月之暗面开放Kimi K3权重与技术报告，港股云厂商受益与竞争传导相互抵消

# 美股盘后催化 7/11: Kimi K3开放权重、长上下文和Day 0 API伙伴上线，冲击闭源模型商业壁垒和AI capex叙事

- 美股会话: 2026-07-27
- 方向: MIXED；类型: 产品/技术；规划重大性: 8/10
- 映射模式: competitive
- 美股驱动（规划器初判，需独立核验）: 事实：Moonshot/Kimi发布Kimi K3权重和技术报告，称其为2.8T MoE、原生视觉、100万token上下文，并开源MoonEP、FlashKDA、AgentENV；Nebius、Together、Modal、Baseten、Fireworks等成为早期部署伙伴，Hugging Face热度快速上升。推断：高性能开放权重中国模型可能压缩闭源模型定价权和训练资本强度叙事，但也可能扩大推理部署、云托管和AI应用需求。

## 本催化相关美股真实收盘行情

- 英伟达 (NVDA.US): 196.51 -4.99%
- 微软 (MSFT.US): 389.1 +1.94%
- 谷歌 (GOOGL.US): 326.56 +2.13%
- Meta (META.US): 593.87 -0.22%
- 亚马逊 (AMZN.US): 231.39 -0.31%
- Nebius (NBIS.US): 187.88 +0.06%

## 亚太映射假设（仅候选，必须查证）

- 阿里巴巴 (9988.HK) [H/竞争替代/云平台/mixed]: 中国开放模型生态活跃利好云托管和企业AI需求，但也加剧通义等模型竞争。；核验问题: 核实阿里云是否上线Kimi K3及对Qwen价格/调用量影响。
- 百度集团 (9888.HK) [H/竞争替代/mixed]: 强开源模型可能压制闭源文心模型差异化，但扩大企业AI应用市场。；核验问题: 比较Kimi K3与文心在中文、代码、Agent和企业API价格上的差距。
- 腾讯控股 (0700.HK) [H/客户/竞争替代/mixed]: 开放模型降低应用开发成本，可能利好腾讯云、广告和游戏工具，但自研模型也承压。；核验问题: 核实腾讯是否接入Kimi K3或推出对标开放权重策略。

## 必答问题

- Kimi K3许可证商业限制是否影响大型企业采用？
- 开源模型对GPU/HBM需求是降低训练资本强度，还是扩大推理总量？
- 美国是否会限制中国开放权重模型在政府和企业环境使用？

## 原始证据全文（15 条）

### [E0744] Kimi (Moonshot AI) | 2026-07-27 15:14 UTC | 盘中
Releasing the model weights and technical report of Kimi K3.

Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.

New model architecture: 2.5x the intelligence per unit of com
URL: https://x.com/Kimi_Moonshot/status/2081760186235289764
Releasing the model weights and technical report of Kimi K3.

Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.

New model architecture: 2.5x the intelligence per unit of compute, not just more params.

Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale.

Model weights: https://t.co/7m7eEg6Y0B
Tech report: https://t.co/yeu6cjpMCT
Tech blog:

### [E0764] 新浪财经7x24 | 2026-07-27 15:17 UTC | 盘中
Kimi AI：发布 Kimi K3的模型权重和技术报告。Kimi K3是我们目前能力最强的模型，这款模型拥有2.8万亿参数的 MoE（混合专家）模型，具备原生视觉理解能力，并支持100万 Token上下文窗口。
URL: https://wap.cj.sina.cn/pc/7x24/5010642
Kimi AI：发布 Kimi K3的模型权重和技术报告。Kimi K3是我们目前能力最强的模型，这款模型拥有2.8万亿参数的 MoE（混合专家）模型，具备原生视觉理解能力，并支持100万 Token上下文窗口。

### [E0783] Kimi (Moonshot AI) | 2026-07-27 15:24 UTC | 盘中
We've open-sourced MoonEP, our high-performance communication library for distributed MoE workloads.

