AI's $600B Reality Check: Beyond the Bubble Hype

The big four hyperscalers just spent six hundred billion dollars on AI infrastructure in 2025. This wasn't funded with debt or speculative venture capital—it was paid for entirely with their own operating cash flow. Last time we talked about AI models evolving. Now we know what that evolution costs. The conversation on Twitter this week wasn't about whether this is a bubble. It was a brutal takedown of the entire bubble thesis. The analyst firm Dick Capital laid out the argument: this spending is rational. It represents less than one percent of global GDP, far below the peak of previous tech build-outs like telecom in the nineties.

And unlike the telecom bubble, which built fiber for a future that wasn't there yet, this AI infrastructure is being deployed to meet demand that is already here. Demand that is paying, and growing faster than any software transition in history. So the money is real, and the need is real. The question is what happens when all this digital intelligence starts moving in the physical world. That’s where the conversation splits. On one side, you have the defense sector. A thread from the Seoul Economic Daily highlights that physical AI for autonomous combat is nearing deployment. But it’s running straight into Moravec’s Paradox.

The things that are hard for AI—like navigating rough terrain or keeping a signal in a chaotic environment—are the very definition of a battlefield. Mission failure isn’t a server crash. It’s a loss of life. On the other side, you have industry. Agile Robots just announced a partnership with Google DeepMind. They’re integrating the Gemini model into over twenty thousand deployed robots. This isn't just an upgrade. This is about creating robots that learn and adapt in real time, building a feedback loop from the factory floor back to the model. So we have two paths for physical AI emerging simultaneously: the factory and the battlefield.

And that brings us to the real fight. Anthropic is now suing the U.S. Department of Defense. The DoD designated Anthropic a supply chain risk because the company refuses to allow its models to be used for autonomous weapons. The DoD’s legal brief is the part that stops you cold. It says they can’t trust Anthropic because warfare requires “privileged access” and “constant tuning” of the AI models. Anthropic’s response is that you can’t just “tune” a deployed model like that. It reveals a fundamental conflict. The military believes it needs total control to alter an AI’s core logic in the field.

The AI’s creator says that’s not how the technology works, and they won't build a backdoor for it. This isn't a philosophical debate anymore. The six-hundred-billion-dollar build-out is forcing the issue. All that compute is powering AI that will either be assembling cars or targeting objectives. The argument over which path we take—and who holds the keys—is now happening in a federal courthouse. The conflict isn't about whether the AI is smart enough. It’s about who gets to tell it what to do.