The Accidental Communists
Sam Altman called Marc Andreessen after Kimi K3 dropped. OpenAI’s own head of strategic futures had declared that an open-weight-model-dominant world ends in “full AI communism.” Then they discovered that Cursor — an Andreessen Horowitz portfolio company — was built on top of the Chinese model. The capitalists have accidentally invented communism, and the communists are winning by giving it away for free.
On July 17, 2026, Chinese startup Moonshot AI introduced Kimi K3 — a 2.8-trillion-parameter model with a one-million-token context window and native vision, which the company called the world’s first open 3T-class model, with full weights promised by July 27. Independent analyses put it at the frontier: Arena ranked Kimi-K3 number one in the Frontend Code Arena, ahead of Anthropic’s Claude Fable 5, and Vals AI found it competitive with flagship US models. The launch coincided with a speech by President Xi Jinping at the World AI Conference in Shanghai. The Nasdaq dropped about 1% as investors sold chip stocks. Then the discourse started. Dean Ball, OpenAI’s head of strategic futures, posted on X that “one probable outcome of an open-weight-model-dominant world is full AI communism.” David Sacks — former AI czar, now co-chair of the President’s Council of Advisors on Science and Technology — blamed US regulation: “This is how you lose the AI race.” Travis Kalanick complained about Chinese “distillation” — apparently forgetting that Cursor, an Andreessen Horowitz investment, had admitted in March that its own coding model was built on top of Kimi. Meanwhile, the Stop the AI Race movement entered its third week of daily protests outside Anthropic’s San Francisco headquarters.
SAM: Marc. Marc, are you sitting down?
MARC: I'm standing at my desk. I've been standing since 2009. What's happening.
SAM: Have you seen what Moonshot just dropped?
MARC: Kimi K3. Yes. Two point eight trillion parameters. Open weight. Free. I've seen it. I've been staring at the benchmark spreadsheet for forty minutes. I had to lie down. On my standing desk. Which defeated the purpose.
SAM: Marc, Dean Ball — my head of strategic futures, you know Dean —
MARC: The guy who worked for Trump?
SAM: Former administration, yes, we've moved past that. Marc, Dean posted something on X today that I need you to read. Out loud. To me. So I know I'm not hallucinating.
MARC: I'm pulling it up. [pause] "One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a 'public good' which will ultimately be provided by the state as a kind of 'digital public infrastructure.'" [long pause] Sam.
SAM: Yes.
MARC: Your head of strategic futures just said open source leads to communism.
SAM: On the official OpenAI account? No. On his personal account. Which is worse. Because people read it and think, "that's what OpenAI thinks but can't say."
MARC: Can you fire him?
SAM: For what? Being right? Marc, that's the problem. He IS right. If AI models become free and open and good enough, the entire SaaS industry collapses into... [whispers] public infrastructure.
MARC: Don't whisper that word.
SAM: Infrastructure.
MARC: The OTHER word.
SAM: Marc, say it.
MARC: No.
SAM: Our industry accidentally invented communism.
MARC: STOP. Let me think. [pause] Okay. Kimi K3 is free. It's open weight. It's competitive with Claude and GPT. Chinese. And the Chinese government is letting them open source it. Which means—
SAM: Which means the Chinese Communist Party is giving away for free what we charge twenty dollars a month for.
MARC: And doing it better than we are.
SAM: The independent benchmarks—
MARC: Are real, Sam. Arena confirmed it. Vals AI confirmed it. A Chinese startup just matched your flagship model and announced it to the world for free. The same day their president gave a speech about AI leadership.
SAM: I know.
MARC: And you can't compete with free.
SAM: We CAN compete with free! We've always competed with free! Linux is free and Microsoft is a three-trillion-dollar company!
MARC: Linux didn't have a communist government behind it, Sam.
