NextFin

Chinese AI Models Gain Ground as OpenAI and Anthropic Costs Surge

Summarized by NextFin AI
  • Chinese AI models are increasingly capturing U.S. enterprise workloads, with their share of tokens used by U.S. companies exceeding 30% since February and peaking at 46% due to significant cost advantages.
  • Open-source Chinese models can be 60% to 90% cheaper than leading American models, prompting companies to prioritize cost-effectiveness over brand prestige in AI deployment.
  • Adoption data shows a dramatic increase in the usage of Chinese models, with GLM 5.2 experiencing a 27x growth in daily token volume and an 80x increase in customer adoption within a week.
  • The enterprise AI market is shifting from model selection to workload management, where cost per task is becoming a critical factor in decision-making, favoring cheaper models for routine tasks.

NextFin News - Chinese AI models are taking a larger slice of U.S. enterprise workloads just as the economics of using top-tier American models get more expensive. OpenRouter says the share of tokens used by U.S. companies on Chinese AI models has stayed above 30% every week since Feb. 8 and has risen as high as 46%, while Vercel says Z.ai’s GLM 5.2 was the fastest-adopted model it tracked in 2026.

The shift is not about geopolitics in the abstract. It is about bill size. OpenRouter’s Justin Summerville said open-source Chinese models can be “60% to 90% cheaper” than leading Anthropic and OpenAI models, and that cost gap is large enough to change how companies route everyday tasks. As a result, more customers are using the premium models only where they really need them and sending routine work to cheaper alternatives that are good enough.

That change in routing is already showing up in usage data. Vercel said GLM 5.2 saw daily token volume grow about 27x and the number of customers using it grow about 80x in its first full week after launch. In other words, Chinese models are not just earning attention; they are earning workload.

The broader message for OpenAI and Anthropic is uncomfortable. Their most advanced models still matter for hard tasks, but the market for routine inference is getting more price-sensitive. When enterprises start measuring AI by cost per useful task rather than raw model prestige, the cheapest capable model can win share quickly.

Why Cost Is Now Beating Brand

The enterprise AI market is moving from model selection to workload management. Once a company starts deploying AI across support, coding, research, and internal tools, token spend becomes a recurring operating cost. Even modest differences in per-token pricing can turn into real money at scale, and that makes the cheapest capable model increasingly attractive.

OpenRouter’s data show that Chinese models are no longer fringe choices. A weekly share above 30% since Feb. 8 suggests they have become a standard part of enterprise routing. The fact that the share has climbed to 46% shows the usage is not static; companies are willing to move more traffic when the price-performance trade-off looks better.

That is why open-weight systems matter so much. They can be deployed privately, tuned for specific tasks, and slotted into workflows without forcing every request through the highest-priced API. For many companies, the practical question is no longer whether a model is the very best in the world. It is whether the model is good enough to save money on the 80% of tasks that do not require the absolute frontier.

“Chinese AI models are particularly attractive to American companies now as AI costs skyrocket,” said Kyle Chan, fellow in the John L. Thornton China Center at Brookings. “Where previously U.S. companies were prioritizing AI adoption regardless of model, now they're getting more cost-conscious.”

That shift in buyer psychology matters more than any single benchmark headline. Once procurement teams become cost-conscious, the premium model no longer gets the default win.

What The Adoption Data Say About Chinese Models

The usage numbers from OpenRouter and Vercel point to a concrete change in behavior, not just enthusiasm for a new release. A model that grows daily token volume 27x and customer adoption 80x in one week is being plugged into real workflows. That kind of jump usually reflects a mix of quality, price, and ease of use that satisfies enough of the market to matter.

The important point is that adoption does not require universal superiority. It only requires enough performance at a meaningfully lower cost. Chinese providers have benefited from exactly that combination. As technical quality has improved, the price advantage has remained large enough to pull usage toward them.

