DeepSeek Was Just the Beginning: China's AI Race Explained

Just months after Chinese AI startup DeepSeek rattled Wall Street with a model that challenged Silicon Valley’s giants, a domestic rival is already claiming to have beaten it. This week, Shanghai-based MiniMax unveiled its M1 reasoning model, asserting that it outperforms DeepSeek’s latest release while using significantly fewer resources. MiniMax, backed by tech behemoths Alibaba and Tencent, claims M1 supports a one-million-token context window — eight times that of DeepSeek’s R1 — and was trained on just 512 Nvidia H800 GPUs.

The announcement is the latest salvo in a blistering artificial intelligence arms race inside China, where a cohort of well-funded startups and established tech titans are locked in a rapid-fire contest for dominance. The sudden ascent of DeepSeek in early 2025 was not an isolated event but a catalyst, accelerating a domestic battle fueled by national pride, vast investment and the pressure of US sanctions.

Who is DeepSeek and why did it shake the AI world?

In January 2025, a little-known Chinese company called DeepSeek released its R1 reasoning model, triggering what reports described as a $1 trillion stock market meltdown as investors feared a cost-effective challenger to Western AI supremacy. The company claimed its models could match or even outperform the latest offerings from OpenAI and Anthropic with far less computational power and capital. DeepSeek said it cost just $5.6 million to conduct the final training run for its V3 model, a fraction of the estimated $100 million-plus spent by OpenAI on its most advanced systems.

Behind the secretive company is Liang Wenfeng, a 40-year-old entrepreneur who previously founded the successful quantitative hedge fund High-Flyer Management. Described as a “true nerd” and a “mystery figure,” Liang transitioned from finance to AI, establishing DeepSeek in spring 2023 with the stated mission of establishing China as a force in the field.

DeepSeek’s key innovation was its pioneering use of a technique called “sparsity,” which makes AI models more efficient by activating only relevant portions of the model to answer a specific query, much like the human brain fires up specific neurons for a task. The breakthrough was so significant that Nvidia CEO Jensen Huang called the R1 model “genuinely a gift to the world’s AI industry.” Within weeks, Microsoft and Amazon began offering DeepSeek models on their cloud services, and AI search engine Perplexity integrated a version it fine-tuned and rebranded as R1 1776.

How are US sanctions fueling Chinese innovation?

DeepSeek’s focus on efficiency wasn’t just an academic pursuit; it was a strategic necessity. For years, the US has sought to slow China's technological progress by restricting its access to advanced semiconductors. Tensions peaked in 2022 and 2023 when Washington hit Beijing with multiple rounds of chip export controls, limiting sales from American firms like Nvidia, whose cutting-edge GPUs are the workhorses of AI training.

This created what one analyst called a “Darwinian pressure” on Chinese developers: do more with less. As Microsoft CEO Satya Nadella noted, “DeepSeek, and R1 in particular, was the first model I’ve seen post some points” in a landscape previously dominated by OpenAI. The restrictions forced companies to develop innovative workarounds. DeepSeek’s sparsity is one example; another is Huawei Technologies Co.’s rapid advancement in creating its own chips and systems, such as the CloudMatrix 384, which analysts say can be more powerful than Nvidia's flagship racks under certain circumstances. This push is part of President Xi Jinping’s national agenda for “self-reliance and self-strengthening” in key technologies.

Who are the other players in China's AI race?

DeepSeek is just one of several highly ambitious startups known as China’s “Little Dragons.” This elite group includes MiniMax, Moonshot AI, and 01.AI, many of which are backed by China’s largest internet companies. The success of DeepSeek reportedly forced many of these rivals to pivot from fundamental research to focus on more immediate applications.

The tech giants are also formidable competitors in their own right. Alibaba Group Holding Ltd.’s Qwen family of models, including Qwen2.5 and the latest Qwen3, consistently rank near the top of global leaderboards alongside models from Google and OpenAI. Baidu Inc.’s CEO has claimed his company can develop models just as good as DeepSeek’s but at an even lower cost.

MiniMax's M1 release shows how quickly the goalposts are moving. By claiming superior performance and efficiency over the very model that disrupted the industry six months prior, MiniMax underscores the hyper-competitive nature of the domestic market, where leadership is fleeting and innovation is constant.

What is the global strategy for Chinese AI?

Many Chinese firms, including DeepSeek and Alibaba, have strategically chosen to open-source their models. This approach serves multiple purposes. On one level, it democratizes AI. On a more strategic level, it undercuts the business model of Western proprietary labs like OpenAI and Anthropic. By making powerful models cheap or free, Chinese companies can accelerate global adoption, embedding their technology in the wider ecosystem before competitors can secure the market.

This strategy also involves what one DeepSeek researcher called “decoupling” values from the models. By allowing developers to customize a model’s ethical standards, a Chinese AI can be adapted to different cultural and political contexts, overcoming censorship-related barriers and making it more palatable for global users. This approach has proven successful, with DeepSeek’s models being incorporated into Western platforms like Microsoft Azure, Amazon Bedrock, and Perplexity.

What challenges lie ahead for China's AI champions?

Despite their rapid innovation, China's AI firms face significant headwinds. The most immediate challenge is the intense competition at home. As MiniMax’s recent launch demonstrates, any company that achieves a breakthrough immediately becomes a target for its domestic rivals to surpass. This relentless cycle puts constant pressure on firms like DeepSeek to innovate continuously or risk being overtaken.

Another major question is commercialization. While many of these companies are flush with venture capital, it remains unclear how they will generate sustainable revenue, particularly with an open-source strategy.

Finally, international scrutiny remains a persistent obstacle. A US House of Representatives committee report alleged “significant” ties between DeepSeek and the Chinese government and accused it of unlawfully stealing data from OpenAI. While these claims are denied, they highlight the geopolitical tensions and trust issues that Chinese tech companies must navigate as they expand globally. Even as they solve complex technical problems, their biggest challenges may be political.

Reference Shelf:

How US Chip Sanctions Are Building a Fortress for Huawei (ARPU)

China's MiniMax Says Its New AI Reasoning Model Beats DeepSeek (Bloomberg)

Who Is the Mysterious Founder of ChatGPT Rival DeepSeek? (Bloomberg)

Why Huawei Can Only Make 200,000 AI Chips a Year (ARPU)