AI valuations are now approaching infrastructure-scale pricing

The companies at the center of AI race are starting to look less like software firms and more like a mix of utilities, defense contractors, and financial assets. The amount of money flowing into AI infrastructure right now is enormous, and honestly, parts of the market are beginning to feel overheated.

The biggest story this week was Anthropic’s new funding round. According to Reuters, the company raised funding at a valuation reportedly approaching $965 billion — a number that would place it near the most valuable public companies in the world on paper. - Anthropic Valued at 965B

That number is staggering when you think about it. Anthropic is still essentially an AI model company selling enterprise access and APIs, yet investors are now valuing it like foundational global infrastructure. At this pace, AI startups are skipping the awkward “profitable business” phase entirely and heading straight to “strategic civilization asset.”

This says a lot about where investor psychology is right now.

Wall Street is no longer valuing AI companies like software startups. Investors are betting that a few dominant AI firms could eventually control the next layer of the digital economy — similar to how Google controlled search or how AWS became core internet infrastructure.

Whether those expectations are realistic is another question entirely.

The other thing that stood out this week is how tightly AI is becoming tied to governments and national security.

The Pentagon’s ongoing AI partnerships and negotiations with companies like OpenAI, Microsoft, Google, and others show that governments now see frontier AI models as strategic infrastructure. AI is no longer just a commercial product; it’s increasingly being treated like energy infrastructure or semiconductor manufacturing.

There’s also growing tension underneath the surface. Some AI companies want defense contracts because the money and influence are enormous. Others are more cautious about military use cases. Anthropic, in particular, has reportedly been more hesitant about unrestricted military deployment of its models — which, in Silicon Valley terms, now qualifies as being “moderately conservative.”

At the same time, the U.S.–China AI competition keeps escalating.

A lot of the real battle is now happening around semiconductors rather than chatbots. Export controls on advanced AI chips continue tightening because governments understand that whoever controls compute controls AI development.

And that brings us to the semiconductor companies — which honestly still look like the biggest winners in the entire AI boom.

NVIDIA remains at the center of everything. Every major AI company still depends heavily on Nvidia GPUs. But this week also showed that large AI firms are trying very hard to reduce that dependence.

Google is expanding TPU deployments. Microsoft is working on custom AI chips. Broadcom has quietly become one of the most important players in AI infrastructure through custom accelerator partnerships.

What’s interesting is that the market narrative is slowly shifting.

A year ago everyone was talking about chatbots. Then it became AI agents. Now the conversation is increasingly about infrastructure — chips, power, cooling systems, cloud capacity, and data centers. AI discussions now sound less like software conferences and more like industrial engineering meetings with unusually high stock valuations.

That shift matters because infrastructure businesses often end up making more durable money than application-layer software companies.

But this week also exposed something else that people in the industry are starting to whisper about more openly: overbuilding.

The AI infrastructure buildout is happening at incredible speed. Massive GPU clusters, new data centers, power agreements, and cloud expansion projects are being announced constantly.

The assumption behind all of this spending is that AI demand will keep growing exponentially.

Maybe it will.

But there are now early signs that some companies want flexibility in case demand doesn’t fully match expectations. Reuters reported this week that Elon Musk clarified SpaceX had not committed to a long-term lease arrangement tied to AI infrastructure expectations the way earlier reports suggested. - Musk says SpaceX not committed to long-term lease with Anthropic

That might sound like a small detail, but it’s important.

It suggests that even major players are trying to avoid locking themselves into massive long-duration infrastructure bets too aggressively.

That’s usually what happens when markets start getting nervous about oversupply risk.

The stock market still loves the AI trade for now. Semiconductor companies continue looking strongest because they’re making money regardless of which AI model eventually wins.

NVIDIA still looks dominant. Broadcom keeps gaining importance. AMD and Micron Technology continue benefiting from the demand cycle. Investors still see multi-year AI infrastructure spending ahead.

But the easy phase of the AI rally may be ending.

The market is starting to ask harder questions:

Right now, the industry feels like a mix of genuine technological transformation and classic late-cycle market euphoria.

Both things can be true at the same time.

The most likely outcome probably isn’t an AI collapse. It’s consolidation. A small number of companies could become extraordinarily powerful while many smaller AI startups disappear under infrastructure costs and competition pressure.

And increasingly, the real power may belong less to the chatbot companies and more to the firms supplying the chips, compute, and energy behind them

- - -