For enterprise buyers, that creates a segmented market. The most demanding tasks still justify premium U.S. models. But routine inference, internal automation, and many coding or research tasks can be routed to a cheaper model if the output is close enough. That is the real opening for Chinese AI companies: they are not replacing every American frontier system. They are taking the work that does not need one.

“Price is doing the work here,” said Harpreet Arora, head of agentic infrastructure at Vercel. “When a task doesn’t need the best model, teams are beginning to route it to the cheapest one that’s good enough, and the recent wave of models coming out of China is winning that trade.”

That is the market in one sentence. The cheapest good-enough model is often the one that gets deployed first and most often.

Why OpenAI And Anthropic Face More Pressure Than Before

OpenAI and Anthropic still have the strongest brands in frontier AI, but their pricing power is getting tested. The more companies see large savings from Chinese models, the harder it becomes to justify paying a premium for every single request. That does not eliminate demand for the best models. It narrows where that demand shows up.

This is the key shift: expensive models are moving from default infrastructure to specialty tools. If that pattern continues, the biggest U.S. labs could find themselves with fewer low-value tokens and a more concentrated mix of high-value tasks. That may preserve usage growth, but it puts more pressure on margins and on the assumption that premium APIs can remain the standard choice for all enterprise work.

The question for customers is simple. If a model is close enough on quality and far cheaper on cost, why keep paying for the premium one on routine work? That question is driving more of the routing decisions now, and it is the main reason Chinese models are gaining ground.

The answer is not that U.S. frontier models have lost relevance. It is that the market has become more selective. The best model still wins the hardest work, but it no longer gets the easy work by default.

What Could Slow The Shift

The move toward Chinese models is real, but it is not automatic. Enterprise buyers still care about security, governance, reliability, and vendor risk. Some companies will keep sensitive workloads on U.S. providers even if the Chinese option is cheaper. Others will require private deployment or strict controls before moving any traffic.

There is also a ceiling on substitution. For the hardest reasoning and code tasks, the best proprietary models can still justify higher prices. That means the biggest gains for Chinese providers are likely to come first in high-volume, lower-stakes use cases, where cost matters most and small quality differences are acceptable.

Even so, the direction is clear. AI procurement is becoming a question of return on tokens, not just model prestige. As long as Chinese model makers keep closing the capability gap while staying materially cheaper, they will keep winning share in enterprise routing.

For OpenAI and Anthropic, that means the fight is no longer just about who has the smartest model. It is about who can stay good enough at a price that companies are willing to pay at scale.

The market is not choosing between American and Chinese AI in the abstract. It is choosing between premium and affordable enough. Right now, affordable enough is winning more of the work.

Explore more exclusive insights at nextfin.ai.

Insights

What historical factors led to the rise of Chinese AI models in the U.S. market?

How do the technical capabilities of Chinese AI models compare to those of OpenAI and Anthropic?

What are the current market trends for enterprise AI models in 2023?

What feedback have users provided regarding Chinese AI models versus American models?

What recent developments have occurred in the pricing strategies of AI models?

What policy changes could impact the adoption of Chinese AI models in the U.S.?

How might the enterprise AI market evolve over the next five years?

What long-term impacts could the shift towards cheaper AI models have on innovation?

What challenges do Chinese AI models face in gaining wider acceptance in the U.S.?

What controversies exist around the security and governance of Chinese AI models?

How do Chinese AI models compare to competitors in terms of cost-effectiveness?

What are some historical cases that demonstrate the importance of cost in technology adoption?

How does the performance of cheaper AI models influence market dynamics?

What are the implications of shifting from brand prestige to cost in AI procurement?

What metrics are companies using to evaluate AI models beyond just price?

What specific tasks are being routed to Chinese AI models instead of premium models?

How significant is the impact of token pricing on enterprise AI workloads?

What strategies might OpenAI and Anthropic adopt to compete against cheaper models?

Search
NextFinNextFin
NextFin.Al
No Noise, only Signal.
Open App