Chinas AI Surge Sparks Controversy as Usage Outpaces US

Global AI Race Intensifies as Chinese Model Usage Surpasses U.S., Anthropic Levels "Distillation" Accusations

A new front in the global artificial intelligence competition has emerged with data indicating Chinese-developed large language models (LLMs) are now being used more heavily than their U.S. counterparts among international developers, even as a leading American AI firm alleges industrial-scale intellectual property theft by Chinese companies. The developments unfold against a backdrop of massive capital inflows, evolving corporate strategies regarding AI integration, and significant advancements within China's domestic semiconductor and autonomous driving sectors.

Anthropic's Allegations and a Shifting Competitive Landscape The AI sector was roiled this week by public accusations from Anthropic, the company behind the Claude models. In a statement, Anthropic claimed three prominent Chinese AI firms—DeepSeek, Moonshot AI (creator of Kimi), and MiniMax—had conducted what it termed an "industrial-scale distillation attack." The company alleged these entities used "24,000 'sock puppet accounts' to engage in 16 million conversations" with the purpose of extracting Claude's capabilities to train their own models, a practice sometimes referred to as model distillation.

The allegation, which echoes long-standing tensions over technology transfer, was met with public derision from Elon Musk. The Tesla and xAI founder posted a message broadly interpreted as mocking the accusation, implying hypocrisy given the foundational training data of many Western models. Neither DeepSeek, Moonshot AI, nor MiniMax provided immediate public comment on the specific charges.

The controversy coincides with a significant milestone in the practical adoption of Chinese AI models. Data from OpenRouter, a major global aggregator of AI model APIs, shows that for the week of February 16-22, Chinese models were called for a total of 5.16 trillion tokens. This not only represented a 127% increase over a three-week period but also substantially exceeded the 2.7 trillion tokens recorded for U.S. models during the same week. Notably, the prior week had already seen Chinese models (4.12 trillion tokens) outpace American ones (2.94 trillion tokens).

Perhaps more tellingly, Chinese models now dominate OpenRouter's global usage rankings. Four Chinese LLMs—MiniMax's M2.5, Moonshot AI's Kimi K2.5, Zhipu AI's GLM-5, and DeepSeek's V3.2—occupied four of the top five spots, collectively accounting for 85.7% of the tokens used among the top five models. The platform's user base, comprised of 47.17% U.S.-based developers with only 6.01% from China, suggests these figures reflect genuine global developer interest rather than domestic usage, marking a potential inflection point in the soft power of China's AI ecosystem.

Market Dynamics and Reassurances from Silicon Valley The competitive fervor is being fueled by unprecedented investment. OpenAI announced it had secured a monumental $110 billion in new funding, a war chest earmarked for further scaling its frontier model research and infrastructure. This follows a period of intense capital deployment across the industry.

Amidst concerns that generative AI might render traditional software companies obsolete, leading tech executives have pushed back forcefully. Nvidia CEO Jensen Huang reiterated his stance that the market has "misjudged" AI's threat to the software sector. "All of these tools we use today... exist for a fundamental and legitimate reason," Huang stated. He argued that AI assistants would become "intelligent software" that leverages these existing tools on behalf of users, dramatically boosting productivity rather than replacing the platforms themselves.

Echoing this sentiment, Salesforce CEO Marc Benioff directly addressed what he called "SaaSpocalypse" theories during an earnings call. "If there was a SaaSpocalypse, it might have been eaten by Bigfoot because there's so many companies using SaaS now—and it's getting better because of AI agents," Benioff quipped. He highlighted that AI-native companies like Anthropic themselves "use a lot of SaaS." Salesforce's own AI platform, Agentforce, reported an Annual Recurring Revenue of $800 million, surging 169% year-over-year, underscoring Benioff's point that AI agents are making enterprise software "more critical," not less.

The Expansion of China's AI Ecosystem The data on model usage aligns with robust growth within China's broader AI infrastructure. A report from Frost & Sullivan revealed that in the second half of 2025, daily enterprise-level LLM calls in China skyrocketed to 37.0 trillion tokens, a 263% increase from 10.2 trillion in the first half. Alibaba Cloud's Qwen model saw its market share nearly double to 32.1%, indicating consolidation among front-runners.

Financial results from major Chinese tech players illustrate the commercial traction. Baidu's AI Cloud unit reported full-year 2025 revenue of RMB 30 billion ($4.1 billion), with subscription-based revenue from AI high-performance computing facilities growing 143% year-over-year in Q4. Huawei, while not breaking out AI-specific revenue, announced its total 2025 sales exceeded RMB 880 billion ($121 billion). The company also revealed its HarmonyOS ecosystem now runs on over 40 million terminal devices.

