Subscribe
Share
Share
Embed
A daily briefing on the AI systems, products, companies, and policy shifts that are just becoming possible.
Want a podcast for your own topics? Join early access: https://www.barelypossible.to/waitlist/?source_path=public_feed&feed_source=rss
Okay kiddos, I'm your boy Tony DeLuca, and this is Barely Possible, the show where we chew through the tech news so you don't have to eat the whole plate yourself. Grab your coffee, pull up a chair, because today we've got two of the biggest companies on earth pointing lawyers at each other, a couple of streaming giants panicking about your attention span, China landing a rocket, and a robot company that wants your favorite chatbot running a machine on a factory floor. Let's have at it.
Let me start with the one that made me put my coffee down. Apple is suing OpenAI. Not a subpoena, not a leak, not a passive-aggressive tweet — an actual lawsuit over alleged trade secret theft. And here's the part that gives it teeth: Apple isn't framing this as some rogue employee who wandered off with a thumb drive. According to the reporting from Sarah Perez at TechCrunch, Apple alleges the misconduct was directed by OpenAI's senior leadership, and it involves a long-time former Apple employee.
Now I want to be careful here, because I don't have the full complaint in front of me and I'm not going to pretend I do. What I've got is the fact of the filing and the shape of the accusation. But let me tell you why the shape matters, because the shape is the whole story.
There's a big difference, legally and reputationally, between "one of our guys took some files" and "your executives told our guy to take some files." The first one is an HR problem with a lawsuit stapled to it. The second one is a claim that the theft was a strategy, a decision made at the top. When you name senior leadership in a trade secret case, you are not asking for a settlement over a severance dispute. You are trying to establish intent. You are trying to say the company knew, the company wanted it, the company benefited. That's the difference between a fender-bender and reckless driving.
And here's the context you already know if you've been listening. These two companies used to be dance partners. Apple built OpenAI's models into the iPhone. That was the deal — Apple didn't have a frontier model of its own that could compete, so it rented one, and OpenAI got distribution into the most valuable pocket in the world. That relationship has been fraying in public for a while now, and this is the frost turning into ice. When your former partner is your rival, all those meetings, all that shared roadmap talk, all those engineers who moved between companies — every one of those becomes a potential exhibit.
So what do you, the builder, take from this? A couple of things. One: the era of frenemies is getting more expensive. When two giants collaborate and compete at the same time, the legal exposure on both sides goes up, and the people caught in the middle are often the employees who moved between them. If you're hiring someone out of a big lab or a big platform company right now, your onboarding needs to include a very boring, very serious conversation about what they can and cannot bring with them, even in their head. Trade secret law doesn't require a stolen file. It can hinge on what someone knew and where they took it.
Two: this is a proxy war over the phone. The real prize here isn't a document. It's the on-device AI layer, the assistant that lives in your pocket. Apple wants to own that and stop renting it. OpenAI wants to be the model everybody rents. When the stakes are the interface to a billion people, the fights get ugly, and they get personal. Watch this one. If Apple's allegations about senior leadership hold up in discovery, it changes how every partnership between a platform and a model lab gets negotiated going forward. Everybody lawyers up harder, everybody trusts less, and the walls between these companies get taller.
I'll flag one more thing and then move on, because I want to be honest about what we don't know. We covered on yesterday's episode the separate mess where the New York Times and a group of publishers filed a sanctions motion accusing OpenAI of hiding and deleting ChatGPT logs. That's a different case, different plaintiff, different facts, and I'm not going to blend them together into some grand narrative that OpenAI is under siege on all fronts — that's the kind of lazy connective tissue I try not to sell you. But I will say this: when a company is fielding a copyright fight from the press and a trade secret suit from Apple in the same news cycle, the legal department is not having a quiet summer. And legal drag is real drag. It pulls executive attention, it slows deals, it makes partners cautious. For a company moving as fast as OpenAI, friction like that is not free.
Now let's move from the courtroom to the factory floor, because there's a story here that I think matters more to builders than its quiet little headline suggests.
Anthropic put out an announcement, dated the ninth, that UST is bringing Claude to physical AI. Now the source text on this one is thin — it's basically the headline and a partnership framing — so I'm not going to invent details that aren't there. But I want to talk about the direction, because the direction is the interesting part.
Physical AI means robots, machines, systems that act in the real world — not just chatbots that write your emails. And what this partnership signals is that the model labs are no longer content to live inside a text box. They want Claude, and by extension the whole class of frontier language models, to be the reasoning brain sitting behind a machine that moves, that inspects, that operates on a factory line.
Here's why I'm dwelling on a two-line announcement. For most of the last couple of years, the language model story and the robotics story ran on separate tracks. Robotics people worried about actuators and sensors and control loops. Language model people worried about tokens and context windows and benchmarks. Those worlds are colliding, and this UST partnership is one more brick in that wall coming down. The bet is that a language model's ability to reason, to plan, to follow instructions in messy real-world language, becomes the layer that sits on top of the physical hardware and tells it what to do and why.
