Chinese AI Accounts for 58% of US Compute Consumption: How Open-Source LLMs Are Disrupting Silicon Valley

From 10% to 58%: A Quiet Migration in AI Consumption

According to an analysis shared by a Japanese tech expert on Bilibili, a striking dataset from OpenRouter — a popular open-source model aggregation platform — is upending assumptions about the global AI landscape. As recently as the first half of 2025, Chinese AI models accounted for less than 10% of total AI consumption in the US. By July 2025, that figure had surged to 58% — Chinese AI had crossed the halfway mark in actual usage.

OpenRouter is one of the most influential model aggregation platforms in the AI developer community. It provides a unified API endpoint that integrates over 400 large language models from different providers. Developers only need to connect to a single OpenRouter endpoint to freely switch between OpenAI's GPT series, Anthropic's Claude series, Google's Gemini series, and various Chinese open-source models — without registering for and maintaining multiple API accounts. This architecture makes it an important window for observing real-world AI model preferences globally, since users are making choices in a completely free market environment with no platform lock-in effects.

The expert specifically emphasized that this refers to "usage volume," not "revenue" — a crucial distinction. In the LLM space, consumption is typically measured in tokens. A token is the smallest unit of text a model processes: one English word corresponds to roughly 1 to 1.5 tokens, while one Chinese character typically maps to 1 to 2 tokens. When we say Chinese models have surpassed the halfway mark in token usage, it means that on the OpenRouter platform, more than half of the actual compute workload — including input prompts and model-generated responses — is flowing to Chinese models. However, since Chinese models are often priced at one-tenth or even less per million tokens compared to American models of equivalent performance, their revenue share is far lower than their usage share. This "high volume, low price" dynamic illustrates precisely how Chinese models are capturing real-world application scenarios through extreme cost-effectiveness. This tipping point occurred mainly between May and June 2025, completing the leap from minority to majority in just two months.

Use cheap, capable Chinese models for lightweight tasks, then switch to more expensive frontier models for high-difficulty work — it's precisely this "switch on demand" mechanism that lets users continuously vote with their usage, and the result is that more and more people are choosing Chinese AI.

Silicon Valley's Collective Pivot: DoorDash and Airbnb Lead the Way

What makes this even more striking is that the adopters aren't fringe players — they're well-known tech companies at the heart of Silicon Valley. The expert cited multiple cases of companies that have publicly or semi-publicly adopted Chinese open-source LLMs.

According to the analysis, US food delivery giant DoorDash is rapidly rewriting code and restructuring company-wide systems, using Kimi for virtually everything except a handful of sensitive components. Kimi is a large language model developed by Chinese AI startup Moonshot AI, founded in 2023 by researchers from Tsinghua University's AI lab. Kimi initially gained fame for its ultra-long context window (supporting 2 million character inputs) and has since continued to iterate on code generation, logical reasoning, and other capabilities. Its open-source version, Kimi-K2, has performed impressively on multiple programming benchmarks and is considered to match or even surpass Claude in code refactoring tasks — at a fraction of the cost.

Meanwhile, Airbnb CEO Brian Chesky publicly stated that Airbnb has adopted Alibaba's Qwen (Tongyi Qianwen) across the board. Qwen is a series of LLMs released by Alibaba Cloud, with its open-source version Qwen2.5 spanning multiple parameter scales from 500 million to 72 billion. It excels in multilingual understanding, tool calling, and long-text processing. Because Alibaba chose to open-source the model weights under the Apache 2.0 license, enterprises can commercially deploy and customize the model without restrictions — an enormous draw for companies like Airbnb that operate globally, require multilingual support, and are highly sensitive to data sovereignty.

An even more dramatic example comes from the well-known AI company Lindy. Lindy is a startup focused on AI-powered automation workflows, with a core product that lets users create AI assistants capable of autonomously executing complex tasks across customer service, sales follow-up, data analysis, and other scenarios. According to the expert, Lindy was once 100% dependent on Claude Code but then switched entirely to DeepSeek, saving hundreds of millions of yen in costs within a year and significantly boosting profits. DeepSeek was founded in 2023 by the Chinese quantitative investment fund High-Flyer. Its open-source model DeepSeek-V3 and reasoning model DeepSeek-R1 attracted global attention in early 2025. DeepSeek's core innovations lie in its Mixture of Experts (MoE) architecture and Multi-head Latent Attention (MLA) mechanism, which enable it to match or even surpass frontier models at training costs reportedly just a fraction of those for comparable US models. This demonstrates that Chinese AI is no longer a "toy" or "curiosity" — it's battle-ready and being deployed at scale in production environments within core US tech companies.

The expert also revealed an intriguing detail: a friend working at a major US tech company (a Microsoft-affiliated firm) told him that while the company hasn't publicly acknowledged it, they've been extensively using Kimi internally for some time. The number of companies willing to openly admit using Chinese AI models is growing.

You Don't Need a Ferrari for Every Trip: The Cost-Performance Logic of Chinese Models

Why would Silicon Valley companies — which sit atop their own cutting-edge research labs — make this pivot? The expert cited an analogy widely circulated in the industry: "You don't need a Ferrari for every trip."

The logic is straightforward: to run to the nearby supermarket, a compact sedan is more than enough — there's no need to drive a Ferrari. When enterprises actually break down their tasks by tier, they discover that the vast majority of scenarios don't require the most powerful, most expensive models. Chinese open-source models like DeepSeek and Kimi — the "economy cars" — perform far better than expected and can easily handle everyday needs.

