From Kimi and DeepSeek to Qwen, why are U.S. companies cutting costs and improving efficiency making Washington anxious?
First, let’s share some major news with everyone—
On July 20 local time, Axios exclusively reported that as Moonshot (Moonshot AI) launches its latest major model, Kimi K3, and Chinese open-weight models such as DeepSeek R1/V3, Alibaba’s Qwen, and Zhipu’s GLM-5 continue to gain popularity in the U.S., the Trump administration is considering reviving multiple restrictions and “de facto bans” on Chinese AI models.
Axios used a strong term to describe the significance of this event: “a major move that could entrench the dominance of OpenAI and Anthropic.” In other words, the U.S. government is considering using administrative measures to remove Chinese AI from the U.S. market.
How do you proceed? Axios outlined several options held by the government:
Executive Order mandates accountability: It would require U.S. cloud platforms or companies hosting or using Chinese open-source models to bear 100% of the cybersecurity breach risk and compliance responsibility.
Entity List sanctions: The Department of Commerce is evaluating whether to add DeepSeek, Moonshot AI, Zhipu, and other laboratories to the Entity List, cutting off commercial licensing.
Spreading FUD and security warnings: The U.S. National Security Agency (NSA) and the Office of the National Cyber Director jointly issued a warning claiming Chinese models contain “backdoors” and security vulnerabilities, compelling U.S. companies to voluntarily withdraw due to compliance concerns.
These proposals were previously suppressed by officials advocating for free competition. But now, opposition has weakened—key figures such as Sriram Krishnan have left, and voices from the security hawks within the U.S. government are growing louder.
Keyword: Cost-performance ratio
Behind this debate lies a reality that makes the United States uneasy.
According to data from the AI aggregation platform OpenRouter, as of early July, approximately 58% of AI tokens processed by U.S. companies on the platform were completed by Chinese models, up from less than one-third in mid-January. Chinese models first surpassed U.S. models in March and reached as high as 63% during the first week of July.
The reason is simple: it's much cheaper.
Using the OpenRouter platform as an example, Anthropic’s Claude Opus 4.7 charges $25 per 1 million output tokens, while DeepSeek V4 Flash costs only $0.18. The AI startup Lindy switched 100% of its traffic from Claude to DeepSeek in June, with the CEO stating this will save millions of dollars over the coming months.
An increasing number of U.S. companies are extensively deploying Chinese open-source models as underlying commercial infrastructure, either privately or publicly, including:
The most popular AI programming tool in the United States, Cursor, has been revealed to have based its next-generation core code model (Composer 2 / K2.5) on an authorized secondary development of the Moonshot AI Kimi framework.
Major e-commerce platform Shopify and leading short-term rental company Airbnb have extensively integrated open-source models such as Alibaba's Qwen into their mid-tier services, such as merchant assistants and customer support systems. Shopify explicitly stated in its technical blog that, compared to proprietary solutions, costs have dropped by 68%.
Food delivery giant DoorDash has acknowledged offloading significant foundational processing tasks to the Moonshot Kimi model; automation tools startup Lindy has fully discontinued Anthropic in favor of DeepSeek.
The performance gap is also narrowing rapidly. The Stanford University 2026 AI Index Report states that the performance gap between Chinese and U.S. AI models has effectively disappeared. Ryan Fedasiuk, a researcher at the American Enterprise Institute, wrote in the report that Kimi K3’s performance has overturned the conventional belief that China’s best models lagged behind their U.S. counterparts by six to eight months—“this gap has now narrowed to a matter of weeks.”
The market is voting with its feet
On July 17, the day Kimi K3 was released, the U.S. tech sector suffered a sharp decline. The Philadelphia Semiconductor Index plunged as much as 5.7% intraday and ultimately closed down 1.63%, retreating more than 20% from its historical high at the end of June, officially entering a technical bear market. By the close on July 20, major players such as NVIDIA, Microsoft, Google, and Meta were all under pressure, as the market reassessed the competitive landscape of the AI industry.
On the other hand, Alibaba (BABA) strengthened逆势, as it is a major shareholder of Moonshot AI, the company behind Kimi K3.
According to Alibaba's fiscal year 2024 financial report, Alibaba invested approximately $800 million in Moonshot AI, acquiring about 36% equity. On July 20, Alibaba's Hong Kong shares surged by up to 6%, and its U.S. shares closed at $114.78. Previously, Alibaba's Qwen model was selected by Apple as the AI service provider for Apple Intelligence in China—capital markets are voting with real money on this relationship.
Bank of America analyst Alex Liu noted: "Despite ongoing hardware/compute constraints in China, K3 proof of pre-training scalability combined with architectural innovations can still deliver step-function improvements for China's flagship models."
Moor Insights & Strategy chief analyst Patrick Moorhead described the market reaction as an "overreaction strikingly similar to the DeepSeek panic," and believes large language models will only "accelerate and expand the reasoning market."
The cautious group reminds us: cheapness isn't the only consideration.
Bessemer Venture Partners partner Byron Deeter believes U.S. companies "will not adopt them simply because of lower token costs." Morningstar analysts note that U.S. companies, government agencies, or firms working with the government will not adopt Chinese open-source weighted models to save on inference costs.
American companies are voting with their feet, choosing value for money. Meanwhile, Washington seeks to block this path through executive orders.
This script is not unfamiliar to us: from smartphones to electric vehicles, these industries initially emerged in the West, but ultimately saw market dominance captured by China.
The same story is likely to be replayed in the AI field.
Author: Seed.eth
Twitter: https://twitter.com/BitpushNewsCN
BitPush Telegram community: https://t.me/BitPushCommunity
BitPush TG subscription: https://t.me/bitpush