AI’s New Battleground: The Fight for Compute and Custom Chips in 2026

Anthropic is negotiating with Samsung to build its own custom AI chips, and it's preparing for an IPO as early as October. In episode 108, we tracked the major company splits and realignments changing the game. This is the next move on that same board: securing the physical supply chain. Because what's becoming clear is that the biggest constraint in AI right now isn't a lack of ideas. It’s a lack of hardware. The fight for compute is now THE central story. This is the clearest signal yet that compute, not cleverness, is the real bottleneck. Anthropic isn't just trying to get faster. It’s trying to get predictable revenue and reduce its operational risk before going public.

It's a strategic shift away from total reliance on Nvidia and rented cloud capacity. They want to own their destiny, and destiny, it turns out, is forged in silicon. And they are NOT alone. Meta just announced it's doubling its AI compute capacity. This isn't a small upgrade. This is a ten-billion-dollar data center in Alberta, Canada, plus new supply deals with Samsung. When a company like Meta spends that kind of money, it's not an experiment. It's a declaration. The race for model dominance has become a race for industrial-scale infrastructure. The winners won't just have the best algorithms. They'll have the most GPUs. So while the giants are buying up chips by the boatload, a different kind of race is happening on the software side.

This is about what you actually DO with all that power. A company called Nous Research just raised funding at a one-point-five billion dollar valuation. That number is less important than what they build: an open-source autonomous agent platform called Hermes. And this isn't just investor hype. Pay attention to the developer numbers. Hermes has over two hundred and fourteen thousand stars on GitHub. Nearly forty thousand forks. That is massive, organic developer adoption. It shows you where the energy is going. It's moving beyond simple chatbots and into AI agents that can operate continuously, on their own, to complete tasks for you. We're talking about a fundamental shift from AI that answers to AI that acts.

That is the promise that just got a billion-and-a-half-dollar vote of confidence. Okay, so we have more compute and smarter agents. But how is this actually changing the way people work, right now? The AI Engineer World's Fair just gave us the raw numbers. And they are STAGGERING. As of today, twenty-seven-point-six percent of all merged pull requests on GitHub are AI-generated. Let that sink in. More than a quarter of the code being shipped is written, at least in part, by a machine. This isn't a future trend. It's a present-day reality. But here's the turn. Here's the part that nobody is putting in the press release. Only forty-eight percent of developers say they always verify that AI-assisted code before committing it.

Less than half. We are building the future on code that we are not consistently checking. This is the core challenge. As one report put it, enterprises don't fail at AI because the models are weak. They fail because nobody inside the building can wire the model into decades of messy workflow. The new frontier isn't just building AI. It's "Harness Engineering"—the discipline of safely and reliably plugging this incredible power into the real world. That's the friction. That’s the work. We now have robots like Boston Dynamics' Spot getting a Gemini brain. It's a genuine milestone, not a stunt. The robotics and frontier-model races are merging. But navigation and reasoning are still being solved separately.

You have this incredible brain, but it's still learning how to reliably walk and chew gum at the same time. So the landscape that’s emerging is one of immense power and immense risk. The hardware is scaling at an unbelievable rate. But our ability to integrate it safely is what will define the next year. It brings to mind what Dr. Fanny Elahi said at the Aspen Ideas Festival. "Don't resist AI—be intentional with it." Some tasks our brains no longer need to carry. Others, we must consciously protect. That’s the real work ahead. The bottleneck is no longer invention. It’s implementation.