This reflects an important technical reality in AI applications: approximately 70% to 80% of enterprise AI calls are "routine tasks" — text summarization, format conversion, basic code generation, simple Q&A, data cleaning, and so on. These tasks don't require the strongest reasoning capabilities or the largest parameter counts; a mid-tier model at an extremely low price can complete them with nearly identical accuracy. Only the remaining 20% to 30% of tasks — those involving complex multi-step reasoning, frontier scientific problems, or highly creative writing — truly require flagship models like GPT-4o or Claude Opus. Once enterprises recognize this, routing the bulk of routine traffic to low-cost models becomes the rational choice, and Chinese open-source models fill that gap perfectly.

From this, the expert extended a broader point: give the market the freedom to choose, and people will naturally gravitate toward the optimal solution. This also indirectly suggests that heavy-handed AI regulation policies may not work as intended — the market is voting with its feet, and the results are breaking down the supposed US monopoly on AI.

The expert also mentioned that Cursor, a prominent AI coding startup acquired by SpaceX for a massive sum, has been found to use Chinese models behind some of its new features and services. Cursor is a deeply customized AI-native code editor built on VS Code, developed by Anysphere. It deeply integrates LLMs into the programming workflow, supporting cross-file code comprehension, auto-completion, intelligent refactoring, and natural language instruction-based coding. Cursor rapidly became one of the most popular AI programming tools in the developer community in 2024, with its valuation at one point exceeding $10 billion. In 2025, reports indicated that Cursor was brought under the acquisition umbrella of Musk's xAI (rather than a direct SpaceX acquisition), a deal that itself reflects the strategic value of the AI coding tool space. When a company whose core product is AI itself starts adopting Chinese open-source models on the backend to reduce inference costs, the signal is especially powerful — even the most cutting-edge AI companies are acknowledging the practical value of Chinese models.

When even "AI-native companies" are running Chinese models behind the scenes, a consensus that "Chinese AI works just fine" is spreading through Silicon Valley.

The Autonomy Advantage of Open Weights

Beyond cost and performance, another key factor in the rise of Chinese models is "open weights." It's important to distinguish between "open weights" and "fully open source" — two concepts that are easily confused. Fully open source means publishing the complete training code, training datasets, data processing pipelines, model weights, and evaluation methods. Open weights primarily means publishing the model's parameter weight files, allowing users to download and run the model locally or on private cloud infrastructure, without necessarily disclosing training data or the full training pipeline. Most mainstream Chinese LLMs (such as DeepSeek, Qwen, and Kimi-K2) currently adopt the open weights model, accompanied by commercially friendly licenses.

Open weights mean enterprises can self-host and maintain full control over their models — a decisive advantage for data sovereignty, privacy compliance, and long-term operational costs. Specifically, enterprises can deploy models on their own servers or in cloud infrastructure within designated regions, ensuring that business data never leaves their jurisdiction and satisfying strict privacy regulations like GDPR (the EU's General Data Protection Regulation). Self-hosting also eliminates the risk of dependency on third-party APIs — no business disruption if a model provider adjusts pricing, changes terms, or faces geopolitical sanctions. Enterprises can also fine-tune models to more precisely fit specific business scenarios, which is typically restricted under closed-source API models. Instead of entrusting core business operations to a closed-source API, enterprises can truly take control.

From Kimi to DeepSeek to Qwen, Chinese open-source LLMs are penetrating global developers' daily workflows through the combination of "low cost + high performance + self-hosted deployment."

A Victory for Pragmatism and the Open-Source Spirit

Finally, the expert noted that the abundance of training courses teaching people to "make money with Claude Code" are essentially teaching people to "drive a Ferrari to the supermarket" — it looks premium but makes little economic sense. Claude Code is a command-line AI programming tool released by Anthropic that can understand an entire codebase directly in the terminal, autonomously execute edits, run tests, and perform Git operations. Because it's powered by flagship models like Claude Sonnet/Opus, API costs for a single complex coding session can run several dollars or even tens of dollars. For individual developers or small teams whose daily programming tasks don't involve complex architectural design or cutting-edge algorithm implementation, using open-source alternatives with equivalent capabilities but prices an order of magnitude lower (such as programming tools based on DeepSeek or Kimi) is often the more rational economic choice. The truly pragmatic approach is to match the right model to the task at hand.

Of course, the data and cases in this article primarily come from a single source (a Japanese expert's analysis reposted on Bilibili), and OpenRouter's statistics only reflect usage behavior on that platform — they cannot be equated with the true share of the global AI market. It's worth noting that OpenRouter's user base consists primarily of independent developers, small and mid-sized startups, and tech enthusiasts. Large enterprises typically consume AI through directly negotiated enterprise API contracts or self-built inference infrastructure, which wouldn't appear in OpenRouter's statistics. Therefore, the 58% figure more accurately reflects the usage preferences of the developer community and SMEs, not the full picture of the AI market. Even so, the trend remains profoundly significant — because today's preferences in the developer community often foreshadow where the enterprise market will head in the next one to two years.

Regardless of the exact numbers, the trend of Chinese open-source AI rapidly penetrating global markets through extreme cost-effectiveness — and even entering the core systems of US tech companies — is a reality that can no longer be ignored.

The essence of this disruption is not a victory of nationalist narratives, but a victory of pragmatism and the open-source spirit.