Built to make expert-parallel communication more efficient at scale, MoonEP helps reduce communication overhead in large MoE training and
URL: https://x.com/Kimi_Moonshot/status/2081762544721781124
We've open-sourced MoonEP, our high-performance communication library for distributed MoE workloads.

Built to make expert-parallel communication more efficient at scale, MoonEP helps reduce communication overhead in large MoE training and inference systems.

Explore on GitHub:

### [E0787] Kimi (Moonshot AI) | 2026-07-27 15:25 UTC | 盘中
We've open-sourced FlashKDA, our high-performance CUTLASS-based implementation of Kimi Delta Attention kernels.

It delivers 1.72×–2.22× prefill speedup over the flash-linear-attention baseline on H20, and works as a drop-in backend for fla
URL: https://x.com/Kimi_Moonshot/status/2081762799202746420
We've open-sourced FlashKDA, our high-performance CUTLASS-based implementation of Kimi Delta Attention kernels.

It delivers 1.72×–2.22× prefill speedup over the flash-linear-attention baseline on H20, and works as a drop-in backend for flash-linear-attention.

Explore on GitHub:

### [E0796] Kimi (Moonshot AI) | 2026-07-27 15:25 UTC | 盘中
We've open-sourced AgentENV in collaboration with kvcache-ai.

AgentENV is a distributed system for running agent environments at scale. Its components power agentic RL training for Kimi K3, with fast snapshot, resume, and fork support for
URL: https://x.com/Kimi_Moonshot/status/2081762978391843020
We've open-sourced AgentENV in collaboration with kvcache-ai.

AgentENV is a distributed system for running agent environments at scale. Its components power agentic RL training for Kimi K3, with fast snapshot, resume, and fork support for large-scale parallel agent workflows.

Explore on GitHub:

### [E0809] Wall St Engine | 2026-07-27 15:28 UTC | 盘中
KIMI K3 WEIGHTS AND TECHNICAL REPORT ARE NOW LIVE

Moonshot AI has released the full Kimi K3 weights under its own license. K3 is a native multimodal MoE model with 2.8 trillion total parameters, 104.2 billion active parameters, 93 layers a
URL: https://x.com/wallstengine/status/2081763551577051471
KIMI K3 WEIGHTS AND TECHNICAL REPORT ARE NOW LIVE

Moonshot AI has released the full Kimi K3 weights under its own license. K3 is a native multimodal MoE model with 2.8 trillion total parameters, 104.2 billion active parameters, 93 layers and a 1-million-token context window. Each token is routed through 16 of 896 experts.

The architecture combines Kimi Delta Attention, Attention Residuals and Stable LatentMoE. Moonshot says these changes delivered roughly 2.5x better scaling efficiency than K2 in its internal validation-loss testing. K3 was post-trained using nine specialist policies across coding, general-agent and reasoning tasks at low, high and max effort, before being consolidated into one model.

Moonshot reports K3 leading its comparison set on BrowseComp, ProgramBench, SWE-Marathon and MCPMark, while nearly matching GPT-5.6 Sol on Terminal-Bench at 88.3 versus 88.8. However, the report explicitly says K3 still trails Claude Fable 5 and GPT-5.6 Sol overall. Some benchmark setups also use different agent harnesses, so the comparisons are not fully apples-to-apples.

One notable new detail: during a 48-hour autonomous run, K3 built and verified a small inference-chip prototype using open-source EDA tools. Internal cyber testing also found 16 previously unknown vulnerabilities across six projects, although an independent assessment recorded zero successful arbitrary-code-execution cases across 41 tasks.

Moonshot also released its FlashKDA kernels, MoonEP communication library and AgentENV infrastructure. The report does not disclose total pretraining tokens, total compute or training cost.