And I'll connect this to something we've been circling on the show. We talked recently about the idea that gaming data might be the training set for physical AI — the notion that if you feed a foundation model enough simulated worlds, it learns something transferable about acting in space. This is the commercial cousin of that idea. It's the part where a services company like UST says, okay, we've got real industrial customers, real machines, and we're going to wire a frontier model into them. That's the difference between a research paper and a purchase order.
For the builder, the takeaway is a question worth sitting with. If the model is going to be the brain of a physical system, then the whole harness problem — the tools, the guardrails, the approvals, the observability — gets a lot more serious when a mistake isn't a bad paragraph but a machine doing the wrong thing in the physical world. If you're building anything that touches hardware, the safety and permission layer is not a nice-to-have. It's the product. Keep an eye on how these physical-AI partnerships handle that, because that's where the real engineering lives, and that's where the real liability lives too.
Alright, let's shift gears entirely. Let's talk about your television, because two of the biggest names in streaming are quietly admitting the same thing this week, and it tells you something about where the whole attention economy is heading.
First, Netflix. There's a report, per Lauren Forristal at TechCrunch, that Netflix could be planning "always-on" live TV channels — 24/7 channels you just tune into, the way you used to flip on the tube and let it run. And the framing in the piece is the tell: this is happening amid signs of slowing engagement. Read that again. Slowing engagement. The company that trained an entire planet to binge on-demand, to pick exactly what you want exactly when you want it, is now looking at bringing back the lean-back channel — the thing that just plays, so you don't have to decide.
And it's not just them. Separately, there's a report from Aisha Malik, also at TechCrunch, that Disney+ is considering a free streaming tier. And the reasoning there is explicit: free content would let Disney+ compete better with free services like YouTube and Tubi, which are eating a growing share of people's viewing time.
Now put those two next to each other, because together they're more interesting than apart. Netflix is worried you're not watching enough, and its answer is to remove the friction of choosing. Disney is worried you're spending your eyeballs on free stuff, and its answer is to give some stuff away free. Both of these companies spent a decade telling us the future was premium, on-demand, ad-free, you're-in-control. And now the pendulum is swinging back toward the thing they replaced — channels that just run, and free tiers that don't ask for your credit card.
What happened? What happened is that the scarce resource stopped being content and started being attention. There is more good stuff to watch than any human being could get through in ten lifetimes. The bottleneck is not "is there something on." The bottleneck is "I'm tired, I don't want to scroll through four hundred thumbnails, just play me something." The paradox of choice is a real tax on people, and the companies that win are the ones who reduce that tax.
Here's the builder lesson, and it's not about streaming. It's about any product where you've given the user infinite options. There's a phase in a product's life where more choice is a feature. You're proud of your catalog, your customization, your flexibility. And then there's a later phase where more choice becomes a burden, where the user is paralyzed and just wants a default, a channel, a "just do the reasonable thing" button. Netflix hit that wall. If your product is maturing and your engagement is softening even though your library keeps growing, ask yourself whether you've handed your users too many decisions. Sometimes the growth move is to take choices away.
And there's a tidy little bookend to this. There's a piece from Amanda Silberling at TechCrunch about a company called Dumb Co, which sells hacked flip phones that sync to your smartphone — the idea being to bridge the always-connected iPhone with the blissful limitations of an early-2000s relic. Same instinct, different corner of the market. People are tired. They want less. They're paying money to have fewer options, whether that's a channel that just plays or a phone that just calls. There's a real, growing appetite for constraint. If you're a founder, "we help people do less" is turning into an actual business model, and it's not a joke.
Now let's get out of the living room and into orbit, because China did something this week that matters for the whole space business.
Per the reporting from Stephen Clark at Ars Technica and Tim Fernholz at TechCrunch, China's state-owned space company recovered its first orbital rocket booster after launch. They landed it, they got it back, and by the reporting they showed a somewhat new approach to how it's done. One quote in the coverage puts it plainly — that they clearly admire the work SpaceX is doing and are trying to replicate it.
Now, on one level, this is a "China copies SpaceX" story, and you can shrug at it. But I'd push back on the shrug, because reusability is the whole ballgame in launch economics. The reason SpaceX flattened the cost of getting to orbit is that they stopped throwing away the most expensive part of the rocket every single time. It's the difference between buying a new car for every trip to the grocery store and, you know, owning a car. When a second serious player — and a state-backed one, with deep pockets and patience — cracks reusability, the price of launch is going to keep falling, and the number of things you can afford to put in orbit goes up.
