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 we've got a fresh plate of tech news to work through today, so grab your coffee and settle in. This is Barely Possible, the show where I read the technical stuff so you don't have to, and I try to tell you what actually matters underneath the headlines.
We've got a big one to lead with today, and it's not a model release, it's not a robot, it's a number. Anthropic went from about nine billion dollars in revenue run rate in 2025 to a forty-seven billion dollar revenue run rate by May. Let me say that again because it deserves to be said slow. Nine to forty-seven, in a matter of months. And the guy telling that story is Matt Murphy over at Menlo Ventures, whose firm led Anthropic's five-hundred-million-dollar Series D. Murphy says in twenty-five years of investing he has never seen growth like this. Not the internet wave. Not mobile. Not the first cloud boom. Nothing.
Now here's the part I want you to sit with, and it's right there in the title of the piece: it's not the model. Murphy's argument, and I think this is the real angle for you if you're building something, is that Anthropic didn't win because Claude is definitively smarter than everybody else's chatbot on some benchmark. It won because it became the thing developers and enterprises actually route their real work through. The coding, the agent workflows, the boring back-office plumbing that companies pay real money for. The model is table stakes. The distribution, the trust, the place it sits in your actual workflow — that's the business.
And I want to tie this right into the next thing, because it's the same lesson wearing a different suit. Google reported earnings, and if you've been listening to the market types wring their hands about whether all this AI spending is a giant bonfire of shareholder money, here's your answer for this quarter. Google Cloud revenue jumped eighty-two percent year over year, up to twenty-four point eight billion dollars. That blew past what Wall Street was looking for, which was around twenty-two and a half billion. And the company said flat out: that growth is coming from enterprise AI adoption, enterprises buying the infrastructure and the solutions.
The number that made me put my coffee down, though, was the backlog. Google's cloud contracting backlog — that's signed work they haven't turned into revenue yet — climbed to five hundred fourteen billion dollars. Half a trillion in committed future work. Sundar Pichai got pressed by analysts on the call about when all that capex, and they're spending somewhere between a hundred eighty and a hundred ninety billion this year on data centers and chips, when that pays off. And his answer was basically: the demand signals look healthier than they did a year ago, including long-term deals. So there's your justification for the spend, at least this quarter — the enterprise is actually opening the checkbook.
So look at those two stories together. Anthropic winning on where it sits in the workflow, and Google justifying the whole spending binge on enterprise cloud demand. The through-line for a founder is the same: in this phase of the game, the money is in being the thing businesses build their operations on top of, not in having the shiniest demo. Enterprise adoption is the story now, not model bragging rights.
Now let me shift from the money to the muscle behind the money, because there's a genuinely strange geopolitical fight that broke open this week and it's going to affect what models you're legally allowed to use.
The U.S. Treasury Secretary, Scott Bessent, doubled down on threatening sanctions against Chinese AI companies. This came hours after a White House official, the science and technology policy chief Michael Kratsios, accused the Chinese firm Moonshot of, quote, improperly distilling Anthropic's Fable model. Bessent's line, and he posted this on X, was, quote, "Open source is not open season on American IP." And he said when Chinese firms conduct covert, industrial-scale distillation attacks that cross into IP theft, sanctions and Entity List designations are on the table.
Let me translate the jargon, because it matters. Distillation is when a smaller model learns from the outputs of a bigger one. It's a totally normal, everyday optimization technique — everybody does it, it's how you make a cheap fast model that mimics an expensive slow one. But it can also cross into stealing intellectual property depending on how you do it and whose outputs you're slurping up. So the accusation here is that Moonshot took Anthropic's Fable and distilled it into their own thing.
And here's where I get skeptical, and the reporting from Rebecca Bellan at TechCrunch flags this too. Some experts flat-out dispute that Moonshot's Kimi K3 could have been built primarily by distilling Fable, because Fable has only been publicly available since July 1st. Kimi K3 came out last week. You don't build a frontier open-weight model in three weeks by copying somebody's homework. The timeline doesn't really work. There's also a separate allegation that Moonshot got its hands on Nvidia's GB300 servers — the Blackwell-generation stuff that's banned from sale to Chinese companies — and accessed them in Thailand. That's an export-control question, which is a whole different animal.
Now, we touched on the open-weights fight a couple days back when we talked about the broader debate over who should own AI, and I don't want to re-plow that whole field. But here's the new wrinkle that actually matters for you as a builder. The reason Washington is suddenly so worked up is that Kimi K3, an open-weight model, has capabilities good enough that it's making people question the business models of the big American labs. If a free-to-download model does ninety percent of what Fable or GPT-5.6 does, why would a customer pay the premium? That's the anxiety under all the national-security language. And the policy response being floated — restricting or effectively banning Chinese open-weight models — could genuinely change which tools are on your menu, at what cost, with what liability attached. So if you're building on open weights, and a lot of you are, keep one eye on Washington, because the ground under your feet is not stable right now.
