Co-Author: Tony Greenberg
For close to a quarter century, the tech savvy and future-positive crowd has been waiting for innovation that felt meaningful. Not just another way to part us from our attention, shared reality, and money as Web2 and Web3 did. Not another niche app, digital addiction, or speculative asset. But real moonshot change, like the steam engine and electricity and space travel, and the Internet and, maybe someday, the flying car.
Now we finally have that breakthrough in generative AI. The markets are sustained on the back of the Magnificent 7. The literati are seriously debating a future of AGI singularity, and whether that will make us superhuman or extinct. And everyone in the general vicinity of the boom / bubble is eager to build fast.
Billions in capital are being poured not just into the pockets of talented devs that catch Mark Zuckerberg’s eye, but the humble plumbing that we have stayed with in the long interregnum between dotcom and AI booms when it was far from cool. This includes data centers and networks, resilient power & cooling infrastructure, and all the smaller puzzle pieces needed behind the scenes to solve the world’s grand challenges let industrious high schoolers cheat on their assignments.
A lot of this spending is being done in gold-rush fashion – staking grand claims in hopes of a first mover advantage. But as old fogeys who have been there through the first revolution, experience teaches us that it’s a marathon, not a sprint, and being first is not always what it’s cracked up to be. As they say, the early bird gets the worm, but the second mouse gets the cheese.
The winners of the first data center gold rush were not the original facilities builders like Exodus that I was part of. They were the vultures that swooped in and bought those data centers and networks for pennies on the dollar in the decade-long gap between pets.com superbowl ads and the YouTube and Netflix era. Given the hasty growth and exorbitant valuations of AI-driven data center projects, history is likely to repeat itself, as overleveraged early movers are overtaken by a more methodical and innovative second generation.
If you’re in the AI infrastructure game, think less about being first and more about building sustainably for the long haul, conserving capital, innovating, and future proofing. Even if AI changes everything eventually, it won’t
necessarily do it on the schedule prescribed by analysts’ hockey stick forecasts. It will be far messier, with troughs of disillusionment still to be survived before we reach the steady state.
Being able to weather disruptions (e.g. another DeepSeek-like innovation that cuts the need for GPUs and power) or disappointment (e.g. the “low-background steel” problem of AI models increasingly training on an internet polluted by their own outputs) is likely to be far more important than getting in the door first with overpriced power deals and engineering that’s already obsolete by the time it goes live.
I know it’s tough to resist the siren song of venture wallets loosening up after 3 years of entrepreneurial starvation. It’s hard to remain on the sidelines or cautious when continuous evolution is disrupted by a true paradigm shift. But as someone who got in on the ground floor of both the boom and the bust of Web 1.0, I’d rather not be the first mouse (or, to harken back to a meme from just about that ancient age, the second deer).
