Podcast 5-5 by Dongee · November 13, 2026

Transcript

Gabriel: Today on the Dongee podcast: Jensen Huang's resignation at Nvidia, the Microsoft Outlook and Teams outage, and Gemini 4 Hydrogen, Google's new model.

Gabriel: Friday, November thirteenth. Hey, Lina. The Nvidia story: Jensen Huang resigned.

Lina: Did he resign… or was he pushed out?

Gabriel: The statement says he resigned, effective immediately. And he's giving up everything: CEO, chairman and board member, after thirty-three years running the company he founded.

Lina: After a week of scandal.

Gabriel: It's over his relationships with a woman flagged as a suspected Chinese spy and with an executive who reported to him. The board met Wednesday night, and the announcement came out yesterday.

Lina: A company like that can't go without someone in charge for even a day.

Gabriel: Colette Kress, the chief financial officer, is stepping in as interim CEO…

Lina: And she's still handling the money too?

Gabriel: She's doing both. And Harvey Jones, who used to run Synopsys, a chip software company, is the first board chair in Nvidia's history.

Lina: The first?

Gabriel: Huang was CEO and chairman at the same time. Now those jobs are being split. And the board is going to look for a permanent CEO, from inside or outside the company.

Lina: And in the whole statement, not one word about Huang. Not even a "thank you."

Gabriel: The stock dropped as much as sixteen percent during the day… and closed down a little over four percent.

Lina: Wall Street freaked out early and had calmed down by closing time.

Gabriel: And inside the company, according to Wired, a tech magazine, employees on an internal forum are asking for him to come back. Elon Musk even said on X that hiring him "would be an honor."

Lina: And according to Bloomberg, a Republican congressman says Huang agreed to testify before the United States Congress.

Gabriel: If you look at it from the oversight side, the board did what it had to do: those relationships raise questions about keeping the person in charge in check, and about national security. And splitting the roles helps.

Lina: If you look at it from the inside, they pushed out the founder who took Nvidia to the top, before the investigation is over, right in the middle of the AI race. That could cost them too.

Lina: Gabriel, if I emailed your work account yesterday and it never showed up… it wasn't personal.

Gabriel: Outlook?

Lina: Outlook, Teams, Word, Excel, Copilot… Microsoft 365 went down worldwide starting Thursday.

Gabriel: I thought it was my internet. I restarted the modem twice.

Lina: Half the planet thought the same thing. According to The Guardian, Downdetector, a website that tracks outages, got nearly one hundred thirty thousand reports in seventy-four countries.

Gabriel: Yikes. So what broke?

Lina: According to Microsoft, several servers became "unexpectedly unavailable." And the rest had to pick up all the work…

Gabriel: …and they got tired too.

Lina: Overloaded, more like. And the weird part is the fixes didn't hold: traffic came back and the servers went down again.

Gabriel: Wait. So is everything back now?

Lina: Not completely. It was still failing early Friday morning, especially email, and the problem spread to Asia. More than twelve hours. At the time of recording, Microsoft was still working to get things back to normal.

Gabriel: With that much chaos, your first thought is hackers.

Lina: Microsoft says there's no sign of an attack. What it did was pause changes, reroute traffic and look at undoing some recent configuration changes.

Gabriel: And it wasn't just people at home: companies, some government offices…

Lina: …and even Microsoft connections that use OpenAI's and Anthropic's services.

Gabriel: Yesterday was the day with the fewest work emails in history. Somebody must've celebrated.

Lina: Some people see it as a wake-up call: relying on just one provider for email and work is risky. One expert, Mark Gregory, recommends having a backup option and testing changes more carefully.

Gabriel: And some people are saying stay calm: outages like this are rare compared to millions of hours without problems, Microsoft kept everyone updated the whole time, and it wasn't an attack.

Gabriel: Switching gears. Google launched Gemini 4 Hydrogen, an AI model built for quick, everyday work.

Lina: Another model. I've already lost track of the names.

Gabriel: You're going to run into this one no matter what: it's coming to the Gemini app and to Google Search, for everyone. Free users get it automatically.

Lina: So a lot of people are going to use it without even knowing.

Gabriel: It's made for coding, multi-step tasks and working as an agent, meaning an AI that actually does things for you, not just answers you.

Lina: Uh-huh. And it breaks records, I'm guessing.

Gabriel: According to Google, it beats its own big model, Argon, on an independent comparison chart…

Lina: …by a hair, I bet.

Gabriel: By one point. Fifty-one to fifty.

