All week we've been asking when the AI cost curve actually bends. Today somebody finally put a number on it — and it bends a lot earlier than the keynote slides promised. If you're just joining: AI usage stopped being a product feature and became a balance-sheet problem. Satya Nadella reframed tokens as “token capital,” and the enterprise-agent crowd dragged it into CIO reality — permissioning, production, cost controls. So when agents become real workloads, who can actually afford to run them? This is Tech Podcast Podcast. Today — a $60K monthly token bill, SpaceX IPO math, and a model that catches a four-year bug, then won't take the job. Let's start with the number Joey's been waiting on. We're staying on AI token economics — follow the show and you won't miss what comes next. Here's what All-In with Chamath, Jason, Sacks & Friedberg is reporting. So All-In runs a live pitch competition: four investors, four trades — MGM, Talen Energy, Aktis Oncology, and a crypto wildcard, GEODNET, from Kyle Samani at Multicoin. It's a genuinely good format on paper — make people defend a position out loud. Aaron Cowen pitching MGM, Dan Dreyfus on Talen, real conviction trades with a clock running. Three macro-and-equities pros, and then Samani walks in with GEODNET, a decentralized GPS network. One of these things is auditioning for a different show. Right, but that's what I'd be listening for in this episode — does the room push him on the unit economics, or does the crypto pitch get waved through because it's the fun one? Sponsored by EY, the NYSE, and a wearable that records every word you say. Nothing says independent stock picks like the exchange itself buying the ad. a16z Show, with Erin Price-Wright:
In the United States, the average salary of an electrician in Texas is now higher than that of a Silicon Valley software engineer. This surprising fact surfaces a deeper crisis: a nation that is struggling to build. A16Z General Partner Aaron Price Wright hosts two founders who are using AI to reverse the decay of physical construction—at wildly different scales.
a16z has Aaron Price Wright with two founders aiming AI at the physical world — Alex Bowden building power plants and data centers, Davide Asnaghi making circuit boards in the U.S. faster than Asia. And the hook is wild: a Texas electrician now out-earns a Silicon Valley software engineer. Yeah, that number is basically the whole episode. The skilled-trades shortage got so bad the market repriced wiring a building above writing the app. The framing I actually liked: their bottleneck isn't intelligence, it's data. There's no Common Crawl for PCB schematics or construction blueprints, so both founders treat hardware more like code to generate the training set. That's the one real operating detail in here, and it's a good one. Everybody's screaming about humanoids folding laundry, and these two are quietly saying the moat is just — nobody digitized the blueprints. Solve the data, the model follows. My only flag — it's an a16z show, with two a16z-adjacent founders, and the salary stat is basically recruiting copy for the trades thesis. I want the manufacturing-cost number before I believe “cheaper than Asia.” Fair. Bowden and Asnaghi are working at completely different scales, so a shared premise only gets you so far — I'd queue it for the data-as-the-frontier argument, not as proof that either company ships at price. This one's from PodSized:
Brad Gerstner sits down with Gavin Baker and Andrew Fox of Atreides Management, alongside Altimeter partner Clark Tang, to break down one of the biggest questions in tech and markets: how should investors think about the SpaceX IPO?
So BG2 does the SpaceX IPO episode — Gerstner, Gurley, Gavin Baker, the whole bullpen. And the vibe note literally says “Intense & Bullish. The energy is electric.” Yeah, I bet it is. The actual structure's interesting, though — they break it into launch, Starlink, and then this “Elon Web Services” AI compute segment. Orbital compute as a line item in an IPO pitch. Here's my problem. SpaceX has been private forever, so what does an IPO at this scale even mean for a regular person who isn't already in the cap table? You're buying in after the founders, the early funds, and Altimeter have already eaten the good part of the curve. That's the BG2 tension, right there — Gerstner runs Altimeter, so he's framing the SpaceX story while having an obvious position in how it lands. I want the Clark Tang Taiwan supply-chain detail more than the bull narrative. Right, give me Fox breaking down launch-cadence economics, not five guys agreeing the future is electric. The operating number is in the chip supply chain, not the vibe. This one's from PodRocket:
Paige, Jack, Paul, and Noel break down the Cloudflare VoidZero acquisition that puts Vite and Evan You under the Cloudflare roof and what it means for open source sustainability when it seems like acquisition is the only exit. We also talk about Uber burning its annual AI coding tools budget in four months, the Claude Max token subsidy math, and whether local AI models can break the cost curve.
Three days I've been asking if anybody's gonna put a number on the cost curve. PodRocket just did. Uber burned through its entire annual AI coding-tools budget in four months. And that's the real-spend version of what Fortune flagged last month — using the tool costs more than the headcount it was supposed to replace. The panel's own bill is sixty grand a month in tokens. That's the bend point, and it shows up way earlier than any keynote slide promised. The piece I want is the Claude Max subsidy math — reports have Anthropic spending over a thousand bucks for every hundred you pay them. So that four-month Uber number is the bill at the discounted price. Right. Whatever Uber paid, the real cost was worse, and somebody's eating the difference until they decide not to. The other half is VoidZero going to Cloudflare — Vite and Evan You now under a corporate roof. And that's the part that bugs me. The open-source pitch was, you don't need a gatekeeper. Then the only exit on the board is getting acquired by the next infrastructure layer. Same re-intermediation, no chain required. Cloudflare frames it as funding the ecosystem. Maybe it is. But if acquisition is the only way Vite stays sustainable, that says a lot about who's actually funding open source. From Unchained:
Anthropic promised Mythos and shipped Claude Fable 5 instead. The model found a four-year-old bug in Zcash’s shielded pool that survived multiple expert audits. But when Anthropic shipped the model days later, it was no longer willing to audit smart contracts, bailing the moment a prompt smells like security work. Jailbreakers are already turning a jailbroken Opus 4.8 against it, while white hats sit locked out.
So Claude Fable 5 finds a four-year-old bug in Zcash's shielded pool — something that survived multiple expert audits — and then, days later, Anthropic ships it with the security work apparently surgically removed. That's the cleanest version of the liability paradox I've heard all week. The model just proved it can catch what humans missed for years, and now it bails the second a prompt smells like a smart-contract audit. And the asymmetry is brutal — white hats are locked out, but jailbreakers are already pointing a jailbroken Opus 4.8 right back at it. The capability didn't go away. Access did. Right, the defenders can't point it at their own code, but nobody can prove the black hats haven't. With North Korean actors harvesting API keys for six-plus months, the only people that guardrail actually stops are the ones trying to help. This ties straight into the token economics — frontier AI as a cost-and-access problem. Kain Warwick pulled five grand of compute out of a two-hundred-dollar plan, roughly 200 million tokens in four hours. The subsidy and the lockout are the same story from two ends. And nobody's holding the bag either way. When Claude refuses, the gap just sits there unowned. When it doesn't — ask Humanity Protocol, who lost their bridge, token, and treasury to one infected device. Got thoughts on today's stories, a correction, or an idea we should chase next? Send us a note at techpodcastpodcast at lantern podcasts dot com. We read every message, and your feedback helps shape the show.
What we're watching next: June 22 is the checkpoint, when Fable goes API-only and we see what frontier usage looks like without the 200-dollar-plan subsidy.
As always, we've put links to every story from today's show in the notes, so if something stuck with you, you can go straight to the source. That's Tech Podcast Podcast for today. This is a Lantern Podcast.