Alright, let me pivot to something completely different, and honestly this is the one that surprised me most in the whole stack today. Let's talk about space privateers. Yes, you heard me. Pirates. In orbit. Sanctioned by the government.
Stephen Clark over at Ars Technica wrote this piece and I loved it because it starts two hundred forty years ago and lands right in the middle of a modern Pentagon fight. So here's the setup. Back in the early days of the United States, when the Navy was a joke and we couldn't fight the British or the French head-on, presidents used privateers. A privateer is a private ship with a government permission slip — it's called a letter of marque and reprisal — that legally lets you go attack the enemy's ships and keep the loot. It's in the Constitution. John Adams loved it, used it against France in the so-called Quasi-War in the seventeen-nineties. James Madison approved it against the British Royal Navy in the War of 1812.
Now here's the beautiful part. Benjamin Franklin, who had backed privateering during the Revolution, turned around after the war and argued it should be banned by everybody. He wrote in a 1785 letter, quote, "This will be a happy improvement of the law of nations. The humane and the just cannot but wish general success to the proposition." And there was another founder, William Whipple, from New Hampshire, who was blunter about it. He said the people who do this, quote, "soon lose every idea of right and wrong, and for want of an opportunity of gratifying their insatiable avarice with the properties of the enemies of their country, will without the least compunction, seize properties of their friends." Two hundred forty years old and it reads like a warning about mercenaries yesterday.
So why is a space reporter digging up Ben Franklin? Because the lines between commercial and military in space are blurrier than in any other domain. Starlink and commercial surveillance satellites are underpinning actual battlefields right now. The tentative global consensus is that yes, those privately-owned satellites can be legitimate military targets. So flip it around: if a private operator goes on offense in space, is that space piracy? Not if they hold a letter of marque. Suddenly a two-century-old legal instrument is relevant again.
And here's the concrete policy fight. Senators Mike Lee and Rep. Tim Burchett actually authored a bill for privateering against drug cartels, and introduced a separate one this month for cyber warfare — Lee called it, quote, "the ultimate in white-hat hacking: stealing from America's enemies at the direction of the president and splitting the profits with the federal government." Meanwhile, last year's defense authorization bill included an amendment that prohibits the Pentagon from fielding a privatized or subscription-based missile-defense intercept capability. In plain English: if we build space-based interceptors for this Golden Dome missile shield, the military has to own them, not lease them from a contractor.
And at the confirmation hearing for the next Chief of Space Operations, Lt. Gen. Doug Schiess, Senator Jack Reed laid down the analogy that I thought was the sharpest line in the whole piece. Reed said, quote, "Some companies want to lease certain space assets that are actually military weapon systems that can be used to destroy other systems and even kill people... We don't lease M4 rifles. We don't lease F-15s." And Schiess agreed — commercial leasing is fine for communications and space awareness, but when it's a weapon system that can harm someone, quote, "that needs to be completely owned by the Space Force and needs to be controlled."
Now why does this matter to you, a builder, not a general? Two reasons. One, the Space Force is about to get a monster cash infusion — the administration asked for seventy-one billion for fiscal 2027, nearly an eighty percent jump, and lawmakers look ready to approve somewhere between fifty and sixty billion. That's an enormous procurement pipeline opening up for commercial space companies. And two, it's the exact same governance question we keep circling in AI, just wearing a space suit. When does a private tool become so powerful that the state has to own it outright rather than rent it? That's the Anthropic-versus-public-infrastructure question. That's the Chinese-open-weights question. And now it's the space-weapons question. The technology keeps outrunning the norms, and every domain is having the same argument about who's allowed to hold the dangerous thing.
Okay, let's come back to Earth and talk about robots and the people whose jobs they threaten, because there's a real labor story developing and I want to be careful with the framing here.
The fresh piece from Jeremy Hsu at Ars Technica is Hyundai pushing back on reports about its striking South Korean autoworkers. Now, we covered the original Hyundai union walkout back a week or so ago — the workers reacting to Boston Dynamics Atlas robots — so I'm not going to re-litigate that. What's new here is Hyundai's denial and the legal context around it, and that's the part worth your attention.
Hyundai is disputing the idea that the strikes were spurred by the robot deployment. The company says the union's demands are about compensation — wage increases, bonuses, extending the retirement age — and that, quote, "Potential deployment of robots in Korean production facilities is not part of the current labor-management discussions." Hyundai's clarifying that the current Atlas plan only covers its Metaplant factory near Savannah, Georgia, starting in 2028. They want twenty-five thousand of these Atlas robots eventually.