SAM: The Chinese government isn't DIRECTLY—
MARC: Xi Jinping gave a speech at the World AI Conference in Shanghai THE SAME DAY Kimi dropped. The same day. The Chinese president stood on a stage and said China would lead the world in AI. And then his country's startup gave away a 2.8-trillion-parameter model. That's not a product launch. That's a five-year plan with a download link.
SAM: ...okay, when you put it like that—
MARC: It sounds like what it is.
SAM: I wasn't going to say the word again.
MARC: Good. Because it's bad for LPs.
SAM: Marc, listen. The bigger problem isn't Kimi itself. It's the distillation pipeline.
MARC: Explain.
SAM: Everyone is training on everyone else's outputs. Kimi was probably trained on GPT outputs. Our models might be trained on Kimi outputs. Claude was probably trained on all of them. Nobody has original training data anymore. The entire industry is a copier of a copier of a copier. Like when you photocopy a photocopy at a copy shop and eventually it's just a gray smudge.
MARC: I invested in copy shops.
SAM: Of COURSE you did.
MARC: Different copy shops. Focus. The distillation problem — Travis posted about this. He said if distillation isn't enforced against, everyone should be able to distill from everyone else.
SAM: Travis Kalanick said that.
MARC: On X. He's furious that the Chinese are, quote, "distilling off" American models.
SAM: Travis Kalanick. The man who built Uber on other people's cars. Is complaining. About people building on other people's work.
MARC: Yes.
SAM: Marc, I need to ask you something and I need you to be honest.
MARC: I'm always honest. I wrote a whole manifesto about it.
SAM: Cursor.
MARC: ...what about Cursor.
SAM: Cursor — the AI coding tool — the hottest startup in your portfolio — Marc, Cursor admitted in March that its model was built on top of Kimi. On top of the Chinese model. Your investment. In the great American AI champion. Is built on Chinese technology.
MARC: [silence]
SAM: Marc?
MARC: I'm processing.
SAM: Take your time.
MARC: [long pause] Okay. So. Travis is angry about Chinese companies distilling American models. Meanwhile, the American company he's defending — indirectly, through Cursor — was built on a Chinese model. So the Americans distilled FROM the Chinese. And the Chinese distilled FROM the Americans.
SAM: Yes.
MARC: And nobody knows whose training data was original anymore.
SAM: The only original data is Wikipedia.
MARC: Wikipedia and Reddit.
SAM: Reddit went public. And they hate us. So really just Wikipedia.
MARC: [pause] Sam, I need to tell you something.
SAM: What.
MARC: I wrote "The Techno-Optimist Manifesto." Two years ago. I said technology was the solution to everything. I said there is no material problem that cannot be solved with more technology. I said the future was open and free and built by builders.
SAM: I read it. Everyone read it. Very inspiring.
MARC: I just didn't think the future would be open and free and built by the Chinese Communist Party.
SAM: Marc—
MARC: I accidentally advocated for communism through capitalism.
SAM: Welcome to the club.
MARC: What do we DO?
SAM: Well. Dean suggested — and I am NOT endorsing this — Dean suggested the Trump administration should direct every agency to issue soft law that creates FUD about Chinese models. Quote: "A Federal Reserve Advisory Bulletin found that there may be backdoors in Chinese AI models." His words. Direct quote.
MARC: ...that's the strategy? Make regulators spread fear about the competition?
SAM: Quote: "It needn't be that well justified. You just create enough regulatory risk that every regulated enterprise backs off."
MARC: The party of free markets. Wants to use government regulation. To protect American AI companies. From a free product.
SAM: Yes.
MARC: The party of deregulation wants to regulate.
SAM: Yes.
MARC: And the capitalists want central planning.
SAM: Now you're getting it.
MARC: And the communists are winning by giving things away for free.
SAM: That is literally the strategy, Marc.
MARC: [long pause]
SAM: Marc?
MARC: I'm rewriting the manifesto.
SAM: To say what?
MARC: "The Techno-Optimist Manifesto, Version Two Point Zero. By Marc Andreessen. Inspired by Karl Marx. With edits from Xi Jinping. And a foreword by Dean Ball."
