Hardware's AWS moment

Software used to be a walled garden. To ship it, you needed a village.

I mean that literally. To put a web app in front of users in 2005, you bought physical servers. You drove them to a colocation facility or built a server room, racked them, wired them, and paid for a fat internet line you mostly wouldn't use. You did capacity planning for a traffic spike that might never arrive, and you over-provisioned anyway, because the alternative was your one machine melting the day you finally got lucky. Founders raised money to buy hardware before they had a single user. Shipping software was a logistics problem, a real-estate problem, and a hiring problem before it was ever a code problem.

Then AWS happened. An engineer sitting in their garage could suddenly deploy software to the entire world on a credit card. But the cheap servers were the least of it. What AWS really handed people was a way to operate. A tiny team could now ship many times a day, watch how real users behaved, and change the product by the afternoon. A company like Instagram could reach tens of millions of people with a team small enough to fit around one table. The gift wasn't infrastructure. It was permission to stay lean, move fast, and sit close to your customer while doing it.

I think the same thing is now happening to hardware. And I'm stress-testing that belief, literally, in my apartment.

The building blocks got cheap

Hardware is quietly getting its own version of S3 and EC2: cheap, standardized blocks you can assemble without owning the machinery underneath.

Sensors are the clearest one. The little chips that give a machine a sense of its own motion and orientation, the IMUs inside every drone and phone, have fallen roughly an order of magnitude in cost over the last decade. Motors are next. An Indian startup called Vimag Labs has patented high-performance electric motors that run without any rare-earth magnets at all, using power electronics and software to do the work the magnets used to. That matters because rare-earth supply is the one chokepoint China still firmly controls, and a motor made of copper, steel, and code can be built almost anywhere. The supply chains that tie all of this together have never been easier to thread, either. A design in your head can be a working prototype out of Shenzhen in a couple of weeks.

And the design itself is getting cheaper to do. I'm building an air purifier for my apartment right now, and AI is doing the parts that used to need a mechanical engineer and a five-figure software license: generating the geometry, running the CFD simulations that model how air actually moves through the housing, helping me choose and source components.

None of these blocks existed at this price, in this form, five years ago.

But cheap parts were never the real point

Here is where I want to be careful, because the obvious reading is wrong. AWS didn't turn everyone into a sysadmin, and I'm not predicting an army of tinkerers soldering their own vacuums. My air purifier isn't the forecast. It's the control experiment. It's how I'm checking, with my own hands, that the blocks actually work.

The forecast is a company.

AWS's real gift to software was never the cheap servers. It was that a handful of people could run a company the way you'd want any company run: small, fast, and obsessed with the person on the other end. That operating model is what I think is finally arriving in hardware, and it's genuinely new here.

Why was this impossible before? Physical products punished iteration. Tooling cost a fortune, injection molds took months, and minimum order quantities meant you placed a huge bet before you learned anything real about what people wanted. So hardware demanded big teams, long roadmaps, and decisions locked years out. You couldn't revise a molded part on Tuesday because you'd heard something on Monday. The distance between the people making the product and the people living with it wasn't a culture problem. It was baked into the economics.

Cheap blocks dissolve that. Quick-turn PCBs, 3D printing, AI-assisted design and a mature ecosystem to bring it together mean a physical product can be iterated on something close to a software cadence. Lean team, customer obsession, rapid iteration: those are the primitives that are actually new to hardware. Not the sensor.

The moats are being ripped out

If the operating model is up for grabs, the incumbents are in trouble, because the very thing that used to protect them now works against them.

I recently took apart an old Dyson vacuum. It has something like a hundred different proprietary screws, the kind of choice that exists only to keep you out. Once I was in, I found something stranger. There was no battery management system. The battery lives in an environment of very rapid charging and discharging, exactly the case where a BMS earns its keep, and Dyson had made a deliberate decision to leave out a chip that costs them a few cents.

Read that decision closely. It only makes sense at scale. When you ship millions of units, a few cents times millions becomes a line item someone is paid to shave, and the person shaving it will never hold the vacuum whose battery dies early. The thing that used to be the moat, enormous scale, is now the thing generating the pressure to cut the corner. It's also what makes the incumbent slow, far from the customer, and locked into decisions made two years ago. They can't run the new playbook. Their size is exactly what forbids it.

Enshittification, up close

I don't think I'm imagining the decline. In the past year I've lost an Apple Watch, a pair of AirPods Pro, a MacBook Pro, a microwave, that Dyson, and an LG display to catastrophic failure. Not one of them was more than three years old.

My air conditioner has an "AI mode." What it actually does is silently reset the temperature to 26 degrees about fifteen minutes after I turn it on. That is the entire feature. Someone shipped a thermostat, called it intelligence, and moved on.

Cory Doctorow has a word for this, enshittification, and he wrote a whole book about it in 2025. The short version is that products get worse on purpose, because the people making them have stopped competing for your loyalty and started extracting from it.

It's a distance problem

Just like with software companies, enshittification is really a problem of distance. The further the person making the decision sits from the person who has to live with the product, the more corners get cut, because each corner is invisible from where they're standing. Dyson's missing chip wasn't chosen by someone close to the user. It was chosen by a spreadsheet, several org charts away from anyone who would ever feel the battery fail.

This is why the software playbook matters so much in hardware, and it's the whole argument in one line. Staying lean, obsessing over the customer, and iterating weekly are not three nice values. They are what keeps that distance near zero even as a company grows. Customer obsession is distance-collapse written as a habit. Rapid iteration is what happens when feedback can still reach the person holding the pen. The incumbent lost this not because its people got worse, but because scale pushed everyone too far from the person using the thing. The new kind of company gets to keep the distance short on purpose.

The next household name will come with a ten-year warranty

I started out assuming the next great consumer electronics company would be a new giant that resets all of this. I've stopped believing "giant" is the right word. Giants are what create the distance in the first place.

The next household name, the one people actually love rather than tolerate, will be a company that runs hardware the way the best companies run software: small enough to stay close, fast enough to keep listening, honest enough to build for the person and not the spreadsheet. It won't win on features or price or a mode with "AI" in its name. It'll win because its products last, and because it still respects you years after you've paid. A ten-year warranty sounds reckless from a boardroom optimizing a margin. It's the most natural promise in the world for a company that never let the customer get far away.

When it arrives, we won't recognize it from a launch event. We'll recognize it from the quiet fact that ten years on, the thing still works, and nobody is surprised.