70% of Americans Oppose Data Centers, YouTube Collateral Damage, and OpenAI's Wiggly Donut

Yesterday in AI  |  10 August 2026

70% of Americans Oppose Data Centers, YouTube Collateral Damage, and OpenAI's Wiggly Donut

Artificial intelligence infrastructure, content filters, and hardware ambitions collided with local communities and environmental limits this weekend. This episode breaks down new polling showing 70% of Americans oppose local data center construction, while Amazon finances a 7.65-gigawatt gas power plant in Texas permitted for 33 million tons of annual carbon emissions.

We explore ByteDance reportedly training a massive 10-trillion parameter model, analyze YouTube's "AI slop" crackdown accidentally burying legitimate human creators, cover OpenAI quietly acquiring slide-generation startup NextSlide, and examine reports detailing OpenAI's $300-$400 screenless "donut" hardware device being designed with Jony Ive.

Feedback? Email mike@yesterdayinai.news or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Transcript

SPEAKER_00 0:05

Hi folks and welcome back to another edition of Yesterday in AI, your daily digest of everything happening in the world of AI in roughly 10 minutes. I'm Mike Robinson. It's Monday, August 10th, and the AI industry spent the weekend crashing into the physical world. Power plants, angry neighbors, YouTube creators getting caught in a filter meant for robots, and, right at the end, a little talking donut that wants to live on your kitchen counter and watch you cook. Let's get into it. We start with a story that might be the biggest one in AI all year, and it has nothing to do with a chatbot. It's about the buildings. A Gallup poll this spring put a hard number on it. Roughly seven in ten Americans do not want a data center built anywhere near them. And here's the part that should worry every tech company with a construction budget. It crosses party lines. About two-thirds of Republicans and 75% of Democrats said the same thing. It gets funnier. When you split the Republicans, the hardcore conservatives oppose these things more than the moderates do, which puts the most conservative voters and the most progressive voters holding hands on the same side of the fence. In 2026, the one thing that unites this country is not wanting a warehouse full of humming computers next door. Here's a quick refresher because the term data center gets tossed around like we were all born knowing what it means. A data center is a giant building packed wall to wall with servers, the machines that train and run the AI you actually use. It takes an enormous amount of electricity to run and a huge amount of water to keep from cooking itself. A big one can pull as much power as a small city. So when a company wants to plant one outside your town, you're not getting a quiet little office park. You're getting round-the-clock noise, a serious spike in local power demand, pressure on your water, and a very large, very ugly, windowless fortress on the horizon. People have noticed and they're organizing. Something like 100 local moratorium proposals are floating around the country right now. One county in Florida, Hernando, already hit pause for a full year on new applications, and this is now moving votes. NPR reported this weekend that data centers have become a live campaign issue, with candidates in both parties running on it. The clearest sign of the mood might be the money. In just the first three months of 2026, roughly $130 billion of data center projects got stalled or blocked. This is the AI build-out running straight into a wall made of city council meetings and losing. And then there's Amazon, which decided to show us all exactly what's at stake. The New York Times reported over the weekend that Amazon is financing a massive private power plant in Picos County, Texas, to feed a new data center. And it's a gas plant, in 2026. We're talking 35 turbines in up to 7.65 gigawatts of capacity, which is a staggering amount of power, enough to run several million homes, and it's permitted to release up to 33 million tons of carbon dioxide a year. And if it gets built to that spec, that one facility would be the single largest source of climate pollution in the entire United States. One plant. To put 33 million tons into something you can picture, that's roughly the same as putting an extra 7 million cars on the road. Here's the honest read. Amazon says building its own power on site means it won't raise electric bills for Texas families, and going off the grid does let them get their AI running faster, which is the whole point. But Amazon's own emissions went up 16% last year, and this is the same company with a public promise to hit net zero by 2040. You can't burn your way to net zero. If the actual plan for powering the AI future runs through the dirtiest single plant in the country, I'd rather someone just say that plainly than dress it up in a press release about helping families. Be honest about the trade-off and let people decide if it's worth it. Why do we need all these power-hungry boxes in the first place? Because the models keep getting bigger. And that takes us straight to a report out of the Financial Times last Thursday. TikTok's parent company, ByteDance, is reportedly training a brand new model with up to 10 trillion parameters. Let me translate that number because it sounds like monopoly money. Parameters are basically the adjustable knobs inside a model, the little dials it tunes over and over as it learns from data. More parameters means more room to store patterns, more raw capacity to work with. Think of it like the number of connections in a brain. 