The Wire Weekly
Episode 11 · June 29, 2026 · 00:08:17.14

This Week's Stories

Full Transcript

Gina Raimondo handed Intel eight billion dollars of your money to build chip factories in America. Then she quit her job to raise half a billion dollars to retrain the workers Intel just fired. I'm not saying there's irony here... but if you squint hard enough, you can see the entire economic policy of the last three years crystallized into one perfect little circle of hell. I'm Travis Wright, this is The Wire Weekly, and this week the pattern is crystal clear: the era of bigger-is-better just hit a wall, and the smart money is sprinting in the opposite direction. Let's start with the biggest move of the week, and it's not about size at all. It's about efficiency. Liquid AI just dropped a model called LFM2.5-230M. That's 230 million parameters. For context, GPT-4 is estimated to have over a TRILLION parameters. Liquid's model is a rounding error compared to the frontier models everyone's obsessing over. And yet... it's beating models four times its size at real-world tasks like data extraction and document processing. And here's the kicker: it runs ANYWHERE. Your laptop. Your phone. An edge device in a warehouse. No cloud required. This matters more than almost anything else I'm covering today. The entire AI narrative for the last two years has been about scale. Bigger models, bigger clusters, bigger energy bills. OpenAI burned billions getting to GPT-4. Anthropic raised billions to chase Claude 3.5. The assumption baked into every pitch deck and every VC memo has been that you need massive compute and massive capital to compete. Liquid AI just said... nah. We're good with 230 million parameters and a cleverly designed architecture. This is what actual innovation looks like when you stop throwing money at the problem and start thinking about the problem differently. Liquid's using what they call "liquid neural networks," which are inspired by biological neurons and designed to be efficient by default. They're not trying to win the benchmarks that OpenAI and Google set up to showcase their own models. They're building for deployment. For real use cases. For businesses that don't want to send every invoice and receipt to the cloud for processing. And the timing here is NOT a coincidence. Because while Liquid is proving that small can beat big, Mistral AI in Europe just hit 400 million euros in annual recurring revenue in under three years. That's not billion with a B. That's 400 million in actual revenue. Not valuation. Not funding rounds. REVENUE. Meanwhile, OpenAI is burning cash so fast they had to restructure as a for-profit just to keep raising. Mistral's out here proving that efficiency and focus and actually selling a product people want beats the frontier narrative every single time. So here's the thread I want you to see this week. Three different stories. Three very different companies. Same exact pattern. First, Liquid AI with their tiny 230-million-parameter model beating billion-parameter rivals. Second, Mistral hitting 400 million in revenue while staying lean and European and decidedly NOT playing the Valley's scale-at-all-costs game. And third... Nvidia's building data centers in Indonesia. Wait, what? Yeah. While every AI company in America is fighting over limited power and rack space in Northern Virginia and Oregon, Nvidia's partner Firmus Technologies just broke ground on a massive AI infrastructure project 3,000 miles south of Singapore. Indonesia. Not exactly the first place you think of when you hear "AI superpower." But here's what Indonesia HAS: cheap energy, available land, a government desperate to be part of the AI economy, and zero NIMBYs filing lawsuits about water usage and grid capacity. This is the same playbook we've seen in manufacturing, in chip production, in basically every other infrastructure wave of the last 30 years. The West hits capacity. Costs skyrocket. Regulations pile up. And the action moves somewhere hungry and ready. Nvidia's not building in Indonesia because they love the Jakarta skyline. They're building there because the West is tapped out and the next phase of AI compute is going to happen where the power and land and political will actually EXIST. So let me connect the dots. You've got models getting smaller and more efficient. You've got European companies proving you don't need to burn billions to build a real business. And you've got infrastructure moving to places where growth is still physically possible. The pattern is this: the era of "just scale it" is over. The era of "build it smart" is here. Alright, wild card time. Let's talk about China minting two robotics unicorns in the same week. Shenzhen Ubtech and Fourier Intelligence both hit valuations over a billion dollars in recent funding rounds. These are companies building humanoid robots. Not demos. Not research projects. Actual commercial robots that work in factories and warehouses and elder care facilities. And while Silicon Valley is still debating whether humanoid robots are even a good idea, China just quietly deployed thousands of them and raised billions to deploy thousands more. Here's what I love about this. The Valley spent the last two years convinced that the future of AI is chatbots and code assistants. Meanwhile, China looked at the same technology and said cool, let's put it in a body and make it DO something. And the funding is following. $2.9 billion in combined valuations for two companies most people outside the robotics world have never heard of. Compare that to Figure AI and Tesla's Optimus, which get 90% of the press and a fraction of the deployed units. And speaking of doing something... General Intuition just raised $320 million to train robots using Fortnite gameplay. Yes, you heard that right. They're taking millions of hours of video game replays, human players navigating complex 3D environments, making decisions, coordinating with teammates, and using that data to teach AI agents how to move and interact in the real world. Because it turns out that the mechanics of building a fort in Fortnite and the mechanics of navigating a warehouse are... kind of similar. Pattern recognition, spatial reasoning, real-time adaptation. It's all there. This is my favorite kind of weird smart. It sounds absurd until you think about it for 30 seconds, and then it's the most obvious thing in the world. Video games are basically simulation engines that millions of people voluntarily generate training data for. Why WOULDN'T you use that to train embodied AI? It's brilliant. And the fact that they raised $320 million means a lot of very serious people agree. Okay. Three things to watch in the next few weeks. First, Securitize is closing a $400 million SPAC merger and going public on the NYSE July 2nd. This is the company that tokenized BlackRock's money market fund. If you've been waiting for real-world asset tokenization to move from "interesting idea" to "actual tradable securities on major exchanges," this is that moment. BlackRock's already in the cap table. The infrastructure is live. This is not a pilot project anymore. Second, watch the backlash to the intel situation. Gina Raimondo's new fund isn't technically illegal, but it FEELS like a revolving door on steroids. You give a company $8 billion in taxpayer money, they lay off thousands of workers, and then you leave government to raise money to retrain those workers for an economy that may or may not need them. Senators Warren and Scanlon are already moving to regulate AI health data sales. I would not be shocked if someone introduces a bill targeting this kind of post-government profit-taking. The optics are just too bad. And third, pay attention to how many more "small model" announcements we get in the next quarter. Liquid AI just proved that efficiency beats scale for a huge range of tasks. Mistral proved you can build a massive revenue business without burning billions. If I'm running an AI startup right now, I'm not trying to out-GPT OpenAI. I'm trying to out-efficient everyone and go to market with something that actually runs on hardware people already own. The next wave is not about who has the biggest cluster. It's about who ships the smartest product. That's the world we're living in. Bigger isn't better anymore. Efficient is better. Deployed is better. Revenue is better. And if you're still betting on the frontier model narrative, you're fighting the last war. I'm Travis Wright. This is The Wire Weekly. I'll see you next Tuesday.