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, this is Tony DeLuca and you're tuned into Barely Possible, where we take the noise out of the AI world and hand you back the signal. Today we've got a strange little sandwich shop, a five percent slice of the biggest AI company on earth, a chip war spreading to Samsung, and a couple of stories that should make any founder think twice about their pricing model. Grab your coffee, let's have at it.
Let me start with the story that I keep chewing on, because it's the one that tells you where the whole AI economy is headed, and it comes from a place nobody would expect. TechCrunch's Julie Bort did a piece with a headline that made me laugh out loud: Jersey Mike's IPO illustrates how bad the AI hype has become. Now, Jersey Mike's. This is a sandwich shop. Subs. Cold cuts. The mike's way, with the onions and the oil and vinegar. This is a place where the biggest technology decision should be whether the slicer is clean. And Julie Bort, just for kicks, went and looked at their IPO documents, figuring surely a sandwich chain would have no reason to mention artificial intelligence. And lo and behold, there it was.
Now why does this matter to you, the person building an actual company? Because when a sandwich shop feels obligated to name-drop AI in the paperwork it files to go public, you are looking at the top of a hype curve. This is the moment where the word stops meaning anything specific and starts meaning we would like a higher valuation, please. I've seen this movie before. In the late nineties every company put dot-com in the name. Later everybody was a platform. Then everybody was a marketplace. Then everybody had a blockchain strategy. Now everybody has an AI strategy, right down to the guys making your Italian sub.
And here's the honest part, the part I want you to hold onto. Being skeptical of the hype does not mean the technology is fake. It means the language has gotten cheap. When the language gets cheap, two things happen. First, real builders get lumped in with the tourists, and that makes it harder to raise money on substance. Second, the counter-reaction is coming. The moment enough of these AI-flavored companies stumble, the pendulum swings, and suddenly investors treat every AI claim as a red flag instead of a green one. If you're a founder, your job right now is to be legible. Say exactly what your AI does, what it replaces, what it costs, and what breaks. The people who survive the hype cleanup are the ones who could always show their work.
And speaking of showing your work versus dressing up your paperwork, let's talk about the biggest paperwork story in the business.
Sam Altman and OpenAI have reportedly proposed handing over five percent of the company's equity to a US sovereign wealth fund. This was reported by TechCrunch's Russell Brandom, and Ars Technica's Ashley Belanger covered the same development, framing it as OpenAI floating a five percent stake to win over the AI skeptics, and noting the number is far lower than what Senator Sanders had been pushing for. The idea, as it's been described, is a fund structured something like the Alaska Permanent Fund, the one that collects oil and mining money and cuts checks to Alaskans. At current valuations, a five percent slice would be worth somewhere in the neighborhood of forty-two billion dollars. And the reporting suggests OpenAI has floated the idea that all the leading AI developers should chip in five percent of their equity, which would rope in Anthropic, Google, Meta, and the rest, though there's no sign any of them have agreed to anything.
Let me give you my read, plainspoken. This is not charity. Nobody hands over forty-two billion dollars out of the goodness of their heart. What this is, is a company that has looked at the political weather and decided the smartest insurance policy money can buy is to make the government a shareholder. Because once Washington owns a piece of you, Washington has a reason to want you to succeed. Your upside becomes their upside. That's the whole game. The Financial Times, as others have noted, pointed at what looks like a fairly clear trade: give the government an ownership stake, and you buy yourself goodwill and you take some of the air out of the argument that AI is enriching a handful of people at everyone else's expense.
Now the sourcing is careful here, and I want to be careful too. The people close to it describe these talks as early stage, even conceptual, and note it might take an actual act of Congress to make it real. So don't picture the government cashing a dividend check next quarter. Picture a trial balloon, floated to see who salutes.
But think about what it signals for the rest of you building in this space. If the frontier labs start normalizing the idea that the state takes an equity cut in exchange for goodwill and lighter-touch treatment, that is a new cost of doing business at the top. And it doesn't stop at the top. The precedents that get set for the giants tend to trickle down as expectations for everybody else. If you're building an AI company you plan to sell or take public someday, the political overhang around this technology is now part of your cap table math, whether you like it or not. That's a genuinely new thing. A few years ago, the government wasn't interested. Now the frontier is offering it a seat at the table before it's even asked. Watch this one. It's not about OpenAI. It's about the shape of the whole industry's relationship to the state.