Why does a founder who isn't building rockets care? Because cheaper launch is infrastructure the way cheap cloud compute was infrastructure. When it got cheap to spin up a server, an entire generation of software companies became possible that couldn't have existed before. When it gets cheap and routine to put hardware in orbit, you get more satellites, more coverage, more bandwidth, more earth-observation data, more of the raw material that downstream businesses build on. This is a supply-side shift that unlocks demand-side products years later. I'm not telling you to go start a satellite company. I'm telling you that the cost curve for the whole layer beneath connectivity and imagery is bending down, and that ripples outward for a long time.
And there's a geopolitical wrinkle you can't ignore. This is state-owned. So the second serious reusable-launch capability on the planet is not a scrappy startup — it's a government program with strategic goals. That means the space economy is going to look less like a free market and more like two national blocs racing, each with its own supply chain. If you're building anything that depends on space-based data or connectivity, start thinking about that split the way you already think about the chip split. Which brings me right to my next thing.
Let's talk chips and money, because there was a Wall Street moment this week that's easy to skim past and shouldn't be.
Per Kate Park at TechCrunch, SK Hynix raised twenty-six and a half billion dollars in what's being called the biggest foreign IPO in US history. And in the same breath, SK Hynix and Samsung are being urged to build new fabs — new chip factories — on US soil. Twenty-six and a half billion. That's not a round number for a startup, that's a country's worth of money for a memory chipmaker.
Here's the thing to understand. Memory is the quiet bottleneck of the whole AI boom. Everybody talks about the fancy processors, the GPUs, the accelerators — but those chips are useless without enormous amounts of fast memory sitting next to them, and the high-bandwidth stuff is in short supply. We've talked before on this show about how the memory shortage is the real constraint under a lot of the AI buildout. So when the biggest memory player on earth raises this kind of money by listing in the US, and then gets leaned on to build factories in the US, you're watching two things happen at once.
One: the market is throwing enormous capital at the supply chain underneath AI, not just at the model companies. The money is flowing down the stack, into the boring, capital-intensive, physical layer — the factories, the memory, the stuff that takes years to build and can't be spun up on a laptop.
Two: there's a national-security overlay on all of it. The pressure to build fabs on US soil is about not wanting the entire AI supply chain to run through a couple of factories in a couple of countries that sit in geopolitically tense neighborhoods. It's the same instinct we just talked about with rockets — the desire to have your own capability, on your own ground, that nobody can cut off.
For the builder, the signal is this. The AI cost story you care about — how much it costs you to run inference, to serve a model, to build your product on top of somebody's API — that cost is downstream of memory supply. If capital is finally flooding into memory manufacturing, and if those fabs actually get built, the medium-term picture is more supply and, eventually, some relief on the input costs that have been squeezing everybody. It won't happen next quarter — fabs take years — but the direction of the money tells you where the smart, patient capital thinks the shortage is. And they think it's memory.
Now let me stay in the money world for a second, because there's a quantum story that's worth a quick, skeptical look.
Per Marina Temkin at TechCrunch, a company called Oratomic raised three hundred million dollars to build what they're calling a viable quantum computer that needs only twenty thousand qubits. The round was co-led by ARCH Venture Partners, Spark Capital, and Khosla Ventures — that's a serious lineup, not tourists.
Now, I'm the guy who's supposed to be skeptical of hype, so let me be skeptical. Quantum computing has been "five years away" for about fifteen years. The graveyard of quantum promises is full. But the "only twenty thousand qubits" framing is the interesting claim, and here's why. A lot of the mainstream approaches to a genuinely useful, error-corrected quantum machine have needed qubit counts up in the millions, because you need mountains of physical qubits to error-correct your way to a handful of reliable logical ones. If Oratomic is saying they've got an architecture where twenty thousand gets you something viable, that's a claim about efficiency, about needing far fewer of the things to do real work.
Should you believe it? I don't know, and neither do you, and neither, honestly, do the investors — that's what a bet is. But three hundred million from names like these means smart people looked at the physics and decided the shot was worth taking. For most founders listening, quantum is not something you're building on next year or the year after. But it is worth keeping in the corner of your eye, because if any of these bets land, the assumptions underneath cryptography — the math that secures basically everything you build — start to wobble. That's not a today problem. But it's a someday problem, and someday gets closer every time somebody raises three hundred million dollars to chase it.
Alright, let me pivot to regulation, because Europe made a move against Meta that I think is a bigger deal than it looks.
Per Ashley Belanger at Ars Technica, the EU is telling Meta to disable autoplay and infinite scroll — or risk massive fines under the Digital Services Act. Read that carefully, because it's not the usual regulatory story. This isn't "delete this content" or "take down that post." This is the EU reaching in and telling a company that specific design mechanics — the autoplay that starts the next video before you've decided to watch it, the infinite scroll that never gives you a stopping point — that those mechanics themselves are the problem.
