0:00 NVIDIA just signaled it's moving into the revenue-sharing business for AI.0:05 This isn't just a new pricing plan; it's a fundamental shift in how access to large-scale AI is bought and sold.0:12 Last week, in episode 100, we landed on how national sovereignty was redefining the AI race, with countries building their own compute.0:22 Now, NVIDIA is redefining the economics of that race for everyone who ISN't a nation-state.0:28 It's a holiday weekend.0:29 The timeline is… quiet.0:31 But the quiet is where you find the signal.0:34 The real work isn't in the hot takes; it's in the foundational shifts that are happening just beneath the surface.0:42 And the threads still echoing from yesterday are all pointing in the same direction.0:48 It’s a story about the industrialization of artificial intelligence.0:52 So while NVIDIA is figuring out how to build the power plants for this new era, the sharpest conversation on the software side is about what to actually DO with all that power.1:04 Greg Isenberg dropped a thread that cuts right through the noise about AI killing software.1:11 His thesis is the opposite.1:13 Agents are not the death of Software-as-a-Service.1:16 They are the next evolution of it.1:18 Think about that for a second.1:20 SaaS, as you know it, gives you tools.1:23 It gives you a dashboard.1:25 You click the buttons, you pull the levers, you interpret the data.1:29 Isenberg’s point is that the next step is not a better dashboard.1:34 It's an agent that both builds the exact dashboard you need for a specific task, and then goes and fulfills the work represented on that dashboard.1:44 The agent becomes both the tool-maker and the worker.1:48 This isn't some far-off prediction.1:50 He's putting his money where his mouth is.1:53 His direct quote was: "If I was starting something new, I’d start an agent business.1:59 And I am.1:59 And I’d invest millions of my own dollars doing it." That’s not speculation.2:05 That’s conviction.2:06 He argues the defensibility in this new world isn't just code.2:10 It’s the classic stuff: distribution, network effects, proprietary data moats, and the raw advantage of being the first one to build a trusted system.2:20 He sees a future where you don't just subscribe to a service; you hire an agent.2:26 And that changes the entire software economy.2:29 Of course, a thesis is one thing.2:31 Seeing it happen is another.2:33 And this is where you have to look past the big names and into the developer chatter.2:39 This isn't just a theoretical debate between venture capitalists.2:44 It's already happening in the code.2:46 The name to watch right now is OpenClaw.2:49 The discussion around it has shifted.2:51 The key phrase being passed around is "rapid practical iteration." That's developer-speak for "it actually works." We're seeing long-running, semi-unsupervised agent tasks inside large codebases move from a purely academic concept to a daily reality.3:08 This means AI agents are being deployed to manage, debug, and negotiate with complex software systems with minimal human oversight.3:17 The whispers are about measurable workflow wins, especially in areas like customer service and system-to-system negotiation.3:26 An agent that can handle a customer support queue or manage API integrations isn't a research paper anymore.3:33 It's a line item.3:35 It's an employee.3:36 And when you see that happening, you look for the next signal.3:40 The second-order effect.3:42 And here it is: Peter Steinberger is building something called "crabhelm." Now, that name might not mean anything to you.3:50 But what it is matters.3:52 It's a dedicated orchestration layer for agent workflows.3:56 Let me be very clear.3:57 You do not build orchestration tools for things that are simple, or rare, or experimental.4:03 You build orchestration when you have a complex, growing, and critical system that is becoming unmanageable.4:11 You build it when you have swarms of agents that need to be deployed, monitored, and coordinated like a factory assembly line.4:19 The very existence of a project like crabhelm is a screaming indicator that the agent economy is getting real, and getting complicated, FAST.4:29 It’s the plumbing and wiring being laid for the new high-rises.4:33 People don't pour that kind of concrete foundation for a tent.4:38 Which brings us all the way back to NVIDIA.4:41 Back to the power plant.4:42 Their announcement wasn't just a press release.4:45 It was a flare sent up to signal the start of a completely new market, built for the world Isenberg and Steinberger are creating.4:54 You have to read the fine print.4:57 The key phrase from NVIDIA was the market's shift from "model training" to "always-on token production." That is EVERYTHING.5:05 Model training is a massive, one-time capital expense.5:09 It's like building the factory.5:11 It takes a huge amount of power and resources, but it's a discrete project.5:16 You finish, and you have an asset.5:18 But "always-on token production"...5:21 that's different.5:22 That's not a project.5:23 That is a UTILITY.5:25 That's the factory running 24/7, churning out tokens, which are the atomic unit of AI work.5:31 It’s a constant, operational expense.5:34 It's the electricity bill.5:35 And how do you sell electricity?5:38 You don't sell people a whole power plant.5:40 You sell it by the kilowatt-hour.5:43 You make it accessible.5:44 You help them finance the appliances that use it.5:48 Now look at NVIDIA's plan again.5:50 They are partnering with AI clouds to deploy what they call "AI factories." And they're doing it through revenue-sharing and credit-support.6:00 This is the move.6:01 They are not just selling you a box of chips anymore.6:04 They are becoming the bank and the utility for the entire AI economy.6:09 They will finance your access to their compute, and in return, they will take a CUT of the revenue that your AI agents generate.6:18 They are building the financial plumbing that allows a developer with a great idea for an agent to get started without needing ten million dollars for hardware upfront.6:29 They are underwriting the agent economy.6:32 So what does it all add up to?6:34 While the surface of Twitter is quiet for the holiday, the tectonic plates of the AI industry are grinding against each other.6:43 We're seeing the birth of a new, full stack, from the silicon all the way to the business model.6:50 NVIDIA is building the utility grid.6:52 Developers like Peter Steinberger are building the circuit breakers and the control panels.6:58 And entrepreneurs like Greg Isenberg are designing the businesses that will run on it.7:04 The game is no longer just about who has the smartest model.7:08 That's table stakes.7:10 The new game is about who has the most efficient, scalable, and profitable AI factory.7:16 The conversations that mattered this week weren't loud.7:19 They were about infrastructure.7:22 About business models.7:23 About the boring, essential work of turning a magical technology into a predictable industry.7:30 The debate is moving on from what AI can know.7:33 The real question now is what AI can DO.7:35 And more importantly, what it costs to get the job done.7:39 Next week, we'll be watching to see who announces the first major deal under NVIDIA's new model.7:46 Because that's when the race truly begins.7:49 It's not about the code anymore.7:51 It's about operations.