And if you want proof that the state is very much in the mood to squeeze big tech, look across the Atlantic.
Google lost its long-running appeal of a record EU fine. Ars Technica's Ryan Whitwam has the details: Google has to cough up four point seven billion dollars. This is the old case about Google bundling its search engine and its Chrome browser with Android, using the dominant mobile operating system to lock in its own products. Google fought this for years. Years. And they lost. The fine stands.
Here's why I bring a European antitrust decision to a room full of founders. Because the through-line from the OpenAI story to this one is the same: the era where big tech operated with a light regulatory hand is over, on both sides of the ocean, in different flavors. In the US, the government's angle right now looks like ownership and cozy relationships. In Europe, it's fines and structural remedies. Different tools, same message. The platforms that got giant by bundling and defaulting are now paying for it.
What does it mean for you specifically? Two things. One, defaults are power, and regulators have now firmly established that abusing default placement is a punishable offense at scale. If your growth strategy secretly depends on being the pre-installed, pre-selected option through a deal with a gatekeeper, understand that ground is legally shakier than it used to be. Two, four point seven billion dollars sounds enormous, and it is, but for a company Google's size it's a cost of doing business, a rounding error against the value of owning search on every Android phone for a decade. That's the uncomfortable truth about these fines. They arrive so late that the anticompetitive behavior has already done its work. The lock-in is baked in. So when I tell you defaults are power, I mean it as both a warning and, frankly, a strategy note: the incumbents will pay the fine and keep the users.
Now let's move from the courtroom to the chip fab, because the hardware story this week is quietly one of the most important for anyone thinking about the AI supply chain.
TechCrunch's Lucas Ropek reports that Anthropic is in discussions with Samsung about a new custom chip. And the timing is the tell. This comes about a week after OpenAI announced its own custom AI chip in a partnership with Broadcom. So now you've got the two leading model labs, OpenAI and Anthropic, both moving to design their own silicon rather than living entirely at the mercy of Nvidia's pricing and Nvidia's waitlist.
We've been circling the chip and compute theme a lot on this show, so let me zoom into the specific thing that's new here rather than repeat the meta-story. The specific thing is diversification of the manufacturing partner. OpenAI goes to Broadcom. Anthropic goes to Samsung. That's not an accident. When you're spending the kind of money these labs spend on compute, being dependent on a single vendor for your most expensive input is a founder's nightmare at civilizational scale. It's the same lesson any of you learned the first time a single cloud provider jacked up your bill or a single API you built on changed its terms. Concentration of supply is concentration of risk.
And here's the part that matters for the smaller players. When Anthropic partners with Samsung to build custom chips, that partnership creates capacity, expertise, and tooling that eventually flows outward. The hyperscaler custom-silicon path is well worn now. But the involvement of Samsung specifically is interesting because it deepens the memory and fabrication story we've been tracking, the one about South Korea's enormous bet on being the backbone of the AI supply chain. If you're building anything that depends on inference cost coming down over time, the fact that the model labs are now designing their own chips is, on balance, good news for you. More competition at the silicon layer eventually means the price of a token drifts down. It just takes a few years and a lot of Samsung's fab time to get there.
Stay with the money for a minute, because there's a warning label attached to all this compute that doesn't show up on the balance sheet.
TechCrunch's Tim De Chant wrote a piece framing it as a warning sign about AI's real cost, courtesy of Google and Amazon. The gist: AI has made it a lot harder for the big tech companies to hit their net-zero climate pledges. All those data centers, all those custom chips, all that inference and training, it runs on enormous amounts of electricity, and a lot of that electricity is not clean.
I don't do this as a lecture. I do it because it's a business risk hiding in plain sight. Here's the chain of logic. AI demand drives compute demand. Compute demand drives electricity demand. Electricity demand, in a lot of places, drives up prices for everybody and blows a hole in the sustainability commitments these companies made when the board of directors thought it made them look good. Now those commitments are colliding with reality. And when a company has to choose between a climate pledge and shipping the AI product that Wall Street is rewarding, we've all seen which one wins.