### [E0853] 华尔街见闻快讯 | 2026-07-27 15:39 UTC | 盘中
Kimi K3开放日：模型权重、技术报告和关键Infra技术同步开放
URL: https://wallstreetcn.com/livenews/3140420
据月之暗面Kimi公众号，该公司发布Kimi K3的模型权重、技术报告，并开源支撑Kimi K3模型训练的关键Infra技术：MoonEP、FlashKDA和AgentEnv。希望以此加速前沿智能的部署与普及，促进AGI研究。 “Kimi K3是我们能力最强的模型：这是一个拥有2.8万亿参数的混合专家（MoE）模型，具备原生视觉理解能力，并支持100万token的上下文窗口。”

现在，每个人都可以下载并部署Kimi K3模型——无论是用于内部研发，还是嵌入到面向终端用户的产品中，均可自由使用（其他使用情形，详见Kimi K3许可证）。

Kimi K3开放日：模型权重、技术报告和关键Infra技术同步开放

### [E0861] Clément Delangue (Hugging Face CEO) | 2026-07-27 15:41 UTC | 盘中
Kimi K3 has been released on HF and already top 1 trending with 4,000+ likes. Fastest release growth ever so far!!
URL: https://x.com/ClementDelangue/status/2081766926955462998
Kimi K3 has been released on HF and already top 1 trending with 4,000+ likes. Fastest release growth ever so far!!

### [E1551] Vipul Ved Prakash (Together AI CEO) | 2026-07-27 18:23 UTC | 盘中
In partnership with @Kimi_Moonshot, we now have K3 on Together APIs with capacity for hundreds of millions of TPM on launch day!
URL: https://x.com/vipulved/status/2081807708634931271
In partnership with @Kimi_Moonshot, we now have K3 on Together APIs with capacity for hundreds of millions of TPM on launch day!

### [E1601] Bloomberg | 2026-07-27 18:38 UTC | 盘中
What Is Moonshot AI’s Kimi K3 Model and Why Is It Making Waves?
URL: https://www.bloomberg.com/news/articles/2026-07-27/what-is-moonshot-ai-s-kimi-k3-model-and-why-is-it-making-waves
Chinese startup Moonshot AI’s new artificial-intelligence model has drawn global attention since its release for its powerful capabilities. Now, Moonshot has made the technology available for public download, a step expected to expand the company’s influence at a time when there is growing US concern about China’s rapid strides in AI development.

Here’s what to know about Moonshot’s Kimi K3 model and how its public availability could change the industry.

What is Moonshot?

Moonshot was founded in early 2023 by Yang Zhilin, a former Tsinghua University professor who previously worked at Meta Platforms Inc. and Alphabet Inc.’s Google. Before the breakout of DeepSeek, he was the poster boy of Chinese AI. An alumnus of Carnegie Mellon University, Yang has written more than a dozen widely cited research papers on natural language processing. When he set out to launch his own company to challenge OpenAI, venture capitalists rushed to back him. Moonshot AI’s valuation soon reached $20 billion as Kimi became China’s most popular ChatGPT alternative.

Why is the Kimi K3 model getting so much attention?

The Beijing-based company’s Kimi K3 AI model delivers performance on industry benchmarks that rivals top-tier offerings from OpenAI and Anthropic PBC, a surprising breakthrough for a firm that has operated in the shadow of local competitor DeepSeek.

Moonshot’s initial announcement in mid-July about Kimi K3 sent AI and semiconductor stocks tumbling on investor concerns that the rise of China’s AI businesses could force US leaders such as Anthropic to pull back on their investments. That could undercut demand at industry powerhouses Nvidia Corp., SK Hynix Inc. and Samsung Electronics Co.

In late July, the company released the model’s weights, essentially the parameters used to steer AI systems toward answers. That will enable developers to download, tweak and host the technology freely, which in turn should broaden Moonshot AI’s user base. Zhilin has said he wants to win users by focusing on openness and greater availability than competing US proprietary systems.

What else is Yang Zhilin known for?

