It started with a simple question: why are RAM prices rising? The obvious explanation is AI, with hyperscalers buying enormous quantities of DRAM to fill data centres and consuming much of the world’s supply. Some PC manufacturers, including Lenovo, have suggested that these higher prices may simply be the “new normal.”
But that got me thinking. If demand permanently increases, why wouldn’t manufacturers simply produce more? That’s what markets normally do, and yet the situation feels more complicated. So I found myself pulling on a thread that quickly became much bigger than RAM pricing.
The Fab Assumption
Today, if you want to manufacture cutting-edge chips, there is essentially one accepted path: you build a semiconductor fabrication plant. You spend tens of billions of dollars and use some of the most advanced machines humanity has ever built to project extreme ultraviolet light through incredibly complex optics, etching patterns measured in nanometres into silicon wafers. It’s an astonishing achievement, but it also made me wonder whether we’ve become so good at this process that we’ve stopped asking whether it’s the right process.
We Don’t Question the Medium
Imagine explaining modern semiconductor manufacturing to someone five hundred years ago: “We shine incredibly precise light through mirrors built to atomic tolerances to carve billions of tiny switches into purified sand.” It sounds ridiculous, yet because it’s the technology we’ve refined over decades, it feels inevitable. Maybe it isn’t. Maybe lithography is just one branch of a much larger technology tree.
An Atomic 3D Printer
My mind naturally wandered towards atom-by-atom manufacturing. Instead of carving material away, what if we built devices atom by atom? Instead of needing billion-dollar fabrication plants for every new process node, perhaps manufacturing becomes something closer to software: you design the structure and press print.
Of course, reality immediately pushes back. Atoms don’t politely sit where you place them, and thermal motion, chemistry, and quantum mechanics all matter. Scientists can already manipulate individual atoms inside laboratories, but doing that trillions of times, quickly, cheaply, and reliably is still science fiction. Still, it raises an interesting question: if atomic manufacturing eventually becomes possible, why would we keep building chips that look like today’s chips?
Why Copy DRAM?
Modern DRAM is beautifully simple. Every bit is essentially a tiny capacitor holding charge—one or zero, billions of times over. That architecture exists because it’s manufacturable with today’s tools, but if manufacturing itself changes, does the architecture need to stay the same?
Perhaps not. Future memory might not be built from billions of identical capacitors. Instead, information could be encoded into molecular structures, atomic arrangements, three-dimensional lattices, or even biological structures. Not binary hardware, but physical information.
Growing Computers
This led to another thought. Nature already manufactures incredibly complex structures—trees, bones, brains, DNA—and none of them use lithography. They grow.
Imagine if computer memory wasn’t fabricated but cultivated, not unlike growing a crystal, a coral reef, or a biological scaffold that naturally assembles itself into the desired structure. Scientists are already exploring self-assembling molecules and DNA-guided nanostructures. We’re nowhere near growing DRAM, but perhaps we’re asking the wrong question. Maybe the goal isn’t to grow today’s memory, but to grow tomorrow’s.
Compatibility Is Everything
The biggest obstacle wouldn’t even be the hardware; it would be compatibility. Every operating system, CPU, programming language, and compiler assumes memory is ultimately binary. But that’s actually solvable.
You don’t replace the computer overnight. Instead, you build a translation layer where the CPU continues talking in ones and zeros, and the controller translates those instructions into whatever encoding the new medium uses. This is exactly how GPUs, SSD controllers, and network cards already hide enormous internal complexity behind familiar interfaces—revolution underneath, compatibility on top.
The Bigger Question
I have no idea whether atomic manufacturing, biological memory, or programmable molecular structures will ever become practical. Maybe none of them will, but they’re useful thought experiments because they force us to ask a deeper question: how many of our engineering decisions are constrained by physics, and how many are simply constrained by the manufacturing techniques we’ve inherited?
History is full of technologies that looked permanent until someone changed the manufacturing process. Perhaps silicon lithography isn’t the final chapter—perhaps it’s just the best chapter we’ve written so far.