Why should a founder care? Because energy is becoming a first-class constraint on this industry, not an afterthought. We talked recently about even Honda pivoting toward data centers, about the enormous compute leases eating up power. The cost of electricity is starting to show up in the cost of your inference. If you're building on somebody else's models, that cost gets passed to you eventually. And if regulators or the public start pushing back hard on the carbon footprint of AI, the companies you depend on could face constraints that ripple down to your product. The clean version of this story is: cheap tokens are not guaranteed forever, and one of the reasons is that the electrons behind them are not cheap or clean or infinite.
Alright, let me shift from the heavy infrastructure to something lighter, but genuinely instructive for product people.
Meta quietly launched an app called Pocket. TechCrunch's Sarah Perez reports it's an experimental app that lets users generate and share interactive mini games using text prompts. They're calling it vibe-coded gaming. You type what you want, the AI builds a little playable game, you share it. It's Meta throwing spaghetti at the wall to see if AI-generated interactive content is a thing people actually want.
Now, I'm skeptical of a lot of these quietly-launched experiments from the giants, and you should be too, because half of them get killed within a year. But I want to point at the pattern, because the pattern is real and it's coming for a lot of categories. The idea that a regular person can describe a piece of interactive software and have it exist a few seconds later, without writing code, that's the whole vibe-coding movement compressed into a consumer toy. For the giants, this is a land grab: whoever owns the tool that turns a casual prompt into a shareable experience owns a lot of attention.
For you, the builder, here's the takeaway. Meta launching a vibe-coded game app is a signal that the bar for creating disposable interactive content just dropped through the floor. That's a threat if your product is a thin layer of easily-generated content. It's an opportunity if your product is the thing that makes generated content good, findable, monetizable, or trustworthy. When creation gets free, the value moves to curation, distribution, and quality control. That's been true in every media wave, and it'll be true here.
Now, from a company giving something away to companies getting caught taking something they shouldn't, let's talk about the FTC, because they had a busy day.
Travel app Hopper agreed to pay thirty-five million dollars to settle FTC allegations. TechCrunch's Lauren Forristal reports the FTC said Hopper used deceptive dark patterns to hide fees and mislead travelers about the costs and benefits of its services. Dark patterns. That's the industry term for the sneaky interface tricks that nudge you into spending more than you meant to, or make it hard to see the real price, or bury the cancel button three menus deep.
And this connects to a separate FTC story on the same day. Ashley Belanger at Ars Technica reports that privacy advocates are urging the FTC to reject Elon Musk's bid to end the privacy monitoring on X, warning that X poses a serious risk to Americans' privacy, particularly given how that data could feed into AI systems. So on one hand you've got the FTC nailing a company for deceptive fee patterns, and on the other you've got advocates asking the FTC to keep a leash on a platform's data practices.
Here's the founder lesson, and I'll say it plainly because it's the kind of thing that kills companies quietly. Dark patterns are a liability now, not a growth hack. There was a stretch of years where hiding fees and making cancellation painful was just considered aggressive growth optimization. Everybody did it, so it felt safe. Thirty-five million dollars later, Hopper is a cautionary tale. If your growth metrics look great because your checkout flow is confusing on purpose, you are borrowing against a future enforcement action. And in an AI world, where you're collecting more data than ever and feeding it into models, the privacy piece stacks on top of the deception piece. Build the honest version. The honest version is slower and it's also the one that doesn't cost you thirty-five million dollars and your reputation.
Let me connect back to something we covered a few days ago, because it's the flip side of this same coin. We talked about Ford bringing back its veteran greybeard engineers after AI tools fell short on quality, the idea that AI is a power tool and not a finished product. The Hopper story and the Ford story rhyme. Both are about the gap between the flashy AI-forward story a company tells and the boring, honest, human work of actually getting it right. The companies that win the next few years are going to be the ones that resist the temptation to let AI paper over things that require real craft, whether that's engineering quality or an honest checkout page.
Now let me do a quick lap through the rest of the board, because there's a bunch of smaller stuff worth knowing.
Tesla sales were up twenty-five percent in the second quarter of this year, per Ars Technica's Jonathan Gitlin. Interesting wrinkle: deliveries outstripped production, which suggests Tesla cleared out some inventory it had sitting around. So it's a strong-looking number, but read it carefully, because moving warehoused cars is not the same as building and selling fresh demand. Good quarter on the surface, worth a second look underneath.