Moonshot’s founder isn’t just an academic. The 33-year-old is a huge music fan who has infused his company with the names of his favorite bands. Conference rooms at headquarters sport names including Radiohead, Led Zeppelin, Queen and Nirvana. Kimi’s subscription plans are named after the tempo guidelines of classical music: Adagio, Andante and Moderato. The startup was originally called the Dark Side of the Moon in Chinese, after Yang’s favorite Pink Floyd album.

What is Moonshot’s business model?

Moonshot is looking to raise as much as $2 billion in new funding at a valuation of $30 billion, Bloomberg reported in June, putting it among the leaders in China’s AI market. The Kimi developer is in the process of dismantling its offshore structure to pave the way for an initial public offering in Hong Kong, after Beijing tightened its grip on overseas listings.

The company’s annual recurring revenue — a gauge of its forward-looking sales — topped $200 million in April, driven by surging demand for its chatbot and large language models. The company sells tiered subscription plans for the chatbot and offers its underlying technology to enterprise clients.

What threat does Moonshot pose to US rivals?

Moonshot claims that the latest version of Kimi excels at coding tasks. US firms are spending hundreds of billions of dollars on AI infrastructure to run and improve their models — much of it going to chipmakers. Coding is one of the most lucrative offerings by advanced model companies, driving revenue for Anthropic and OpenAI ahead of their planned IPOs. If Moonshot can provide a viable alternative, that could upend the business case for its competitors.

Chinese models have already priced their services aggressively, undercutting US rivals. The weighted average cost of performing a standardized intelligence task using Kimi 2.6 or DeepSeek’s V4 Flash model ranges from 33 cents to just 2 cents, compared with $2.75 for the same task on Anthropic’s Claude Fable 5, according to benchmarking site Artificial Analysis.

Moonshot’s latest offering is priced at a premium by comparison. Kimi K3’s cost on the same ranking is estimated at 95 cents.

Are Chinese AI companies playing fair?

There’s the possibility that Kimi and others in China have already benefited from Anthropic’s work and investment. Earlier this year, Anthropic accused Moonshot, DeepSeek and MiniMax of training their models based on Claude’s capabilities, in a process known as distillation, with “industrial-scale campaigns” that are “in violation of our terms of service and regional access,” according to a February blog post. Moonshot didn’t respond for a request for comment regarding Anthropic’s accusations. Chinese officials have pushed back strongly at the allegations that Chinese firms stole US intellectual property.

Moonshot has positioned itself as an underdog. “Despite being a highly competitive model overall, K3 nonetheless exhibits a noticeable gap in user experience compared with Claude Fable 5 and GPT 5.6 Sol.”

### [E1602] Bloomberg | 2026-07-27 18:38 UTC | 盘中
What Is Moonshot AI’s Kimi K3 Model and Why Is It Making Waves?
URL: https://www.bloomberg.com/news/articles/2026-07-27/what-is-moonshot-ai-s-kimi-k3-model-and-why-is-it-making-waves
Chinese startup Moonshot AI’s new artificial-intelligence model has drawn global attention since its release for its powerful capabilities. Now, Moonshot has made the technology available for public download, a step expected to expand the company’s influence at a time when there is growing US concern about China’s rapid strides in AI development.

Here’s what to know about Moonshot’s Kimi K3 model and how its public availability could change the industry.

What is Moonshot?

Moonshot was founded in early 2023 by Yang Zhilin, a former Tsinghua University professor who previously worked at Meta Platforms Inc. and Alphabet Inc.’s Google. Before the breakout of DeepSeek, he was the poster boy of Chinese AI. An alumnus of Carnegie Mellon University, Yang has written more than a dozen widely cited research papers on natural language processing. When he set out to launch his own company to challenge OpenAI, venture capitalists rushed to back him. Moonshot AI’s valuation soon reached $20 billion as Kimi became China’s most popular ChatGPT alternative.

Why is the Kimi K3 model getting so much attention?

The Beijing-based company’s Kimi K3 AI model delivers performance on industry benchmarks that rivals top-tier offerings from OpenAI and Anthropic PBC, a surprising breakthrough for a firm that has operated in the shadow of local competitor DeepSeek.