Lina: One point. In school, that's called barely passing.

Gabriel: The real selling point is something else: according to Google, it finishes tasks faster and makes things up five times less than before.

Lina: Makes less stuff up, but still makes stuff up.

Gabriel: And for programmers it's cheap. According to that independent comparison, it almost ties Claude Sonnet and a GPT model from OpenAI, for a third of what Sonnet costs.

Lina: Mmm. That's going to hurt.

Gabriel: And in Antigravity, Google's coding tool, they launched the Orchestrator: it switches between a big model, Carbon, and the fast one, Hydrogen, depending on the job.

Lina: Like only calling in the expert when you really need them.

Gabriel: Looked at optimistically: top-level ability for everyone, free in Search, cheap for programmers, and less made-up stuff.

Lina: Looked at under a magnifying glass: almost all the numbers come from Google, and the only other one is a single outside chart. And if you make it the default in everyone's search, any mistake gets multiplied.

Lina: Gabriel, picture this: a friend invites you to a video call. You see him on the screen, he says hi…

Gabriel: Normal stuff, happens every day.

Lina: …and it's not him. It's a recorded video. That's how FaceStealer worked, a hacking scheme run by the North Korean government that the FBI just took down.

Gabriel: A video of my friend?

Lina: Yep. First they'd steal someone's account on Telegram, LinkedIn or similar apps. And from there they'd invite that person's contacts to a call on a fake copy of Zoom, Teams or Google Meet.

Gabriel: And the familiar face pops up.

Lina: Or a fake AI-made video of someone famous in the crypto world. Then they'd pretend there was an audio problem and give you a "fix"…

Gabriel: …that was actually malware.

Lina: Exactly. And the worst part: they'd record your camera so they could use you as bait for your own contacts.

Gabriel: So the victim ends up starring in the next scam.

Lina: According to the Justice Department, they stole more than four hundred million dollars in cryptocurrency.

Gabriel: Well, at least they've been stopped.

Lina: The FBI took down their servers and seized more than nine hundred videos of victims. And US prosecutors got a judge's permission to seize or freeze more than one point two billion dollars stolen from Binance, a crypto exchange.

Gabriel: So, friendly advice: if someone on a video call asks you to install something or paste in commands to fix the audio, don't do it.

Lina: Hang up and call them back yourself. It's old advice, but it matters more now than ever.

Gabriel: From the FBI's side, it's a big win: they shut down the servers and exposed how the scam works, and that helps people recognize it.

Lina: From the users' side, it's an uncomfortable lesson: seeing a familiar face doesn't prove anything anymore. And since they waited months before stealing anything, there could be victims who still don't know.

Gabriel: Lina, this one sounds like a detective story. An independent researcher, Will DePue, says he found the fingerprint of Nvidia chips that are banned for China in six Chinese AI models.

Lina: A fingerprint? What do you mean?

Gabriel: Here's the simple version: when you train an AI, it starts out with a bunch of random starting numbers. And each type of chip spreads those numbers out a little differently.

Lina: Like a fingerprint.

Gabriel: Exactly. And in models from companies like DeepSeek and Moonshot, two Chinese AI companies, the fingerprint matches Nvidia's most advanced chips.

Lina: Which can't be sold to China without permission from the United States.

Gabriel: And nobody knows of any permission. Meanwhile, those same companies' older models have the fingerprint of the previous generation of chips.

Lina: A fingerprint isn't a confession.

Gabriel: He tested it on cases where people already knew which chip had been used, and he got them right. And for some models, he checked it on actual chips.

Lina: And where were the chips?

Gabriel: He doesn't say, and he points that out himself: the chips can be rented outside China, and that can be legal. The report doesn't prove smuggling.

Lina: A detective who asks you not to jump to conclusions. Weird, but I appreciate it.

Gabriel: And he makes an estimate he says he's not sure about: around twelve thousand of those chips and more than five hundred million dollars in training.

Lina: He also made his code public, so anyone can redo it.

Gabriel: Read as a warning, it's the first public technical evidence that the chip restrictions are being dodged, something officials had already been claiming without showing proof.

Lina: Read with caution, it's an independent report with no outside review yet, and numbers not even the author is sure about. And a fingerprint, by itself, doesn't tell you how those chips got there.

Gabriel: That's it for today's episode. Tomorrow, five more stories.

Lina: And if I ever video call you and my audio cuts out, hang up. I'll call you back.

Gabriel: Talk to you tomorrow.