On the money side, a couple of fund stories. TechCrunch's Dominic-Madori Davis reports Melinda French Gates' firm, Pivotal Ventures, is backing Magnify Ventures' new forty-six point six million dollar Fund Two. Early-stage capital is still flowing to the right people. And Marina Temkin reports that Thiel Capital's Jack Selby has been picking up stakes in hot startups like Etched, the Nvidia-challenger chip company we mentioned recently, through his firm Copper Sky Capital, which is raising a three hundred million dollar second fund. The Etched connection is the thread there, tying back to that custom-silicon gold rush we keep circling.
Quantum had a moment too. TechCrunch's Anna Heim reports that IQM, a full-stack quantum company out of Finland, went public on the Nasdaq at a valuation of around one point nine billion dollars, and in a refreshing bit of honesty, the company itself admitted the future of the technology is uncertain. I'll give them credit for that. A quantum company saying out loud that quantum's future is uncertain is more truthful than a sandwich shop pretending it needs an AI strategy.
A couple more. Lime, the scooter and bike-share outfit, finally went public after years of uncertainty, per Sean O'Kane at TechCrunch, and the honest reason is right there in the filing: they need the funds to help pay down around a billion dollars in liabilities. That's less a triumphant IPO and more a company using the public markets as a life raft. And in a story that should worry anyone in aviation software, O'Kane also reports that Boeing-owned Wisk Aero is accused of firing a manager who raised safety concerns, with the former software manager claiming Wisk rushed software testing ahead of a crucial flight test. Rushed software testing, silenced whistleblower, aircraft. That's a combination that has burned Boeing before, and it's not the kind of thing you want in a company building autonomous air taxis.
On the security beat, Dan Goodin at Ars Technica flagged a newly discovered piece of macOS malware called PamStealer, notable because it uses some clever tradecraft to stay hidden. The bigger point Goodin makes is that a lot more effort is going into Mac infostealers than there used to be, which is a reminder that the old comfortable assumption that Macs don't get malware is dead. If your team runs on Macs, and half of you do, don't get complacent.
And one piece of consumer-facing news that quietly signals a business-model shift worth watching. Ars Technica's Scharon Harding reports that Plex, the media server company, replaced its old lifetime pass with a five-year membership pass for two hundred fifty dollars. Same price, a lot less permanence. Two hundred fifty used to buy you forever. Now it buys you five years. That's a company nudging its customers toward more frequent payments, and it's the exact tension every subscription founder wrestles with: the customer wants to pay once, the business wants recurring revenue. Watch how Plex's community reacts, because it's a live case study in how much goodwill you can burn when you take away a beloved lifetime deal.
Let me pause on that Plex story for one more beat, because it actually connects to a bigger theme running under this whole episode. Sony announced, per Scharon Harding again, that it will stop making physical PlayStation discs in 2028 and is shutting down parts of its digital store, with one quote in the piece capturing the mood perfectly: we will own nothing, it's truly sad. Plex trading a lifetime pass for a five-year one, Sony killing discs, these are the same story. The industry is steadily converting things you own into things you rent. For founders, that's the direction of gravity in nearly every category: ownership becomes access, access becomes subscription, subscription becomes the relationship. Just understand that every time you make that trade, you're spending trust. Do it too aggressively and, like the Hopper situation, it eventually catches up with you, whether from regulators or from customers walking away.
Alright, let me bring this home.
If I zoom out on today's menu, the connective tissue is trust and cost, and how the AI boom is stress-testing both. OpenAI offering the government a five percent stake is a company spending equity to buy political trust. Google's four point seven billion dollar fine is the price of having burned regulatory trust years ago. The energy story is the real cost of AI showing up where the pledges said it wouldn't. Hopper's thirty-five million dollar settlement is the price of burning customer trust with dark patterns. And a sandwich shop putting AI in its IPO is the moment the word itself started spending its credibility.
Here's what I want you to walk out with. The cheap era is ending on both ends. Cheap regulatory treatment, cheap trust, cheap energy, cheap hype language, all of it is getting repriced right now, in real time. The founders who thrive through this aren't the ones with the flashiest AI claim. They're the ones who are honest about what their product does, honest about what it costs, and honest with the people who use it. In a world where everybody from a sub shop to a frontier lab is invoking AI, being the one who can actually show your work is going to be the rarest and most valuable thing you own.
That's the show. I'm Tony DeLuca, this has been Barely Possible. Be skeptical, build honest, and I'll see you tomorrow with a fresh plate.