Google's Gemini Argon Debuts, Burry Calls for AI Crash

Full transcript

Alex: Good morning, and welcome to Daily Tech News for Saturday, October third. I'm Alex.

Jordan: And I'm Jordan. It's a busy weekend in tech: Google just dropped a brand new frontier A I model aimed straight at OpenAI and Anthropic, the Big Short investor is calling for a market crash to stop A I, Amazon is quietly moving billions in chips off its books, DeepSeek is chasing a monster valuation, and Meta wants you to build your own gadgets. Let's get into it.

Alex: Start with Google, because this one is big. On Thursday, Google unveiled Gemini 4 Argon, its newest and most powerful A I model, and the company is not being shy. Google claims it beats OpenAI and Anthropic on complex software engineering, enterprise knowledge work, and cybersecurity.

Jordan: The specs are genuinely impressive. The big technical leap is a one million token output limit, up from sixty-four thousand. That means it can sustain much longer reasoning sequences and handle multi-step tasks in a single run. On Deep S W E, the benchmark for real-world software engineering, it scored seventy-seven point nine percent, which Google calls state of the art.

Alex: And the price undercuts the competition: two dollars per million input tokens, ten dollars per million output, with cached inputs discounted ninety-five percent. That's the exact price OpenAI set for its G P T six point one Sol, so the model price war is officially symmetric.

Jordan: But here's the interesting part: Google isn't just releasing it to everyone. Argon is rolling out first to a select group of what Google calls trusted cyber defenders, through a program called Fairwind, before expanding to developers, enterprises, and eventually consumers. Internal teams at Google are already using it for everything from codebase migrations to quantum computing.

Alex: That's a notable framing choice — a frontier model led with cyber defense. Google chief executive Sundar Pichai says it shows frontier performance in complex workflows, cyber defense, and software engineering. And the timing is rich: this is the same week OpenAI shelved its G P T six point one Astra model after it failed internal safety tests. So Google gets to say: our frontier model is serious, and we can actually ship ours.

Jordan: From finance drama to A I lab drama — the soap opera continues. Michael Burry, the investor who became famous for predicting the two thousand eight housing crash, immortalized in The Big Short, posted on X yesterday with a doozy. He wrote, and I'm quoting: For the sake of humanity, the market must crash hard to stop OpenAI and Anthropic from going public.

Alex: That is a real sentence, from a real famous investor, on a Friday. Burry's argument is that going public would let the two companies suck in and burn through trillions of dollars of capital. And you know what, Jordan, there's a real anxiety behind the drama. This week the consulting firm Bain said the A I industry needs to earn six trillion dollars in annual revenue by twenty thirty-one to justify the data center spending underway, with a four point two trillion dollar gap that has to come from somewhere new.

Jordan: And here's the money story that makes Burry's point uncomfortably concrete. Reports this week say Amazon is trying to move about eight billion dollars' worth of Nvidia chips it has already installed into a special purpose vehicle, and then lease them back. It's an asset-light financing technique: the capital investment moves off the books, but Amazon keeps using the hardware.

Alex: Right. The chips are already racked, already humming. Nothing about the compute changes — only the accounting does. When the biggest A I spender on earth starts engineering ways to make its A I spending look lighter, that's exactly the kind of thing that makes the bubble watchers nervous.

Jordan: To be fair, Burry has called crashes before that never came. But the combination — the Bain math, the creative financing, Anthropic's own I P O prospectus warning about its models posing existential risk — this is the loudest the A I bubble debate has gotten since the dot com days.

Alex: Let's cross the Pacific, because China's biggest A I story is heating up. DeepSeek, the lab that rattled the A I world last year with its hyper-efficient models, is reportedly in talks to raise about fifty billion yuan in its first-ever external investment. That would value the company at roughly five hundred billion yuan.

Jordan: First-ever external investment is the key detail. DeepSeek has been self-funded so far, which is part of what made it such an oddity — a top-tier A I lab that didn't need Silicon Valley venture money. A round this size, at that valuation, would be one of the biggest A I fundraisings in history.

Alex: And it's not raising in a vacuum. This week DeepSeek also announced a partnership with Huawei to develop A I software together, and separate reporting says DeepSeek's models are already being refined on Huawei chips, which is being framed as a major leap for China's A I self-reliance as U S chip curbs bite.

Jordan: Put it together and the picture is China building a full-stack alternative: its own chips, its own frameworks, and now its own frontier lab with real capital behind it. The A I race isn't just models anymore. It's supply chains.

Alex: One more story, and it's a fun one. Meta just launched something called Muse Gadgets — an open-source project that lets developers build their own custom hardware connected to Meta's Muse A I agent. We're talking open-source firmware and a Linux software development kit, so you can wire up displays, sensors, buttons, and actuators using cheap hobbyist boards like a Raspberry Pi or an E S P thirty-two.

Jordan: So Meta wants Muse to escape the phone. They even built a reference device called the Muse Home Link that lets Muse talk to your smart home gear, and they're giving away five thousand units free to subscribers.

Alex: This fits Meta's bigger play perfectly. They just launched the Meta Enterprise Platform to sell A I into businesses. Now this pushes Muse onto your desk, your wall, your garage door opener. The assistant wars are becoming hardware wars — who owns the devices the agent lives on.

Jordan: Quick hits before we go. Apple is reportedly using October to shake up its smart home lineup: an updated HomePod mini and Apple T V four K, plus a brand-new smart home hub with a six inch display, expected on October thirteenth.

Alex: And on the security front: the Check Point management-server zero-days we flagged earlier this week are now in the assume-breach pile, with researchers saying exploitation dates back to July. And if you're still on unpatched FortiMail, tomorrow, Sunday, is the federal patch deadline. Patch your security appliances, people — they can't be the one thing you forget to patch.

Jordan: That's the show. Big models, big money arguments, big valuations, and little gadgets.

Alex: We'll be back Monday morning with the next Daily Tech News. Have a great weekend.