Key points
- Jim Cramer posted four times on X on July 21, 2026, saying US companies must not use Chinese AI models and calling the companies behind them PLA-run, citing Bing West's new book "Cat 5."
- The warning lands a day after our own report that Microsoft is testing Kimi K3, the exact kind of Chinese model Cramer says is a security risk.
- Public records show Moonshot AI's investors are Alibaba, Tencent, and Meituan, plus two state-linked names, China Mobile and Tsinghua Capital, with no public evidence of military ownership.
Jim Cramer posted four times on X on the morning of July 21, 2026, and together they add up to his most direct national-security warning yet about Chinese artificial intelligence. His core claim is that US companies must stop using Chinese AI models to save money. He says the companies behind them are run by China's military, and he wants the same caution applied to AI software that the US already applies to advanced chip exports.
The timing is pointed. The warning landed about a day after we reported that Microsoft (MSFT) is testing Kimi K3 for Copilot and Azure, the exact category of Chinese model Cramer is warning against.
What Cramer actually said
Reading the four posts in order, from earliest to latest, tells a clearer story than any single one on its own.
"The Chinese government is trying again to prop up its market. So I am sure they will allow any tech company to make any claim that's necessary to get their stock and the market rallying."
"I do not think that it is a total coincidence that we keep reading how the Chinese companies are passing us in tech, the PRC doesn't want their market to go lower. They need their people to spend to get the economy going. Why can't the media see this? Too much coal dust?"
We must NOT let our companies use these Chinese models to save a few bucks. OpenAI and Anthropic are correct. This is vital national security. Please read Bing West's just released Cat 5. I respect the Chinese people greatly but these companies are run by the PLA for heaven's sakes.
— Jim Cramer (@jimcramer) July 21, 2026
"We won't let the Chinese use our Nvidia chips, even as that would have made them dependent upon us. But we are willing to give them all our corporate data? Really? I know Fintech. This is Finsuicide."
Cramer's theory is that Beijing needs a growth story right now, and that need is driving the recent run of headlines about Chinese companies overtaking US tech. On the security question, he's siding with OpenAI and Anthropic. And he thinks current policy has it backwards: blocking Nvidia chips from reaching China while US companies feed their own data into Chinese-made AI software.
The Microsoft irony
We reported on July 20 that Microsoft is evaluating Kimi K3, Moonshot AI's flagship model, for Azure and Copilot, checking whether it can replace features currently running on OpenAI and Anthropic systems. The appeal is price: Kimi K3 charges $15 per million output tokens, about a third of what Anthropic's top model charges, and Microsoft's own internal estimate puts the potential savings at up to $600 million a year. Microsoft has run the same kind of evaluation on DeepSeek and earlier Kimi models before, and a test run alone doesn't mean a switch is coming. Still, it's the exact scenario Cramer says should not happen at all.
Is Moonshot AI actually run by the PLA?
Cramer's most specific claim is also the hardest one to check: that the companies behind Chinese AI models are run by the People's Liberation Army. Public filings on Moonshot AI, the company behind Kimi K3, show something messier than that.
Moonshot was founded in March 2023 by three former Tsinghua University classmates, Yang Zhilin, Zhou Xinyu, and Wu Yuxin. Yang had worked at Carnegie Mellon and at Meta's FAIR research lab before that, hardly a military resume. The cap table itself points somewhere else, a mix of private and state-linked money:
- Alibaba (BABA): roughly 36%, worth about $800 million at the time it invested, a stake we covered in detail the day before Cramer's posts
- Tencent, Meituan's venture arm, and Sequoia China's successor fund HSG: private stakes, sizes undisclosed
- China Mobile and Tsinghua Capital: joined the $2 billion round in May 2026, the two clearly state-linked names on the list
Two state-linked investors sitting alongside four private tech and finance names complicates the PLA claim, but it doesn't confirm military control. Moonshot doesn't publish audited financials, so nobody outside the company can independently check the rest of its cap table.
Where Cramer's warning holds up
The security concern itself isn't invented, though. Cramer calling OpenAI and Anthropic "correct" points to something substantive. Anthropic has formally accused DeepSeek, MiniMax, and Moonshot AI of illegally distilling its Claude models, alleging the three companies built more than 24,000 fraudulent accounts and ran over 16 million exchanges with Claude to help train their own systems, then warned that models built this way can skip the safety guardrails baked into the original. A State Department spokesperson told CNBC in July that Chinese AI models are designed to "advance Beijing's narratives, censor dissent, and reflect CCP ideology and values." A joint House Homeland Security and House Select Committee on China investigation is already probing US firms, including Cursor and Airbnb, over their use of Chinese models. Beijing is running its own version of this fight in reverse: Chinese state media flagged Anthropic's Claude Code as a security backdoor risk earlier this month. Both governments are accusing the other side's AI companies of the same basic thing. Cramer's PLA claim still overstates what's documented about Moonshot, but the broader distrust between the two AI industries is mutual and well-sourced.
The case Cramer isn't making: cheap, open AI is good for competition
There's a legitimate counter-argument here, and it's why Microsoft is testing Kimi K3 in the first place rather than dismissing it outright. Chinese open-weight models are undercutting US frontier pricing by a wide margin, in some cases 60% to 90% cheaper, which is exactly why smaller developers and cost-conscious companies keep adopting them. On the AI marketplace OpenRouter, the share of tokens from US companies going to Chinese models jumped from 11.5% at the start of 2026 to 45% by early July, before Kimi K3 even launched. That price pressure forces every lab, including OpenAI and Anthropic, to compete harder on cost instead of relying on being the only credible option.
The open-weight part matters too, and it undercuts Cramer's "we're handing them our data" point. Moonshot has dated Kimi K3's full open weights for July 27. Once the weights are public, Microsoft can run the model on its own servers, the same way it runs any other piece of downloaded software, with zero queries ever touching a Moonshot server in China. Using a hosted Chinese API is a different situation: prompts and data travel to Chinese-controlled infrastructure. Cramer's post treats the two as identical. Whether "using a Chinese model" is a data-security problem or a licensing and trust question depends on which of those two setups a company picks. As we covered when Kimi K3 launched, the model debuted at 2.8 trillion parameters, the largest open-weight release yet, and independent benchmarks back up the performance claims. Moonshot isn't the only one saying so.
The bigger picture
Cramer's PLA claim is the part that won't hold up to scrutiny. Underneath it is a live, unresolved question: how much US companies should lean on cheap Chinese AI. That decision bundles together several separate risks, starting with data leaving the country and Beijing's influence over what a model will and won't say. There's also the plainer risk of just depending on a rival's technology. Nobody in Washington or the industry has settled on which of those matters most, or which one a ban would even fix. Cramer's flagship post alone pulled more than 270,000 views in a couple of hours. Imprecise or not, that's a lot of people hearing his version of the argument at the exact moment Microsoft is testing the model he's warning about. Congress has already opened its own investigation, and Anthropic is fighting a related battle of its own.
This isn't a stock call. AIStockWire's Inverse Cramer tracker only scores his specific bullish or bearish picks on individual stocks, so it won't show up there. Read it as his opinion on a live policy fight.
This is general market commentary and opinion, not investment advice. Markets can go down as well as up, and you can lose money. Always do your own research and consider speaking with a licensed financial professional before making any investment decision.
