China's AI Models Now Process Tokens on Par with US

Key Takeaways

• Chinese AI models processed 4.37 trillion tokens in one week versus 4.98 trillion for US models
• China's digitally-delivered service exports grew 15.2% annually over 20 years to $245 billion
• Chinese models hit a record 12.96 trillion tokens in early April, outpacing US models by 4x

The Numbers Tell the Story

China's open-source AI models are processing data at volumes that rival their American counterparts. According to a Jefferies research report citing OpenRouter data, China's top nine AI models processed 4.37 trillion tokens in the week ending April 26. US models processed 4.98 trillion in the same period.

The gap is narrow. And it's shrinking fast.

That April 5 figure is striking. Chinese models processed more than four times the token volume of US models in a single week. While this appears to be a peak rather than a sustained rate, it demonstrates the raw capacity China has built.

Cheap Energy Powers the AI Engine

Jefferies points to two factors behind China's AI processing surge: cheap energy and abundant computing power. Running large language models requires massive amounts of electricity. Countries with lower energy costs have a structural advantage in AI infrastructure.

China has leveraged this advantage to build what Jefferies calls "digitally-delivered knowledge services." The country ranked as the world's sixth-largest exporter of these services in 2025. Its exports reached $245 billion, up from $14.4 billion in 2005. That's a 15.2% annualized growth rate over two decades.

India still leads this category with $328 billion in exports. But China's trajectory signals a shift from physical manufacturing to digital services. AI processing is becoming an export product.

Open-Source as Competitive Strategy

The token counts tracked by OpenRouter measure open-source model usage. This distinction matters. While US companies like OpenAI and Anthropic dominate commercial AI with proprietary models, China has invested heavily in open-source alternatives.

Open-source models can be deployed freely by anyone with computing resources. They're popular with developers, researchers, and companies that want to avoid vendor lock-in or API costs. High usage numbers suggest Chinese open-source models are gaining traction globally.

The strategy mirrors China's approach in other tech sectors. Offer capable alternatives at lower cost. Build adoption through accessibility. Establish market position before competitors can respond.

Geopolitical Context Complicates the Picture

The Jefferies report connects AI competition to broader geopolitical tensions. While the AI numbers favor China's momentum, other economic indicators show stress in allied economies.

Germany's GfK consumer climate index dropped 5.2 points to minus 33.3 in May. That's the lowest reading since February 2023 and the sharpest monthly decline since October 2022. German Chancellor Friedrich Merz publicly criticized US strategy on Iran, stating "the Americans clearly have no truly convincing strategy."

“The Iranians are clearly stronger than expected. The Americans clearly have no truly convincing strategy.”

— Friedrich Merz, German Chancellor

These comments mark an unusual break from Berlin's typically cautious approach toward Washington. European leaders are questioning the value of hosting over 40 US military bases with 85,000 troops. Some have floated direct negotiations with Iran to secure shipping access through the Strait of Hormuz, which handles 34% of seaborne crude oil and 19% of LNG.

Domestic Pressures Add Uncertainty

US polling shows 61% disapproval for military strikes on Iran and 63% disapproval of Trump's economic handling. These numbers add pressure for diplomatic solutions over military escalation.

The UAE's decision to exit OPEC on May 1 signals fractures within the Gulf region. Energy markets are already volatile. Additional instability could disrupt the computing infrastructure that both US and Chinese AI development depends on.

What This Means for AI Development

The token processing race is a proxy for AI capability. Models that process more data can learn faster, handle more users, and demonstrate market viability. China's numbers show it can compete at scale.

US advantages remain substantial. American companies lead in model architecture, commercial applications, and access to global talent. Export controls limit China's access to cutting-edge chips. But these barriers haven't prevented China from building competitive open-source alternatives.

The competition will likely intensify. Both nations view AI leadership as essential to economic and strategic power. The token counts are just one measure. But they suggest the race is closer than many assumed.

Frequently Asked Questions

How many tokens did Chinese AI models process in the record week?

Chinese AI models processed 12.96 trillion tokens in the week ending April 5, 2025, compared to 3.03 trillion for US models.

What is driving China's AI processing growth?

Jefferies cites cheap energy and abundant computing power as key factors enabling China to scale AI infrastructure cost-effectively.

How does China rank in digital services exports?

China ranked sixth globally in digitally-delivered services exports in 2025 at $245 billion, behind India's $328 billion.

Why does open-source matter in AI competition?

Open-source models can be deployed freely by anyone with computing resources, making them harder to restrict through export controls or licensing.

What is a token in AI processing?

A token is a unit of text that AI models process. It can be a word, part of a word, or punctuation. Token counts measure how much data models are handling.

Source: Tech-Economic Times / ET

Crimson Desert Endgame Combat Update

The new article is about a major update to the video game Crimson Desert, featuring new 'Rematch' and 'Re-blockade' mechanics for endgame players. This is entirely unrelated to the original article's topic regarding Chinese AI token processing statistics.