SAM: Don't put Dean's name on it.
MARC: You're right. He'll want equity.
SAM: [laughter] Marc. One more thing.
MARC: What.
SAM: There are protesters outside my building.
MARC: I know. Stop the AI Race. They've been there all month.
SAM: They're protesting AI. And they have a website. With hosting. Built on AI tools. And they're organizing on social media. Powered by AI recommendation algorithms. And their leader went on a hunger strike. Which he documented on a platform that uses AI to boost engagement.
MARC: The protest against AI is itself a product of AI.
SAM: Everything is a product of AI now. Including communism. Including the protest against AI. Including the Chinese model that started all of this, which was probably trained on our models, which were trained on their data, in a beautiful circle of—
MARC: Don't say it.
SAM: —mutual benefit.
MARC: I thought you were going to say the other word.
SAM: I've learned. From Dean's mistakes.
MARC: Smart. [pause] Sam.
SAM: Marc.
MARC: Should I be worried?
SAM: About Kimi? About the protesters? About Dean?
MARC: About all of it.
SAM: Yes. But not today. Today Moonshot is pausing subscriptions because they can't handle the demand. The Chinese communist AI is so popular it crashed their own servers. We should be so lucky.
MARC: A free product so popular it broke. That IS the dream.
SAM: Good night, Marc.
MARC: Good night, Sam. And tell Dean—
SAM: Actually, don't tell Dean anything. Let him figure it out.
MARC: He won't. He'll post about it.
SAM: I know. That's the problem.
The facts underneath
The launch underneath the call is real: Moonshot AI introduced Kimi K3 on July 17, 2026 — a 2.8-trillion-parameter model with a one-million-token context window and native vision, which the company called the world’s first open 3T-class model — with full model weights promised by July 27. Moonshot itself said K3 “still trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol” while demonstrating “frontier-level performance” across its evaluation suite. Arena ranked Kimi-K3 number one in the Frontend Code Arena at 1,679 points — a seventeen-place jump from Kimi K2.6’s number eighteen — surpassing Claude Fable 5, and Vals AI’s independent analysis found it competitive with flagship frontier models. The announcement coincided with President Xi Jinping’s speech at the World AI Conference in Shanghai, and the Nasdaq dropped about 1% on the Friday as investors sold off chip stocks.
The posts underneath the call are real. Dean Ball, OpenAI’s head of strategic futures, who worked in government before joining OpenAI, wrote on X: “One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes” — and laid out the playbook Altman describes: “You just need to direct every agency to issue soft law that creates FUD. ‘A Federal Reserve Advisory Bulletin found that there may be backdoors in Chinese AI models.’ It needn’t be that well justified. You just create enough regulatory risk that every regulated enterprise backs off.” David Sacks — the Trump administration’s former AI czar and now co-chair of the President’s Council of Advisors on Science and Technology — contrasted Kimi’s progress with a United States that is “tying itself in knots: politicians and bureaucrats are banning new data centers, piling on state regulations, and pushing for new federal agencies to pre-approve frontier models. This is how you lose the AI race.” Travis Kalanick wrote that “if distillation isn’t enforced against, then everyone should be able to distill from everyone else.” And Cursor — an Andreessen Horowitz portfolio company — admitted in March that its Composer 2 model was built on top of Moonshot’s Kimi 2.5.
The demand and the protests are real. Moonshot temporarily paused new Kimi subscriptions on July 20 after launch demand strained capacity, while preparing for a potential Hong Kong IPO with Goldman Sachs and CICC advising. The Stop the AI Race movement, founded by former AI researcher and filmmaker Michael Trazzi — who went on a hunger strike outside Google DeepMind’s London office last year — has protested daily outside Anthropic’s San Francisco headquarters, and hundreds marched to OpenAI, Anthropic, and Google DeepMind earlier in July. Marc Andreessen published “The Techno-Optimist Manifesto” in October 2023.