But here's the tell, and the Wall Street Journal caught it. The union is making, quote, "unprecedented demands seeking to enshrine job protections in the era of robots and AI." So even if the robots aren't formally on the table, the shadow of the robots is absolutely shaping the negotiation. The union had already warned Hyundai, quote, "Remember that without labor-management agreement, not a single robot using new technology will be allowed to enter the workplace."
And this is where it gets genuinely important for anyone thinking about automation strategy. South Korea passed something called the "yellow envelope law" that took effect in March. It expands labor rights and, critically, it recognizes management decisions that significantly affect working conditions as subject to collective bargaining. Labor experts told the Korea Times that means large-scale automation using robots has to be negotiated with human workers. Read that again. In South Korea, you may now legally have to sit down with the union before you can roll robots onto the floor. That is a structural constraint on automation timelines that a lot of these humanoid-robot business plans have not priced in.
And it's not just Korea. The U.S. autoworkers, the UAW, are trying to organize that Georgia Metaplant, which is currently non-union. Shawn Fain, the UAW president, warned at the convention in June against, quote, "the threat of humanoid robotics and mass automation." So if your whole thesis is "we deploy twenty-five thousand humanoids and cut labor costs," the humans and the laws are going to have something to say about the schedule. That's the real risk in the robot story, and it's not technical, it's political.
Now let me connect that to a Tesla story, because Elon just gave us a beautiful reality check on the technical side of the same dream. Tesla's second-quarter shareholder letter came out, and the company quietly walked back its production timelines on three products — the Cybercab, the Semi, and the Megapack 3. They're no longer promising volume production of those in 2026. And they scrubbed the language from the previous letter about Optimus, the humanoid robot, reaching volume production too.
Musk himself said of Optimus on the call, quote, "This is going to be the hardest product to scale manufacturing that we've ever made at Tesla, because everything on the robot is new." Now, coming from the guy who's spent years telling investors the robot future is right around the corner, that's about as close to humility as you're going to get. The financials tell the story of a company burning cash to chase this transition — net income down five percent to one point one billion, operating expenses ballooning forty-seven percent, negative free cash flow of a billion dollars in the quarter, and capex projected at twenty-five billion this year, about three times what they historically spend. Revenue was actually up nicely, twenty-eight point two billion, best delivery quarter in a while. But the bottom line got squeezed because they're plowing everything into the next generation.
So put Hyundai and Tesla side by side. Hyundai's robot problem is the unions and the law. Tesla's robot problem is that building the things at scale is brutally hard and everything's new. Both of them are betting the company on humanoids, and both of them just gave you a reason to be skeptical of the "any minute now" timeline. If you're building in the physical-robotics space, the lesson is that the demo is the easy part. Manufacturing and labor are where dreams go to get renegotiated.
And speaking of people betting big on the physical world, here's a name from the past resurfacing. Travis Kalanick — yeah, the Uber guy who got pushed out in 2017 — his robotics company, Atoms, raised one point seven billion dollars in a round led by Andreessen Horowitz. This is tied to a company name he revealed back in March, so I'm framing this as the older reveal now getting its funding. Ben Horowitz is joining the board, and here's the twist that made me raise an eyebrow: Uber, the company that threw Kalanick out, also joined the round. Reconciliation, or just good business? Probably some of both.
Atoms is built on top of his ghost-kitchen play, CloudKitchens, and back in March he'd acquired Pronto, the heavy-industry automation outfit run by Anthony Levandowski — another name that carries baggage. Kalanick's pitch, in his own words, is that he wants to, quote, "digitize the physical world," building what he calls "atoms-based computers where CPU is manufacturing, storage is real estate, and network is transportation." Ben Horowitz posted "Travis Is Back" and said the goal is to do for the physical world what Uber did for transportation. Look — one point seven billion is real money and a16z doesn't write those checks casually. But the specifics of what Atoms actually builds are still vague, and when the pitch is mostly vibes about conquering "Nature" as the final boss, I keep my wallet in my pocket until I see the product. File it under: interesting, watch it, don't believe it yet.
Let me give you a palate cleanser, because not everything in tech is about who owns the killer robots. Science Corporation — this is Max Hodak's company, the guy who co-founded Neuralink and left in 2021 — got European regulatory approval to sell a vision-restoring chip. The device is called PRIMA, and this is one where the underlying approval traces back to last fall, so I'm framing it as the news catching up rather than something that happened this morning.
Here's what it does, and it's genuinely moving. It treats age-related macular degeneration, which destroys the light-sensitive cells at the back of your eye and makes it impossible to read or recognize faces. Patients get a small chip implanted behind the eye in an hour-long outpatient procedure, then wear camera-equipped glasses that beam the world to the chip. And Hodak told TechCrunch, quote, "One of our patients in France finished a 300-page novel a little while ago, and sent us the book. We have a sketch on the wall that one of our patients drew of the Sydney Opera House." People who lost their vision, reading novels and drawing landmarks again.
