China’s AI development puts US rivals in ‘death zone’
A surge of model introductions from China’s AI industry is swiftly closing the divide with Silicon Valley, establishing what has been characterised as a perilous environment for those lacking innovative technology or disruptive pricing strategies. Alibaba Group Holding Ltd. has emerged as the latest entrant in a series of Chinese launches that are making significant strides in global benchmarks. This week, the company unveiled its Qwen3.8-Max, its most advanced model to date, which appears to match or even surpass Anthropic PBC’s flagship Fable 5. Two weeks earlier, Moonshot AI’s Kimi K3 demonstrated performance on par with the most expensive US options, all while operating on a significantly smaller budget. This development has raised new concerns regarding the efficacy of US chip sanctions aimed at curbing China’s technological advancement. Elsewhere, ByteDance Ltd. has eclipsed all competitors in video generation with its Seedance, the latest iteration being version 2.5, which has just been released. DeepSeek, the original Chinese disruptor of US hegemony in artificial intelligence, made a notable return this summer with V4 Flash, a groundbreaking model in terms of pricing. What was previously an isolated shockwave has evolved into a cascade of unforeseen depth, providing not only low-cost alternatives but also advanced capabilities in reasoning, coding, and complex tasks that present genuine competition to OpenAI and its American counterparts. “The most important change since January 2025 is that China’s progress no longer looks like a single-company breakthrough,” said Poe Zhao. “The first DeepSeek moment looked exceptional. The recent releases suggest China now has a repeatable system for producing models close to the global frontier.”
When considered alongside Z.ai’s GLM-5.2 released in June, which was the highest-ranked open-source model globally at that time, this rapid deployment of five models within eight weeks indicates that Chinese developers are either approaching or, in certain instances, surpassing the US pioneers typically regarded as leaders in frontier AI. The fundamental value proposition extends beyond merely offering low prices; it encompasses a commitment to radical efficiency. In benchmark testing conducted by the independent evaluator Artificial Analysis, the execution of a complex real-world workload incurs a cost of merely $0.03 when utilising DeepSeek V4 Flash, in stark contrast to the $3.15 associated with Claude Fable 5. When China requests cents while American companies demand dollars, the competition between the two nations for AI clientele — particularly those in other regions — begins to appear significantly altered. “A lot of countries are looking at the competition between US and China, and they don’t want to take a side right now because the competition’s just starting,” said George Chen. Both President Donald Trump and his counterpart Xi Jinping have made AI leadership a national priority. “When Xi visits the White House in September, because of Kimi, because of Alibaba and the model they just released last night, Xi will definitely feel like he has, in Trump’s language, more cards to play.” The rapid advances are intensifying the technological friction between the US and China, leading Washington to issue threats of sanctions due to concerns over intellectual property following Moonshot’s Kimi breakthrough.
Trump has indicated that his administration is considering the threat posed by China in relation to the necessity for enhanced safety measures on products such as Anthropic’s Fable or OpenAI’s GPT series. “We must exercise caution in both directions,” he remarked. “We do not wish to impose restrictions that would lead to a situation where we find ourselves in second place to China.” To attract developers, AI competitors must either undercut DeepSeek on pricing or exceed it in terms of raw performance. On a widely circulated benchmark chart from Artificial Analysis, a so-called DeepSeek death zone has now emerged. Increase the price for the same product or reduce capabilities while maintaining the same price, and you may as well forgo the effort altogether. Establishing clear boundaries around that danger zone has become crucial for sustained viability. Mid-market competitors are currently facing significant pressure to either reduce prices to align with the Hangzhou-based startup or invest substantially in developing more advanced models. The larger GLM-5.2, Kimi K3, and Qwen3.8-Max models occupy a superior performance tier relative to it. “Where I think DeepSeek V4 Flash has really changed the economics is agentic workloads,” said Dermot McGrath. When researching hundreds of companies, McGrath noted that he continues to utilise Anthropic’s Claude Code as a framework to formulate an action plan and summary brief. He subsequently delegates execution to DeepSeek within the identical environment, which deploys specialised AI agents to conduct queries and complete the task.
The cost of AI work has increased in recent months, while the emphasis on efficiency has become even more pronounced. That creates an opportunity for services to enhance, rather than replace, the most advanced US products. “A few months ago I wouldn’t have done that. The Chinese models weren’t as reliable at tool calling or long-running agent workflows,” McGrath said. Alibaba, addressing his exact pain point, made improvements in the so-called long-horizon execution of its new Qwen release. Previous models “were slower, more brittle and sometimes didn’t fully understand what they were supposed to do.” Beijing-based Z.ai has prioritised the enhancement of its long-horizon performance with the latest release. Chinese companies have engaged in a rapid iterative process over the past year, driven by internal competition that has expedited timelines and enhanced focus. DeepSeek disrupted the Chinese contest as significantly as it did the global landscape. Moonshot and Alibaba undertook the more audacious strategy of developing exceptionally large models, each exceeding 2 trillion parameters, reflecting their advanced capabilities. They subsequently moderated their energy and computing costs by activating only segments of the model sequentially. OpenAI and Anthropic, in contrast, maintain confidentiality regarding their parameter counts and refrain from disclosing specific figures. Parameters serve as the synaptic connections within an AI brain, facilitating the storage, processing, and response mechanisms through the integration of additional information. The two leading AI laboratories in the United States are preparing for initial public offerings, aiming for market valuations of no less than a trillion dollars, a remarkable target that relies on high-margin business strategies.
Yet China’s inexpensive and open-source competition now poses a challenge to their pricing power. “If there weren’t these Chinese open-source models, OpenAI and Anthropic would be laughing all the way to the bank,” said Kai-Fu Lee. “Now there’s an alternative, and it’s cheaper.” Chinese companies are indeed prioritising the acquisition of users, developers, and recognition in tech leadership over the potential for substantial short-term profits. “China’s AI providers have become trapped in a brutal price war, prioritising market share over profitability,” stated Rob Lea. Beyond core language models, China’s AI momentum is most evident in its dominance in the field of video generation. The market has been left wide open following OpenAI’s decision to shelve its Sora tool, citing the high costs associated with operating a resource-intensive service. Kuaishou Technology’s Kling AI video tool, stepping into the void, recently secured backing from Alibaba and Tencent Holdings Ltd. in a $2.8 billion financing round. A significant part of China’s strength stems from the readiness of companies and investors to overlook apprehensions regarding short-term returns. No company has exemplified that approach more than ByteDance, according to BI’s Lea. It has invested heavily in establishing its Seedance as the preeminent authority in quality for AI video creation, thereby intensifying the competitive pricing landscape. To what extent is the aggressiveness being assessed? “A 99% discount to prevailing market rates,” Lea said.