Moonshot’s initial announcement in mid-July about Kimi K3 sent AI and semiconductor stocks tumbling on investor concerns that the rise of China’s AI businesses could force US leaders such as Anthropic to pull back on their investments. That could undercut demand at industry powerhouses Nvidia Corp., SK Hynix Inc. and Samsung Electronics Co.

In late July, the company released the model’s weights, essentially the parameters used to steer AI systems toward answers. That will enable developers to download, tweak and host the technology freely, which in turn should broaden Moonshot AI’s user base. Zhilin has said he wants to win users by focusing on openness and greater availability than competing US proprietary systems.

What else is Yang Zhilin known for?

Moonshot’s founder isn’t just an academic. The 33-year-old is a huge music fan who has infused his company with the names of his favorite bands. Conference rooms at headquarters sport names including Radiohead, Led Zeppelin, Queen and Nirvana. Kimi’s subscription plans are named after the tempo guidelines of classical music: Adagio, Andante and Moderato. The startup was originally called the Dark Side of the Moon in Chinese, after Yang’s favorite Pink Floyd album.

What is Moonshot’s business model?

Moonshot is looking to raise as much as $2 billion in new funding at a valuation of $30 billion, Bloomberg reported in June, putting it among the leaders in China’s AI market. The Kimi developer is in the process of dismantling its offshore structure to pave the way for an initial public offering in Hong Kong, after Beijing tightened its grip on overseas listings.

The company’s annual recurring revenue — a gauge of its forward-looking sales — topped $200 million in April, driven by surging demand for its chatbot and large language models. The company sells tiered subscription plans for the chatbot and offers its underlying technology to enterprise clients.

What threat does Moonshot pose to US rivals?

Moonshot claims that the latest version of Kimi excels at coding tasks. US firms are spending hundreds of billions of dollars on AI infrastructure to run and improve their models — much of it going to chipmakers. Coding is one of the most lucrative offerings by advanced model companies, driving revenue for Anthropic and OpenAI ahead of their planned IPOs. If Moonshot can provide a viable alternative, that could upend the business case for its competitors.

Chinese models have already priced their services aggressively, undercutting US rivals. The weighted average cost of performing a standardized intelligence task using Kimi 2.6 or DeepSeek’s V4 Flash model ranges from 33 cents to just 2 cents, compared with $2.75 for the same task on Anthropic’s Claude Fable 5, according to benchmarking site Artificial Analysis.

Moonshot’s latest offering is priced at a premium by comparison. Kimi K3’s cost on the same ranking is estimated at 95 cents.

Are Chinese AI companies playing fair?

There’s the possibility that Kimi and others in China have already benefited from Anthropic’s work and investment. Earlier this year, Anthropic accused Moonshot, DeepSeek and MiniMax of training their models based on Claude’s capabilities, in a process known as distillation, with “industrial-scale campaigns” that are “in violation of our terms of service and regional access,” according to a February blog post. Moonshot didn’t respond for a request for comment regarding Anthropic’s accusations. Chinese officials have pushed back strongly at the allegations that Chinese firms stole US intellectual property.

Moonshot has positioned itself as an underdog. “Despite being a highly competitive model overall, K3 nonetheless exhibits a noticeable gap in user experience compared with Claude Fable 5 and GPT 5.6 Sol.”

### [E1661] Bloomberg | 2026-07-27 19:00 UTC | 盘中
Kimi Panic Is a Sputnik Moment for ‘Open’ AI
URL: https://www.bloomberg.com/opinion/articles/2026-07-27/kimi-panic-is-a-sputnik-moment-for-open-ai
The release of Moonshot’s Kimi K3 has spurred plenty of noise in Washington, marked by threats of a crackdown and accusations of stealing. But the signal from Silicon Valley says something different: Open models are becoming essential to AI leadership, and the US is falling behind.

