Chinese AI Model Kimi K3 Sparks U.S. Security Concerns Despite Spending Gap

The Chinese AI model Kimi K3 has triggered renewed discussions in Washington about potential restrictions, even though the U.S. invests 23 times more in private AI than China. According to Stanford's 2026 AI Index Report, private AI investment in the U.S. is projected to reach $285.9 billion in 2025 compared to China's $12.4 billion.

Concerns Over Kimi K3 and U.S. Response

Developed by the Chinese startup Moonshot AI, Kimi K3 is an open-weight model accessible for download and operation on private servers. Its strong performance in coding benchmarks last week caused a significant reaction, unsettling U.S. chip stocks. This resurgence of competitive pressure has brought back discussions within the Commerce Department about adding Chinese AI laboratories to the Entity List, a trade blacklist that previously targeted Huawei in 2019.

also the National Security Agency considered issuing public warnings about Chinese AI capabilities, while the White House explored making American companies legally accountable if they hosted Chinese AI models that suffered breaches. These proposals were shelved due to concerns about innovation, but increased apprehension among security officials has revived these ideas.

David Sacks, an external White House AI advisor, stated that the leading closed-source AI developers, who dominate model revenue, aim to have the government suppress open-source competitors like Kimi K3.

Despite the controversy, demand for Kimi K3 remains strong. Moonshot AI halted new subscriptions within 48 hours of its launch and is preparing for an initial public offering in Hong Kong.

Examining the AI Spending Comparison

The reported 23-fold investment gap between the U.S. and China is somewhat misleading. Stanford's report notes that China's total AI funding includes approximately $184 billion from state-run guidance funds between 2000 and 2023, which is not reflected in the private investment figures.

U.S. AI funding is heavily concentrated in a few major companies, with the number of funding rounds exceeding $1 billion nearly doubling to 28 in 2025, led by OpenAI's $40 billion raise. Despite this, many U.S. firms prefer using Chinese AI models due to cost advantages. For instance, DeepSeek's V4 Pro charges $0.87 per million output tokens compared to Anthropic's Claude Fable 5, which costs $50 for the same volume.

Coinbase CEO Brian Armstrong mentioned in June that the exchange operates GLM 5.2 and Kimi K2.7 Code, cutting its AI expenses by about 50%.

This price competition has tangible impacts. DeepSeek's rise contributed to wiping out $589 billion of Nvidia's market value in one day in January 2025. The ban proposals circulating now originally emerged after that market reaction.

The effectiveness of potential bans is uncertain, as Kimi K3's model weights are already publicly available on repositories, according to reports from Tom's Hardware.

This material is informational and does not constitute financial advice.