From Silicon Valley to Europe, Why Are Enterprises Turning to Chinese AI Models? The Cost-Performance Revolution Is Rewriting Tech Rules

One-line conclusion: Twenty of the world's top 50 most-used AI models now come from China, and in June Chinese model token usage was 85% higher than US models—companies like DoorDash, Siemens, and Airbnb are switching to Chinese AI driven by costs 5-10x lower and capabilities rapidly catching up.

Think the AI race is only about American winners? The reality might surprise you.

According to a viral post by financial analyst Kobeissi Letter (over 266,000 views, 1,500 likes), a startling data point emerged:

Of the world's top 50 most-used AI models, 20 come from China—up 400% since 2025. During the same period, US models fell from 33 to 28.

Even more dramatic: in June, Chinese model token usage reached 98 trillion, a 113% month-over-month increase. US model usage was 53 trillion, up 43%. Chinese model usage is now 1.85x that of US models—and in May, it was only 1.24x.

Kobeissi Letter data: Chinese models surpass US models in both count and token usage among the world's top 50 AI models

Who's Using Chinese AI?

This isn't fringe experimentation. Globally recognized enterprises are adopting Chinese AI models at scale:

• DoorDash: The delivery giant uses DeepSeek models to reduce AI costs
• Siemens: Industrial powerhouse integrates Chinese AI into production optimization systems
• Airbnb: The booking platform uses Chinese models to improve search and recommendation algorithms

The Financial Times reported that these companies are drawn to Chinese AI for a straightforward reason: cheaper, increasingly capable, and easier to run on their own infrastructure.

Financial Times report: From Silicon Valley to Europe, enterprises turn to Chinese AI to cut costs

The Cost-Performance Revolution

Why aren't companies sticking with OpenAI, Google, or Anthropic?

The answer is simple: money.

US frontier AI model API prices have been climbing steadily. The cost per call for GPT-4 and Claude, when multiplied by billions of requests for enterprises, becomes astronomical.

By contrast, DeepSeek and other Chinese models cost 5 to 10 times less. More importantly, their capabilities are closing rapidly.

OpenRouter, a platform that tracks LLM token usage, shows Chinese models surpassed US competitors in token consumption this year. This isn't just about being cheap—if the quality were terrible, enterprises wouldn't use them.

Ali Ghodsi, co-founder of Databricks, shared a surprising finding on Twitter: in internal benchmarking across 11,000 employees and three hyperscalar clouds, the Chinese model GLM 5.2 performed "exceptionally well."

His tweet received over 416,000 views and 1,500 likes.

The Geopolitical Complexity

This shift carries particular sensitivity in Europe.

Last month, the Trump administration imposed export controls on Anthropic's Mythos and Fable models. This move forced European businesses to confront a fundamental question: what are the risks of depending solely on American AI technology?

Tech observer @andre_banandre noted that 46% of enterprise AI token traffic now runs on Chinese models. DeepSeek alone outpaces Google, Anthropic, and OpenAI individually.

This isn't an ideological choice—it's an economic one. When the cost difference is 10x and the quality gap is narrowing, it's hard for businesses to resist.

CoinBureau report: Global enterprises turn to Chinese AI to reduce costs

What Does This Mean?

For the Tech Industry

• US AI giants' pricing power is being challenged—if a Chinese model delivers 80-90% of the quality at 10-20% of the cost, what would enterprises do?
• Open-source AI is becoming mainstream—Chinese models' advantage partly comes from open-source strategies, allowing enterprises to run models on their own servers without worrying about API limits or data leakage

For Investors

• Chinese AI company valuations may be severely underestimated
• US AI giants may face long-term margin pressure

For Everyday Readers

• AI is becoming more democratized—no longer just US tech companies holding the technology
• Lower costs mean more businesses can use AI to improve efficiency, potentially benefiting consumers

Frequently Asked Questions (FAQ)

Q1: Are Chinese AI models really better than US ones?

A1: Not "better"—but "better value." In many tasks, Chinese models deliver 80-90% of US model quality at 10-20% of the cost. For enterprise applications that don't need top-tier capability, this is extremely attractive.

Q2: What is DeepSeek?

A2: DeepSeek is a Chinese AI company known for its high cost-performance language models. Its models are widely used on platforms like OpenRouter, and token consumption has surpassed most US competitors this year.

Q3: How is the US government responding?

A3: The Trump administration imposed export controls on some Anthropic models last month. Ironically, this accelerated European companies' diversification—if US tech has supply risks, why not diversify?

Q4: Is the 46% enterprise AI token figure on Chinese models accurate?

A4: This comes from @andre_banandre's analysis based on data from OpenRouter and similar platforms. OpenRouter is an aggregation platform that lets enterprises flexibly switch between models to optimize cost and performance.

Q5: What about data security? Isn't it risky to use Chinese models?

A5: A valid concern. But many Chinese models support private deployment—running on the enterprise's own servers without data leaving the company infrastructure. This mitigates data leakage concerns to some extent.

Q6: What impact does this have for Hong Kong and Taiwan?

A6: Both Hong Kong and Taiwan are key nodes in the global semiconductor and tech supply chain. The rise of Chinese AI may increase demand for local compute capacity (GPUs, data centers) and create new collaboration opportunities.

Q7: Can US AI companies maintain their lead?

A7: In top-tier research capability, US companies still lead. But in commercialization and cost-effectiveness, Chinese models are closing fast. The future landscape may be a dual-track system: "top-tier uses US, large-scale uses Chinese."

Tags: #ChineseAI #DeepSeek #AIcost #enterprise #open-source #costperformance