The $3 Trillion Server Farm Dream
Somewhere between a chip shortage and a collective tech hallucination, the world decided to spend three trillion dollars building datacentres.
Three. Trillion. Dollars.
It sounds nonsensical, and almost certainly it is.
When you’re funneling more than the GDP of the UK into windowless warehouses with rapidly aging NVidia GPUs something is going wrong.
The Industrial GPU Revolution
The potential to power the artificial intelligence revolution is too much for most. The risk is that it will end like every over-leveraged, over-promised tech fantasy before. Anyone with an eye on history will know how this story ends, and many will end up losing their shirt in the carnage which will result.
Datacentres are today’s coal mines. They’re powering the new Industrial Revolution with vast quantities of electricity and questionable optimism which speaks more to “it works on my machine” than a real revolution.
Computing’s great advantage has never the massive investment. It’s never been the potential. It’s been the idea that a skunkworks project can get going with nothing more than an idea and a laptop. This particular revolution is seeing that go away, and as we lose the foundation of the industry perhaps it will never come back.
The $3 Trillion Build, the $1.5 Trillion IOU
That’s one issue. Another is the precise origins of this overblown spending. Of the projected $3 trillion in global data centre spending only around half will come from cash flow while the rest will be borrowed (that means banks and private credit firms).
It’s an echo of 2008 where throwing money around with no real expectation of payback blew up in rather spectacular fashion. You’d be well advised to stay clear of this chaos if you can, as the results are seldom anything other than messy.
FYI, Meta alone has borrowed $29 billion for datacentre expansion. That’s a lot of servers for a company best known for sending your aunt “People You May Know” notifications after having a private chat about them within earshot of their mobile.
The Legacy
After the dotcom bubble burst we were left with thousands of miles of cabling, leaving a worthy infrastructure behind. Worth the pain? Well, no, but the bubble did leave something more than a brutal explosion that destroyed the chance of retirement for many.
This bubble won’t do the same. That’s because data centres depreciate twice as fast as the revenue they bring in. Hardware ages faster than the business models built on it, and those loans still need to be paid back.
It’s like tech debt accumulating faster than features can be shipped. Like our monolith decomposition project where we move some code out of the monolith while at the same time we ship more production code into the monolith. Bad times, I know.
Leveraged Purchase of NFTs
Microsoft, Google, Meta, Amazon, NVidia (of course) are all in. Their quarterly reports are euphoric. Nvidia hit a $5 trillion valuation. OpenAI is valued at $500 billion. Apple and Microsoft flirt with $4 trillion each.
It’s an orgy of optimism, and everyone’s pretending the music will never stop.
Yet let’s ignore the FAANG gravy train. Most companies playing with gen AI have seen no return on their investments with MIT claiming that 95% of organisations using generative AI pilots aren’t seeing any ROI.
So we’ve got trillion-dollar valuations built on technology that mostly doesn’t work commercially yet. If you’ve never heard of vapor-ware and the promise of never this might just be the best introduction.
Companies are speculating about returns without customers to pay the bills. Yet the real risk isn’t that the datacentres will sit half empty. It’s that the debt used to build them could ripple through the global economy. You can’t “git revert” a trillion-dollar bet.
Hype as a Service (HaaS)
Every cycle has its irrational exuberance. The dot-com boom had domain names (remember how much they changed hands for?). The crypto craze had whitepapers. The AI era has datacentres.
Tech companies call them “the backbone of the AI revolution”, but it’s really the same backbone we’ve been using for years that is hosting files, storing emails, keeping Zoom calls alive.
AI might be the hot new tenant, but cloud computing still pays the rent.
The danger is that we’ve confused infrastructure growth with innovation. We’re scaling the servers, not the smarts. That’s going to cause us problems, and what is astonishing is that seemingly people are unaware of this.
Conclusion
If software teaches us anything, it’s that technical debt is always due given enough time.
Right now, the world is accumulating financial debt to build physical infrastructure for virtual intelligence. And we’re doing it faster than we can prove the business model works.
That’s not a revolution. That’s a refactor waiting to happen.
When this all shakes out, some of these datacentres will indeed power the future. But others? They’ll stand as silent monuments to a moment when we mistook scalability for success.
That’s a sad future, and it’s the one that is coming towards us at an astonishing rate.
About The Author
Professional Software Developer “The Secret Developer” can be found on Twitter @TheSDeveloper.
The Secret Developer once had a dream of owning £3 trillion in assets. They wasted it on chocolate.