10 trillion is an absolutely staggering number of knobs. For scale, the biggest Chinese open model, Kimik3, has about 2.8 trillion. Anthropic's most powerful system, Mythos, is estimated to be around 8 trillion, though Anthropic never publishes the real figure. So honestly, everybody is squinting and guessing. If the FT has this right, ByteDance is aiming past all of them. And ByteDance of all companies has a reason to swing this big. This is the outfit that runs TikTok, whose entire empire was built on one of the best recommendation engines ever made. The thing that decides what video you see next. Pushing to the frontier of raw model size is the next logical move for a company that already thinks in billions of users. I keep the excitement on a short leash though. This is a report, not a shipped product, and the model is still in the early training stage, and that usually takes several months. And bigger does not automatically mean smarter. We spent half of July watching cheaper, smaller models from Deep Seek and Alibaba go toe to toe with the giants for a fraction of the cost. So a 10 trillion parameter model is a gigantic, wildly expensive bet, and expensive bets miss all the time. What it does tell you is that the raw scale race is very much still on, and every one of those knobs has to be trained and then run somewhere. Somewhere with 35 gas turbines and a very unhappy town nearby. It all connects. Now these giant models happen to be spectacular at one thing in particular, pumping out content, endlessly, for basically free. And that has turned into a giant headache for YouTube. All year YouTube has been cracking down on what everyone politely calls AI slop, the tidal wave of low effort machine-cranked videos clogging the site. Back in January it wiped out 16 channels that together had about 35 million subscribers and 4.7 billion lifetime views, all under a policy against what it calls inauthentic content. That sounds reasonable. Here's the backfire, and it's the whole story. The algorithm changes that came with that crackdown are now hammering a bunch of creators that never touched AI at all. If you run one of those faceless channels, no talking head, just narration over stock footage, or rain sounds for sleeping, or a quiet history explainer, YouTube's system increasingly lumps you in with the machine slop and quietly buries you or cuts off your money. The filter built to catch the robots cannot reliably tell a robot apart from a hardworking human who simply doesn't want to be on camera. And it's a real bind, I'll give YouTube that. The slop problem is legitimate and getting worse by the week. But when your only tool is a sledgehammer, it's the honest little channels that get smashed, while the actual slop factories just spin up fresh accounts by Tuesday morning. If you watch this stuff or make it, your feed just became a battlefield. While YouTube is busy trying to keep AI content out, OpenAI spent time trying to push AI further into the drier parts of your actual workday. On Saturday it came out that OpenAI has quietly acquired a startup called NextSlide. What did NextSlide do? It took your rough notes, your documents, your half-baked research, and turned them into finished editable slide decks. The team is already folded into ChatGPT. The deal apparently closed earlier this year, and the price, as always with these, is a secret. I find this one quietly revealing. OpenAI is buying an entire task here, the presentation, which might be the single most dreaded ritual in all of office life. Nobody in the history of work has ever finished building a slide deck and thought, what a rich and meaningful use of my Tuesday afternoon. That is exactly the kind of grinding busy work first in line to get handed off to a chatbot, and there's a bigger pattern underneath it. OpenAI has stopped shopping for clever little features. It's swallowing whole jobs one startup at a time. So if a decent chunk of your week is turning notes into slides that somebody skims for eight seconds in a meeting, well, that's the piece the machine is coming for first. Consider it fair warning, and, if we're honest, maybe a small mercy. Which brings us to where OpenAI actually wants to end up, and it's the perfect place to close. OpenAI wants more than a spot in your browser tabs. According to a report from Bloomberg's Mark German last Thursday, its very first piece of hardware is a small screenless speaker you'd set on a table. Picture something about the size of a hockey puck, or, as roughly everyone on the internet immediately called it, a donut. It's got a camera, it's got environmental sensors, and here's the really strange part. It's got moving parts. Little motors, so the thing physically shifts and reacts, apparently to feel more alive and less like a gadget. It's being designed with Johnny Ive, the man who shaped the iPhone, through the hardware startup OpenAI bought last year for $6.5 billion. Foxconn would build it. The price is reportedly $300-400, and they're aiming to reveal it later this year for a 2027 release. So mark your calendar for a while from now. Let's just ponder this picture for a second. The plan is a little donut on your kitchen counter with a camera and a microphone that gently wiggles at you while it listens to and watches everything happening in the room. Some of you just found your dream gadget and you're already pre-ordering. The rest of you just felt a small cold shiver and glanced up at your ceiling. Me? I'm mostly stuck on the realization that we're pouring gas plants, trillions of parameters, and the smartest people alive into an AI future so it can build your slides, guard your video feed, and best of all, jiggle on your countertop. What a time to be alive. And that's the show. If you have feedback from me, email Mike at yesterdayNai.news or connect with me on LinkedIn, X or Blue Sky. If you enjoy Yesterday in AI, please take a minute to write and review the podcast wherever you listen, or share it with a friend. Thanks for tuning in today. Stay curious, and I'll see you tomorrow.