That's a genuinely different theory of regulation, and every builder should understand it. For years, the fights were about content: what's allowed, what gets removed, who's liable for what a user posted. Now the argument is about mechanics — the engineering choices that keep you glued to the screen. The EU is essentially saying the design pattern is a harm, independent of whatever content is flowing through it. Infinite scroll and autoplay are the two most successful engagement patterns of the smartphone era. They are in basically every consumer app on earth. And a major regulator is now treating them as something you might get fined for.
Here's the connection I can't resist, and it's a real one, not a forced one. Remember what we just talked about with Netflix and Disney? Netflix is trying to reduce user choice with always-on channels, Disney's chasing the free-and-easy model — companies voluntarily leaning into lean-back, low-friction consumption. And now here's the EU coming from the exact opposite direction, saying some of that friction-free design is manipulative and has to go. So you've got the market pulling toward "make it effortless to keep consuming" and the regulator pulling toward "make people choose on purpose." Those two forces are about to collide inside every consumer product team in Europe. If you're building anything with a feed, anything with a player, anything designed to keep someone in the flow — the ground under those default patterns is no longer solid, at least not on that continent. Design your engagement mechanics like they might one day have to be defended to a regulator, because increasingly they might.
Now let me do a quick lightning round of a few things worth knowing before we wrap, because there were a couple of smaller items that a builder should have on the radar.
Bluesky's interim CEO, Toni Schneider, dropped the "interim" — he's the permanent CEO now, per Lucas Ropek at TechCrunch. Schneider ran Automattic, the company behind WordPress, and he's a partner at True Ventures, and he says he's "all in" on the platform. I mention it because Bluesky is the most serious attempt at a genuinely open, decentralized social protocol that's gotten real traction, and having a seasoned operator commit fully instead of keeping one foot out the door is a signal that somebody thinks there's a durable business there, not just a protest movement against the other platforms. If you build in the social or open-web space, the direction Bluesky takes under a permanent CEO matters.
There's also a lawsuit worth a founder's attention. Per Sarah Perez at TechCrunch, the college app Fizz has expanded its lawsuit against rival Sidechat, alleging that a venture investor at Maveron shared Fizz's confidential information — information obtained during a fundraising meeting — with the competing startup. Now, this is an allegation, unproven, and I'm not going to convict anyone on a podcast. But the scenario is a founder's nightmare, and it's worth naming out loud: you walk into a pitch meeting, you open the kimono, you share your metrics and your strategy because that's what raising money requires — and the allegation is that some of that made its way to your direct competitor. Whether or not it's true here, let it be a reminder. When you pitch an investor who also backs, or might back, a competitor, be deliberate about what you hand over and when. The most sensitive numbers can wait until there's a term sheet and a confidentiality obligation with real teeth. Trust is the currency of fundraising, but trust and a signed NDA are better than trust alone.
And one for the smart-home crowd, because it's a small story with a big lesson. Per Scharon Harding at Ars Technica, a firmware update bricked Philips Hue Bridge Pro devices — the little hub that runs your smart lights — and Philips is giving out free replacements, but affected users have to reconfigure everything from scratch. To Philips' credit, free replacement is the right move. But the lesson for anybody shipping connected hardware is old and permanent: a bad update can turn a working device into a paperweight in an instant, and the trust you burn doing that takes far longer to earn back than the money the replacements cost. Test your firmware like the device is going to fail in your own living room, because for some poor customer, it just did.
Okay. Before I let you go, let me tie a bow on the day, because I think there's an honest thread running through most of this, and it's not one I want to overstate into a grand theory.
What I keep noticing is that the industry is being forced, all at once, to reckon with the physical and legal reality underneath the software dream. Apple and OpenAI aren't fighting over an abstraction, they're fighting over people and secrets and a real device in your hand. Anthropic and UST want a model reasoning inside a real machine. China landed a real booster. SK Hynix raised real money to build real factories full of real memory chips. The EU is regulating the real mechanics of real screens. Even the streaming giants are admitting a very human, very physical truth — that people get tired and just want the TV to play.
For a long time the story of this industry was that software eats the world and the messy physical stuff gets abstracted away. What this week says is that the messy physical stuff — the factories, the lawsuits, the regulators, the tired human on the couch — is pushing back, and it's back in the middle of the story. If you're a builder, that's not bad news. It just means the moat is increasingly in the parts that are hard to fake: the supply chain, the trust, the design choices you can defend, the relationships you don't poison. The clever demo is cheap now. The durable stuff is the physical and human reality around it.
That's the menu for today. As always, the links to the pieces we talked through are in the show notes if you want to go deeper on any of it. Take care of yourselves out there, be good to the people who pitch you and the people you pitch to, and I'll be right back here next time. I'm Tony DeLuca — thanks for spending part of your day with me.