Product launches are accelerating the ecosystem's maturity. Alibaba Cloud's "Bailian" platform launched a "Coding Plan" offering API access to four leading open-source models—Qwen3.5, GLM-5, MiniMax M2.5, and Kimi K2.5—allowing developers to switch between models freely, a service it claims is unique among global cloud providers. Huawei Cloud released the public beta of its CodeArts coding agent, which integrates models like GLM-5.0 and DeepSeek-V3.2, claiming a 30% token saving on equivalent tasks and a "zero data leakage" promise by storing code locally.

Beyond software, AI adoption is hitting consumer scale in applications. Ant Group disclosed its "AI Pay" feature on Alipay has surpassed 100 million users, processing over 100 million transactions, claiming it as the world's first AI-native payment product to reach that dual milestone. Its health-focused "A-Fu" app also surpassed 100 million total users.

Semiconductor Sector: Innovation and Supply Pressures The insatiable demand for AI compute is reverberating through the semiconductor supply chain, particularly in China where domestic substitution efforts are ongoing. Several Chinese chipmakers, including Xin Jie Neng, Hongwei Technology, China Resources Microelectronics, and Silan Microelectronics, have announced price increases generally above 10%. Industry analysts attribute this "wave of price hikes" to soaring demand from AI data centers and the new energy vehicle sector.

Despite supply chain pressures, there are signs of technological progress and financial health. Cambricon, a leading Chinese AI chip designer, released a performance forecast showing a dramatic turnaround for 2025, with revenue of RMB 6.497 billion (up 453.21% year-over-year) and net profit attributable to shareholders of RMB 2.059 billion, swinging from a loss to profitability. Hygon Information, a domestic CPU leader, projected Q1 2026 revenue growth between 62.91% and 75.82%.

On the research front, a team from Peking University announced a breakthrough in ferroelectric transistors, a key component for memory and logic chips. The researchers created what they claim are the smallest and lowest-power ferroelectric transistors to date, published in Science Advances, which could provide core device support for improving the computing power and energy efficiency of future AI chips.

Strategic Maneuvers in Autonomous Driving The application of AI in autonomous driving illustrates the complex strategic calculations Chinese manufacturers are making. Great Wall Motor, a traditional automotive powerhouse, is pursuing a dual-track approach to close the gap in smart driving technology. The company is simultaneously leveraging external suppliers while building a substantial in-house R&D capability.

According to industry insiders, Great Wall has established a self-research team exceeding one thousand personnel, integrating smart cockpit and smart driving divisions. This team is led from the company's engineering hub in Baoding, while a Shanghai-based unit focuses on algorithm and software development. The company's Chief Technology Officer, Wu Huixiao, stated that smart driving now accounts for 50% of Great Wall's total R&D expenditure, with annual investment reaching RMB 1 billion.

To rapidly deploy capabilities across its vehicle lineup, Great Wall has structured a multi-supplier strategy across three compute platforms. For its mainstream ADC 2.0 platform, it partners with companies like Zhuoyu and Momenta to deliver features like highway NOA (Navigate on Autopilot). For mid-to-high-end models (ADC 3.0 and 4.0), it has partnered deeply with autonomous driving company Yuanrong Qixing (DeepRoute.ai) to implement end-to-end models and the more advanced Vision-Language-Action (VLA) architecture on Nvidia's Orin and Thor chipsets.

"We are persisting with advancing both cooperation and self-research simultaneously," a Great Wall internal source stated. The strategy appears to be yielding commercial results. The company's premium brand, Wey, saw sales grow 86% in 2025 to 102,000 vehicles, a lift attributed in part to the rollout of its VLA-powered systems. Executives have stated the goal is to democratize high-end assisted driving features, with models like the ORA 5 already offering city NOA at a price point of RMB 120,000-130,000 ($16,500-$17,900).

Brokerage International Investment and Standardization The global nature of the AI push was further highlighted by Hyundai Motor Group's announcement of a planned 9 trillion won ($6.3 billion) investment in AI, robotics, and hydrogen businesses in South Korea's Saemangeum region. The plan includes building an AI data center and a robotics manufacturing facility.

Within China, efforts to systematize and lead in emerging fields are accelerating. The country released its first national-level standard system for humanoid robots and embodied intelligence, covering the entire industrial chain from components to safety and ethics. This move aims to catalyze规范化 development in a sector attracting significant interest, evidenced by XPeng Motors' CEO He Xiaopeng announcing plans to achieve mass production of robots, flying cars, and robotaxis by 2026, with a new generation "IRON" humanoid robot starting production by year-end.

Similarly, Xiaomi CEO Lei Jun outlined the company's strategic focus for its next five-year plan, highlighting core underlying technologies including chips, AI, and operating systems as critical to its ambition of becoming a global hard-core technology firm.

The confluence of events—from allegations of corporate espionage and shifting usage patterns to massive investments and strategic industrial pivots—paints a picture of an AI landscape where technological advancement, geopolitical friction, and market competition are inextricably linked. As Chinese models gain traction globally and domestic companies vertically integrate AI across chips, software, and end applications like cars, the industry appears to be entering a new phase of both decentralized innovation and intensified rivalry.