invest.