First came the panic. Treasury Secretary Scott Bessent declared sanctions and Entity List designations were “on the table,” while accusing Chinese firms of intellectual property theft via distillation, a standard technique in which one model’s outputs are used to improve another. White House Office of Science and Technology Policy Director Michael Kratsios echoed these sentiments. Soon, reports began swirling that the US was considering a crackdown on Chinese AI models themselves.

Then something kind of incredible happened: Silicon Valley revolted. A coalition of 179 startups and tech founders signed a letter addressed to Kratsios and Commerce Secretary Howard Lutnick, urging against a ban of foreign models. Recent Chinese releases such as Kimi K3 show “open distribution is now a competitive strategy,” the letter stated. US leadership requires not just better American open models, but continued access to those already available worldwide, it added. Often described as open source (though more precisely called open weight), these AI models make their inner workings available publicly, allowing developers to download, run and fine-tune them.

Days later, Nvidia Corp. founder and Chief Executive Officer Jensen Huang joined X and used his first-ever post to share a separate letter signed by heavyweights including Microsoft Corp., Alphabet Inc.’s Google, and even OpenAI. It similarly defended open models, but went even further, challenging Washington’s characterization of theft. Policymakers should not “conflate legitimate model-development techniques with misappropriation,” it stated. Distillation “is a widely used technique for model improvement, evaluation, and validation.”

It’s a big day for those of us who have been saying this even when it wasn’t cool. The industry’s response marks a significant shift. Advocates of open AI for years have argued that US leadership cannot rest solely on closed systems controlled by just a handful of labs. Now much of Silicon Valley is making the same case — and loudly enough that this once-wonky debate has gone mainstream, with even 24-hour diner Denny’s weighing-in.

So how will this change the business dynamics of the lucrative AI boom? For many firms, the sudden support isn’t exactly altruism for the cause of shared knowledge. Open weight AI doesn’t mean that revenue will be eroded, but it will likely be redistributed. It’s already been happening, bringing more capital flowing toward companies selling the infrastructure for AI to run on, including chipmakers, cloud service providers and application developers.

Nvidia would be an obvious winner; the more developers can download and deploy open models, the greater demand for hardware to operate them. Amazon.com Inc., Microsoft and Google also stand to gain by hosting AI systems on their cloud platforms. Expanding support for Chinese models could be a key way for American firms to capture revenue from technology developed abroad.

The change does, however, leave companies like Anthropic PBC (the most glaring outlier in Silicon Valley to have not signed onto either of the letters) and OpenAI exposed. They’ve been vulnerable because the real AI race may be less about who can build a God-like computer first than who can diffuse intelligence across the economy fastest.

Washington would be wise to listen to its own industry, but it’s still very possible that the Trump administration may impose restrictions. A reflexive crackdown, however, would backfire. Banning Chinese models in the US would not stop global developers from using them, putting American researchers at an inherent disadvantage. Not to mention undoing delicate and necessary progress in AI safety discussions with Beijing.

Nor would doing so mitigate the cybersecurity threats open systems might pose; everyone else in the world could still access them. And recent events have also complicated the claim they are more risky regarding cybersecurity. The high-profile incident last week of an unreleased OpenAI agent going rogue during an evaluation, compromising systems at model repository Hugging Face garnered global headlines over cyber risks. Yet Hugging Face said in a blogpost that it used GLM 5.2, developed by Beijing-based Zhipu, as part of its defensive response. The control these tools offer can be dual-use for defenders as well as attackers.

Kratsios also claimed Moonshot violated US export controls by using Nvidia chips to develop Kimi K3. The two firms haven’t publicly addressed the allegations; it remains unclear whether it’s accurate. But even if it is, the episode would further expose how porous the current policy is. At a time when Beijing has reacted to chip restrictions by rejecting Nvidia and attempting an all-out effort to create homegrown alternatives, Washington’s policy responses have still focused on holding China back versus keeping America ahead. This strategy may buy time, but in the long run has incentivized these Chinese investments in domestic options.

