The Wire Weekly
Episode 5 · May 4, 2026 · 00:07:52.29

This Week's Stories

Full Transcript

Katie Haun wrote the first check to Coinbase when every regulator in America thought crypto was a scam. Now she just raised a billion dollars to fund AI agents that own wallets, trade assets, and move money without asking permission. Not AI that HELPS you trade. AI that trades. We've officially crossed the threshold where software doesn't need you anymore... it just needs your capital. I'm Travis Wright, this is The Wire Weekly, and if you thought crypto was moving fast, wait until you see what happens when the bots start holding the bags. Alright, let's talk about Haun's move because this is way bigger than a funding round. Katie Haun's firm just closed a billion dollars across new funds, and here's the part that matters. She's expanding from crypto infrastructure into what she's calling the agentic economy. That's the polite Silicon Valley term for AI systems that own stuff, make decisions, and execute transactions autonomously. We're not talking about Siri reminding you to buy milk. We're talking about software entities with wallets, with capital allocation strategies, with the ability to enter into contracts and settle trades faster than any human ever could. This is the same person who saw Coinbase when it was just Brian Armstrong and a regulatory nightmare, and turned that bet into one of the defining venture wins of the last decade. So when she pivots hard into AI agents holding crypto, you better believe every founder in San Francisco is rewriting their pitch deck right now. The thesis here is simple but profound. Crypto gave us programmable money. AI gave us programmable intelligence. Put them together and you get economic actors that don't sleep, don't have emotions, and can move at machine speed. We've been talking about this for months on the show, but now the capital is real. A billion dollars real. And that changes everything. Now here's where it gets really interesting, because three other stories dropped this week that all point to the same future. Nvidia just invested in DeepInfra, a company trying to commoditize AI inference and crush Nvidia's own margins. Microsoft built an AI agent that reads legal contracts like an actual lawyer, not like ChatGPT pretending to be one. And Meta just admitted that ninety percent of their AI model is basically dead weight, zeros doing nothing, wasting compute on parameters that don't matter. You see the pattern? The AI infrastructure wars are heating up, the agent layer is getting REAL capabilities, and the entire industry is waking up to the fact that bigger doesn't always mean better. Let's break it down. Nvidia writing a check to DeepInfra is absolutely wild. DeepInfra just raised a hundred and seven million in a Series B led by Nvidia and Samsung. Their whole pitch is optimizing inference so you can run AI workloads cheaper and faster than the big cloud providers. Nvidia makes its money selling chips at insane margins. DeepInfra wants to make those workloads so efficient that the margin pressure forces everyone to rethink pricing. And Nvidia FUNDED them. That's not a hedge, that's a signal. Nvidia sees the future where inference becomes a commodity and they want a seat at that table even if it cannibalizes their current business. Meanwhile, Microsoft just launched Legal Agent inside Word. This isn't a chatbot that summarizes documents. This is an AI that tracks negotiation history, flags risky clauses, and handles edits the way a junior associate actually works. It's purpose built. It's vertical. And it's a direct shot at every legal tech startup that thought general models would be good enough. They won't. And then there's Meta admitting that most of their massive two trillion parameter Llama model is... basically padding. Sparsity. Zeros. Parameters that contribute nothing but still burn energy and compute. The race to bigger models is hitting a wall, and the next wave is about efficiency, not size. So put it all together. Haun is funding the agents. Nvidia is funding the infrastructure that makes agents cheap to run. Microsoft is building the vertical tools that make agents useful in specific domains. And Meta is figuring out how to make models smaller and smarter. This isn't three random stories. This is the entire AI stack getting rebuilt in real time. Okay, now for the story that made me laugh out loud this week. Parallel Web Systems just hit a two billion dollar valuation five months after their LAST funding round. They raised a hundred million dollars at that valuation led by Sequoia. And here's the kicker. The founder's previous job? He was at a marquee company everyone knows. The product? Still being built. The traction? Unclear. The valuation? Two BILLION. This is the purest distillation of founder reputation as currency that I've ever seen. I'm not saying it's a bad investment. Maybe Sequoia knows something we don't. But let's be honest here. The fastest path to a decacorn isn't building product. It's building product while everyone still remembers your last job title. Five months. That's less time than it takes most startups to close a seed round. And they're sitting at unicorn times two. If you're a founder grinding it out without the pedigree, without the network, without the Sequoia speed dial... this one stings a little, doesn't it? But hey, that's the game. And in this market, narrative beats numbers every single time. So what should you be watching this week? Three things. First, keep your eyes on the Anthropic funding drama. They're considering offers that would value the company at over nine hundred billion dollars. That's not a typo. NINE HUNDRED BILLION. They're burning two point seven billion a year and they're positioning themselves as the safety first alternative to OpenAI. If that deal closes, it resets the entire valuation bar for AI companies. Second, watch what happens with China and AI infrastructure deals. They just blocked Meta's two billion dollar acquisition of Manus, a robotics AI company. This isn't about antitrust. This is about China deciding that AI infrastructure has a nationality, and if you build the rails that agents run on, you're a strategic asset. Expect more blocks, more scrutiny, and more fracturing of the global AI supply chain. And third, pay attention to the security conversation in crypto. Mythos just got breached and the entire industry is realizing that all the best practices we've been following, multisig, cold storage, audits, might not be enough anymore. The threat models are evolving faster than the defenses. If you're holding assets, managing a treasury, or building infrastructure, this is your wake up call. Oh, and one more thing. Goldman Sachs just banned Claude AI for their Hong Kong staff. Not globally. Just Hong Kong. That's not a tech decision, that's a geopolitical one. The AI decoupling is happening inside banks now, and if you think that won't affect how capital flows and where innovation happens, you're not paying attention. Here's the thing. We're watching two massive shifts collide in real time. AI is getting autonomous. Crypto is getting institutional. And the people with the capital, the Hauns, the Sequoias, the Nvidias, they're all betting that the future belongs to agents that own assets and make decisions without waiting for human approval. That's exciting as hell. It's also a little terrifying. Because once you hand the keys to the bots, you better be damn sure you built the right guardrails. Mythos learned that the hard way. Goldman is learning it right now. And the rest of us? We're about to find out. I'm Travis Wright. This is The Wire Weekly. See you next Tuesday.