The six months that changed AI

In 2026, AI's biggest stories are no longer just about models. They're about everything it takes to bring those models into the real world.

In February 2026, Anthropic’s Claude Code, powered by the breakthrough Opus 4.6 model, had already begun to redefine software development with agents. Everyone seemed fascinated by OpenClaw, a personal AI agent which could do helpful tasks on your behalf — if it didn’t get hacked first. Chinese open-weight models were spreading fast; rural communities were pushing back against AI data centers; and AI was becoming a far more consequential cybersecurity concern.

At the time, these seemed like separate stories, in a sign that the AI beat had become impossibly broad. Who could possibly keep up? But six months later, it turns out they were all part of a major industry shift. While AI models remain headline-making news, you can no longer understand the model story without understanding everything around it, including the infrastructure, enterprise deployment, cybersecurity, and policy decisions shaping how AI is built, deployed, and governed.

I didn’t fully understand that, to be honest, when I launched Ground Level AI in June. It’s true that I intentionally focused on infrastructure, enterprise AI, cybersecurity, and geopolitics, because I found those areas fascinating and increasingly important. But I also worried I was trying to cover too much. Would someone want to read about these different topics on the same Substack? Would it be better if I launched, say, a newsletter devoted to AI security news, for example, or one squarely focused on the AI data center boom?

Now, however, I'm realizing these stories are all related, no matter where you enter the AI beat. That's because the first few years of the generative AI boom were about building better models. Now, as frontier models have become dramatically more capable, the challenge has shifted to all the messy stuff that comes afterward. None of it — building the infrastructure to support the models, deploying them inside enterprises, making sure they are secure, generating the compute and tackling governance — is easy or straightforward.

No AI story stands alone anymore

Take Anthropic's Claude, for example. Claude Code turned traditional AI model advances into a story about digital infrastructure and enterprise deployment, as developers and organizations worked to figure out how to run AI agents at scale. Anthropic’s Mythos model became a cybersecurity story when Anthropic held back its release over security concerns. And when Fable was taken down by the U.S. government, the conversation quickly shifted to geopolitics, sovereign AI, and the risks of relying too heavily on any single model provider. Suddenly, even enterprise model choice and Chinese open-weight models became part of the same story as AI model routers grew in popularity.

Or how about the AI data center boom? I originally thought of this as a physical infrastructure story—that is, how America's AI ambitions were reshaping towns and cities when a mega AI data center came to town. But data centers have also become inherently a geopolitical issue because compute has become a strategic national asset, and protecting AI infrastructure is increasingly a matter of national security. Data centers also exemplify the fierce competition among companies like OpenAI, Anthropic, Meta, Google, and Microsoft, who are all racing to generate the massive computing power needed to train and run frontier models. That’s an energy story. An economics story. And ultimately, an AI deployment story, because all of that compute exists to power the next generation of AI agents and systems.

The messy stuff becomes the story

To be clear, I’m not saying that this is all entirely new. Even when I first started covering AI in 2022, AI had already become a legal story with a spate of early copyright lawsuits. Generative AI quickly became a political story with Sam Altman testifying before Congress in mid-2023. And as soon as ChatGPT came out, enterprise companies were already working overtime to figure out how to use these tools to increase productivity.

But there's also no doubt that the parts of AI that once felt like secondary concerns—infrastructure, governance, policy, and security—have become central to the story over the past six months. The models and tools — ChatGPT, Claude, Gemini — still generate plenty of drama, along with their human company CEOs like Sam Altman and Dario Amodei. But increasingly, it's the messy work of deploying, securing, powering, and governing AI that's driving some of the industry's biggest stories.

When I launched Ground Level AI nearly seven weeks ago, I found myself worrying I was trying to cover too many seemingly unrelated stories. Did readers really want a data center story one day, then one on cybersecurity the next? A piece on open-weight models followed by one on enterprise AI adoption?

I think the answer is yes. The AI beat has matured in much the same way the industry has. Covering AI increasingly means covering infrastructure, security, enterprise adoption, politics, and economics—not as separate beats, but as different parts of the same evolving story: how AI is meeting the real world.

In case you missed it