The New Reality: Chinese Models Carry Nearly Half of Enterprise AI Traffic
Chinese AI models are ultra-low-cost large language systems such as DeepSeek and Qwen that enterprises call over APIs for tasks like customer service, content generation and analytics, and they are rapidly winning market share because they deliver comparable performance at a fraction of the price of Western alternatives. The data is stark: since February 8, Chinese-origin models have carried at least 30% of enterprise token volume on OpenRouter, hitting a weekly peak of 46%, up from 11% a year earlier and 4.5% in the first half of 2025. This is not an experiment at the margins; it is a structural shift in where serious AI workloads run. After a week of intensive meetings across Asia, executives describe AI usage fees as a new, escalating cost category that sits uncomfortably alongside traditional per-screen or seat licenses. Rising bills are driving buyers to search aggressively for cheaper Chinese AI models as an escape hatch from the token-based cost trap.
The Cost Shock: Why DeepSeek and Qwen Are Winning Procurement
The core advantage of Chinese AI models is cost, and the gap is wide enough to force board-level conversations. In China, comparatively inexpensive energy helps data centers run at much lower operating cost than their Western counterparts. Backed by government support and private capital, local providers now use pricing as a strategic weapon. According to benchmarking data, a standardized intelligence task with DeepSeek can cost as little as USD 0.02 (approx. RM0.09), compared with about USD 2.75 (approx. RM12.65) using Anthropic’s Claude. On the API side, DeepSeek’s V4 Flash is priced at USD 0.14 (approx. RM0.64) per million input tokens, while OpenAI’s GPT-5.5 is USD 5.00 (approx. RM23.00) for the same volume. Many large customers report savings of up to 90% when switching workflows to these Chinese AI models. At that scale, the question is no longer whether to try a cheaper provider; it is which legacy contracts to unwind first.
Enterprise Behaviour: Model Arbitrage Becomes Standard Practice
Enterprises are responding with cold, economic logic: they are mixing models and arbitraging the price–performance spread. AI aggregation platforms and routing services let teams send different tasks to different engines through a single interface. High-value or security-sensitive work may still go to premium US models, but routine workloads are increasingly redirected to cheaper Chinese alternatives. DeepSeek already accounts for 17.6% of all tokens routed through OpenRouter—around 5.13 trillion tokens a week—making it the single largest vendor on the platform. Alibaba’s Qwen holds 13.9%, or 2.77 trillion tokens weekly. This is what real DeepSeek enterprise adoption looks like: procurement teams systematically routing non-sensitive queries to Chinese engines to protect budgets. Airbnb’s chief executive has said the company runs Qwen for customer service because it is "fast and cheap", and other firms point to similar savings when they stop paying premium API rates for every low-complexity call.
OpenAI’s Pricing Problem and the Open-Weight Threat
This shift is a direct challenge to Western AI vendors whose business models depend on premium services and consumption-based usage fees. Both OpenAI and Anthropic have cut prices this year, but neither has closed the gap to DeepSeek’s or Qwen’s rates, and every buyer who notices becomes a data point in next year’s budget discussions. The threat is not only about cheaper hosted APIs. Many Chinese developers are embracing open-source and open-weight strategies, allowing organizations to deploy models on their own infrastructure or through third-party clouds. A striking example is Moonshot AI’s Kimi K3, a 2.8 trillion parameter open-weight model with a million-token context window whose full weights are scheduled for release on July 27. Once that happens, enterprises can fine-tune and self-host frontier-class AI without sending a single API call to established Western providers. Regulators are starting to push back—questioning companies about their ties to Chinese AI and raising data security concerns—but enforcing bans on open-weight models already downloaded and locally deployed is far from straightforward.
What It Means for AI API Pricing—and Everyday Users
The direction of travel is clear: the next phase of AI adoption will be decided by economics, not brand loyalty. Chinese AI providers are entering the market with a cost structure that Western rivals struggle to match, and price competition from these models is forcing Western labs to rethink token-based pricing rather than treat it as a permanent, high-margin pillar. For ordinary users, the impact is already visible. In fields like digital signage, lower-cost AI can become the catalyst for large-scale deployment, enabling network operators, retailers and media owners to rely on AI-generated content, automated campaign creation, audience analytics and real-time adaptation without blowing up their operating expenses. Enterprises will keep routing sensitive analytics and code generation to trusted Western models, but they will outsource commodity content and customer interactions to cheaper Chinese engines. If OpenAI and its peers do not redesign their pricing around this reality, they risk watching more of their customers’ tokens—and budgets—flow elsewhere. The conclusion is blunt: in an AI world now defined by price, any provider charging a premium must prove it is worth the difference every single day.
