Hardware sets the floor. Software sets the ceiling.
Every hardware brand that wins is a software company underneath. Apple didn't beat the market on silicon or screens. Tesla's powertrain isn't the part nobody else can build. The gap, in both cases, is software.
I argued this in a room before the news made it obvious
A while back, inside a company that builds hardware for a living, I argued we should solve a physical conflict with an FPGA and an algorithm — not with more hardware.
The point wasn't elegance. It was cost. Done right, the algorithm absorbs a problem the mechanism would otherwise have to pay for, and the bill of materials goes down instead of up.
That wasn't the default position in the room. The default was better hardware. Tighter tolerances. More of the physical thing.
I wrote that argument up earlier, in a piece called Soft for Hard. I'm picking it back up because the rest of the world just made the same call at a scale that's hard to ignore.
Then SpaceX bought Cursor
SpaceX is buying Cursor, the AI coding tool, at a reported $60B.
The deal matters less than the signal. The hardest hardware company on earth decided the next phase of competition is won in software — not because a rocket needs an editor to write its code, but because the leverage has moved.
That's the thing I was arguing about in a conference room. It's now happening in public.
It plays out two different ways, depending on who you sell to.
Consumer: software stands in for hardware you don't have
Consumers buy outcomes. They don't care whether full color is voxel-level or pixel-level, whether the gamut is 500,000 colors or ten million. They care whether it works, whether it's fast, whether it's cheap, whether it's fun. Good enough is the whole spec.
In May 2026, Snapmaker hired Ratdoux — the developer behind Full Spectrum, a slicer fork that fakes full color.
It isn't real full-color printing. It's an optical trick at the slicing layer: stack semi-transparent filament in alternating thin layers and let the eye blend two colors into a third. Three or four real filaments, arranged to look like many.
It costs thousands of tool changes and ten to forty-plus hours per model. Up close you see banding. It leans on translucent material.
And for a lot of buyers, it's already enough.
That's the consumer reality. The user doesn't care whether it's "true" full color or just looks like it. They care whether the result clears their bar. Snapmaker's hire is a good move precisely because of that: at the consumer tier, simulating the result in software can beat chasing it in hardware.
The rule underneath: software can cover for weak hardware. Hardware can't cover for a weak software experience.
Professional: algorithms lower what the hardware has to be
Professionals do care about parameters. But the path to a parameter doesn't have to run through more hardware.
Use the algorithm to resolve the physical conflict, deliver a better whole-system result, and take hardware cost out at the same time. That's the cheaper road.
The DJI people who left to build Bambu Lab did exactly this — precise software scheduling instead of brute mechanical precision. Bambu didn't win by building the most exquisite hardware. It won on software.
That's not parity. That's coming in a level below on hardware and beating the parameter anyway.
Algorithms aren't magic
At the (near-)industrial end, there's a physical floor an algorithm can't fill in. High-precision manufacturing, high-load and extreme environments — materials science, precision optics, thermodynamics stay as hard gates. Software can optimize. It can't conjure a difference in physical scale that isn't there.
The algorithm is a multiplier, not a replacement.
There's a second constraint that kills more software-first hardware teams than the first one: cost and supply chain. An algorithmic moat needs a real supply chain and real cost control underneath it. A beautiful algorithm you can't manufacture at cost is a slide, not a product.
DAU
Software redefines where a hardware company competes. It doesn't replace the hardware.
When drawing a board stops being the barrier — and AI-assisted hardware design isn't far behind AI-assisted coding — the competitive axis moves. Not whose spec sheet is higher. Whose data flywheel spins faster, whose software experience is better, whose algorithms understand the user.
Call it DAU.
Data × Algorithm × UX.
- Data — the private dataset from your vertical.
- Algorithm — how efficiently you schedule the hardware.
- UX — the integrated hardware-software experience.
SpaceX didn't pay $60B for a coding tool because it needs one to build rockets. It paid because the next decade goes to whoever figures out the software first.
I made that bet in a conference room before the news caught up. I'm more sure of it now.