Washington may be attempting to contain China’s latest AI breakthrough the best ways it knows how. But the panic has forced Silicon Valley to finally admit that open models are a strategic asset. If that ends up actually producing a competitive, open ecosystem in the US, Kimi will have strengthened American AI more than any attempts to suppress it ever could.

More From Bloomberg Opinion:

OpenAI's Sandbox Breach Shows Containment Risks: Parmy Olson Kimi K3 May Prove to Be a Gift to America's AI Firms: Dave Lee Humanoids Have Bodies. They Still Need Brains: Catherine Thorbecke

Want more Bloomberg Opinion? OPIN <GO>. Or subscribe to our daily newsletter.

### [E0762] Nathan Lambert | 2026-07-27 15:17 UTC | 盘中
Kimi K3 license. It's inspired by MIT but distinctly non-commercial, where any company making over $20M/yr must get a specific commercial deal (and display Kimi K3 if over 100M users or $20M/mo revenue)
URL: https://x.com/natolambert/status/2081760901020201086
Kimi K3 license. It's inspired by MIT but distinctly non-commercial, where any company making over $20M/yr must get a specific commercial deal (and display Kimi K3 if over 100M users or $20M/mo revenue)

### [E1723] Tae Kim | 2026-07-27 19:17 UTC | 盘中
lol. Kimi isn't really free open source. There goes that bear narrative.

"The $20 million trailing 12 months revenue threshold is comically low. Everyone meets it and will have to negotiate a commercial, revenue-sharing contract with Kimi.
URL: https://x.com/firstadopter/status/2081821265136865674
lol. Kimi isn't really free open source. There goes that bear narrative.

"The $20 million trailing 12 months revenue threshold is comically low. Everyone meets it and will have to negotiate a commercial, revenue-sharing contract with Kimi."

### [E0845] The Information | 2026-07-27 15:37 UTC | 盘中
China’s Ministry of Commerce on Monday struck back against Washington’s recent accusations that Chinese AI firms are stealing U.S. intellectual property by distilling American AI models, saying the claim has no factual basis or legal suppor
URL: https://www.theinformation.com/briefings/beijing-strikes-back-washingtons-ai-distillation-accusation
China’s Ministry of Commerce on Monday struck back against Washington’s recent accusations that Chinese AI firms are stealing U.S. intellectual property by distilling American AI models, saying the claim has no factual basis or legal support.

“China has always opposed the US politicizing and weaponizing science, technology and trade issues,” the ministry’s spokesperson said. The ministry also criticized the U.S. accusations and potential sanctions against Chinese AI firms as “typical acts of AI hegemony.”

Some Trump administration officials last week accused Chinese startup Moonshot AI of using U.S. frontier models to train the firm’s recently released model, Kimi K3, in a practice known as distillation. The Information [reported](https://www.theinformation.com/articles/u-s-investigates-chinese-ai-companies-access-chips-amid-moonshot-accusations?rc=jnr9wn) last week that the Bureau of Industry and Security, an agency within the U.S. Department of Commerce, is formally investigating whether Chinese firms like Moonshot are accessing advanced Nvidia chips that are subject to U.S. export controls.

Kimi K3, released on July 16, has disrupted Silicon Valley by delivering capabilities that rival or even surpass top U.S. models on some benchmarks.

The ministry rejected the distillation claim, saying the U.S. ignores the fact that the release time of the Chinese AI model was very close to that of the top U.S. model, leaving little time for distillation. “Some of China’s model capabilities are already in a leading position,” the spokesperson said.

“For any act that substantially harms China’s interests, China will take all necessary measures to firmly safeguard its legitimate rights and interests,” the spokesperson said.

## 研究交付要求

这是一条已经从整场材料语义拆出的独立催化，不要重新把它压回大盘总评。核实美股异动原因和传播时序；用结构化供应链、内部 KB 与公司披露坐实每个亚洲候选；逐腿查真实行情和本地市场 price-in；候选不成立也要逐条给出有证据的 AVOID。