But here's the builder's lesson buried in the feel-good story, and I want you to catch it because it's shrewd. Hodak said the whole point of the vision business is to fund the harder, longer-term brain-interface work. His words: "we don't get to do the bio hybrid stuff long term if you don't have a great vision business. That's what's really financing the rest of it — that's the thing that investors know how to build spreadsheets around." That right there is a masterclass in sequencing a deep-tech company. You find the near-term thing that generates real revenue and that investors can model on a spreadsheet, and you use it to bankroll the moonshot. Every founder chasing a twenty-year vision should tattoo that on their arm.
Now shift from hardware to the softer, product side of things. There's a nice little contrarian startup story I want to flag. A company called Yope raised twelve point three million, led by Northzone, to build a private social network with no algorithms, no ads, and no public content. Their whole thing is "micro communities" — small groups of real-life friends and family sharing photos and messages privately.
Now, I've heard the "anti-algorithm social app" pitch a hundred times and it usually dies because these things can't grow without creator content to keep people hooked. But Yope's numbers are actually interesting. Nearly fifteen million registered users, ten to twenty million pieces of content shared weekly, and the CEO Bahram Ismailau claims more than half their users open the app at least five days a week. His insight was sharp: he said young people, quote, "don't have any place to self-express if they are not ready to be an influencer." That's a real gap. Not everybody wants to perform for strangers; some people just want to share a dumb photo with six friends. The design borrows from old MySpace — customizable walls, collage layouts. It's monetizing through subscriptions instead of ads.
Why do I flag this for a founder audience? Because it's a live test of whether you can build a scaled consumer social product on privacy and subscriptions instead of surveillance and advertising. If it works, it's a real counterexample to the idea that the only viable social model is the attention-and-ads machine. And there's a nice bit of contrast this week, because in the same batch of news, the big social-media addiction lawsuit against Meta — the one that was heading to a bellwether jury trial — got dropped. That's older March-era litigation catching a new development. The plaintiff walked away without any payment, TikTok and YouTube had already settled, and Meta said the outcome shows they, quote, "will not back away from defending ourselves against baseless lawsuits." So the legal pressure on the addictive-engagement model eased up a bit this round, right as a startup is trying to prove you don't need that model at all. Interesting timing.
Let me do a quick lightning round of the announcement stuff, because the labs were busy pumping out press releases and I want to give you the signal without the marketing.
OpenAI put out a bunch of things. There's OpenAI Presence, which is an enterprise agent platform for deploying voice and chat agents into customer and internal workflows — which, notice, is exactly the enterprise-workflow real-money game we opened the show with. There's a national science push with the Department of Energy and the national labs. And there's Project Camellia, a data-center build in Effingham County, Georgia, with the usual commitments about energy and jobs and community investment. That last one connects to the electricity story we've been tracking — these AI infrastructure builds are landing in real counties with real power demands.
And Anthropic, beyond the revenue story, launched something genuinely useful for our audience: the Anthropic Economic Index connector for Claude. This is current, launched this week. It lets you just ask Claude questions grounded in real data about how AI is actually being used in the economy — which occupations use AI the most, what tasks teachers use Claude for, what's being automated and how it's changed over the past year. Now, big caveat they state themselves: it reflects patterns in Claude usage, not the whole labor market. So it's a lens, not the truth. But if you're trying to figure out where AI is actually biting in your industry, that's a free data source worth an afternoon of poking around. And it pairs with their separate two-hundred-million-dollar Economic Futures Research Fund — that agenda dropped a bit earlier and got recirculated this week — which is funding big studies on how to cushion workers from AI displacement. Which, notice, is the exact same labor-disruption anxiety we heard from the Hyundai union and Shawn Fain. Same fear, being approached from the lab's philanthropic side.
So let me tie the whole thing together before I let you go, because there's a spine running through today that I didn't set out looking for but it emerged. Almost every story today is a version of the same question: as these powerful tools go commercial, who gets to own the dangerous or valuable part, and who gets squeezed?
Anthropic and Google are winning by owning the enterprise workflow layer. Washington is fighting over whether Chinese open-weight models should be allowed to erode that ownership. The Space Force is drawing a hard line that weapons must be government-owned, not leased from privateers. The Hyundai union is using new law to force a seat at the table before robots can take the floor. Tesla is discovering that owning the robot dream is a lot harder than announcing it. And Max Hodak figured out how to fund the moonshot with a business investors can actually model. Ownership, control, and who bears the cost. That's the whole plate today.
That's the show, kiddos. I'm Tony DeLuca, and this has been Barely Possible. Go build something worth owning, and I'll see you tomorrow.