As a CEO who has built AI-powered supply chain platforms from the ground up, I’ve sat in the trenches with my engineers and watched a single five-minute AI agent interaction cost over $5,000 in compute tokens. I’m here to tell you: the AI arms race isn’t being run on a level playing field, and almost nobody is talking about what that means for the rest of us.

The token cost wall is real — and it’s shutting the door on innovators.
One five-minute AI agent session on a limited data set can cost a mid-size company over $5,000 in compute tokens. This isn’t a prototype problem — it’s the everyday cost of doing AI business. The only companies that can absorb this without blinking are Meta, Google, Amazon, and Musk’s Grok. Everyone else is being priced out of the race before it even begins.

Amazon is not giving you “free” infrastructure — they’re buying your future.
When Amazon Web Services offers to build your entire data modernization infrastructure for free — work that a trusted contractor quoted at over $500,000 — that’s not generosity. AWS subcontractors build lock-in so deep that leaving later costs more than what you saved upfront. For $20 million revenue companies, you can’t say no. That’s how Amazon is quietly owning the AI futures of thousands of businesses.

The stock market is propped up on AI froth — and the fundamentals will crack it.
Wall Street is asking the wrong questions. Instead of debating which AI company wins the quarter, investors should be asking: how will these companies afford the token costs required to deliver on everything they’ve promised? The compute bills at the scale they’re describing don’t pencil out — and when that math catches up, it will look a lot like the junk bond collapse. People are buying the promise, not the business model.

It’s Facebook 2008 all over again — they’re not telling you what they’re building
with your data.
Just as Facebook built a social playground without disclosing the data extraction happening underneath, AI companies are moving at full speed without transparency about what they’re collecting, how they’re using compute, or what the long-term cost to the user ecosystem will be. We’ve already lived through this story. We know how it ends.

There IS a way for smaller companies to compete — but it requires a walk-before-you-run discipline.
The path forward for businesses that aren’t in the Big Four is radical transparency with clients, surgical AI implementation focused on the highest-value use cases, and being ruthlessly selective about which problems you actually need AI to solve. Add as much value as you can without breaking the bank. Most companies trying to ‘do AI’ are sprinting before they can walk — and they’ll be the ones holding the bag.