I started at a telecom startup in 1998. It was a great place to be – the engineering team was beyond sharp. The product they designed was ahead of its time, and the components they designed with were cutting edge. This isn’t meant as bragging as it was the only way to survive back then. Startups don’t have the luxury of playing it safe on components. Instead you design with what will (hopefully) be available in volume by the time you’re ready to start manufacturing. So you are betting on the curve, not sitting behind it.

We were acquired by a major networking company more than seven billion dollars. (If you were in telecom a few decades ago, you likely already know who I’m talking about. I’m leaving the names out deliberately. Not to protect anyone, but really because the names don’t matter. It is the pattern that matters.)

For the first year after being acquired we were pretty much left alone. The acquirer was smart enough to know it had paid a premium for something that was working, and they didn’t want to break it. But then their corporate policies kicked in. The one that stuck with me and that I still think about had to do with component selection.

At hardware or telecom startups, you can design with components that will be commercially available at manufacturing scale by the time your product is ready to ship. You are basically always working  ahead of the supply chain. JIT (Just In Time) manufacturing mattered, but this wasn’t recklessness. It was how you stayed competitive. You can’t design with yesterday’s components and expect to ship a product that holds up against the other products on the market when you finally are ready to ship.

The acquiring company had a different rule. They only allowed the engineers to design with components that were already available in high enough volumes for planned manufacturing at that moment. Not projected availability. Not components that would be ready in 12 to 18 months when the product would actually need them. You had to use quantities available right now for the basis of your design choices. Whatever was on the approved vendor list, in stock, today.

I understand the logic. When you’re manufacturing at the scale a company like that operates, you can’t afford to design a product around a component that turns out to be supply-constrained when you’re ready to build a million units. The procurement team had likely been burned before and the policy made sense from a risk management perspective.

But here’s what that policy actually produced: an internal development team that was structurally prevented from staying current. While the startup world was designing with what would be next, the internal team was locked into what was already proven and available in bulk. Every product cycle started a step behind. The gap between startups and enterprise firms continued to grow.

And here’s the part that should have been obvious to someone in a finance meeting somewhere: that same company, over the years that followed, spent enormous multiples acquiring other firms to stay relevant. Each acquisition was priced at a premium because the internal teams couldn’t keep up. The policy designed to protect manufacturing economics was directly generating the acquisition premiums they kept paying.

I’ve written about this dynamic before in a different context, about how large firms’ risk-averse vendor policies prevent them from leveraging differentiated technology, while startups run right past them using the very features those large firms are too cautious to touch. That procurement story from my early career is the same story. Sure, different input, but same output: institutional caution creates technical debt, technical debt creates competitive lag. That competitive lag gets solved by writing a big check to whatever startup didn’t have the same constraints.

I’m not writing this to relitigate decisions made twenty-five years ago. I’m writing it because I’ve been watching the same pattern repeat itself in real time, just with a different scarce resource.

The resource isn’t optical components this time, it is AI compute credits.

[IMAGE 4]

The structural tension that killed innovation in that telecom acquisition wasn’t malice or incompetence. It was a policy designed for the purpose of protecting manufacturing economics. And it had a predictable secondary effect no one accounted for. Large organizations are full of policies like this and each one makes sense in isolation. But together they create a thick layer of institutional conservatism that startups simply don’t have.

When I later wrote about why startups were eating large companies’ lunch on technology, the responses I got were always some version of “but the enterprise has stability, relationships, and scale.” True. They also have procurement committees, approved vendor lists, change management processes, and a dozen other friction points that slow the journey from “this is a good idea” to “we’re actually doing this.”

Startups have a different problem: survival. And survival turns out to be a remarkable forcing function for efficiency. When you don’t have excess budget, you don’t have room for waste. You figure out what actually works. And you better be fast, because the alternative is running out of runway.

That asymmetry, institutional caution on one side and survival-driven discipline on the other, is about to play out again in AI. I’ll cover the specifics of what’s happening right now in Part 2. But the setup is the same as it was in 1998, and if you were paying attention then, you already know how this ends.

The acquirer, eventually, buys whoever figured it out.

[Image 5]

This is Part 1 of a three-part series. Part 2 covers what tokenmaxxing actually was, why large organizations fell into it, and what it cost them. Part 3 looks at the overcorrection that’s already underway and why lean firms are positioned to win the next phase.

Photos from Unsplash

Photo by İsmail Enes Ayhan on Unsplash

Photo by Rafael Rex Felisilda on Unsplash

Photo by Homa Appliances on Unsplash