The Worry of Rent-only Compute.

The current supply chain crunches, mostly around RAM, but also around GPUs, caused mainly by speculative data-center buildouts, have seen a rash of worry to the tune of: ‘What if the powers that be have decided we will only rent our compute?!’

There hasn’t been much talk about how that would look in practice, even though I’ve seen it mentioned several times now. None of the mentions included the term thin client, for example.

There also doesn’t seem to be much of an attempt to look at the history of computing or how markets tend to work. A good sign the worry has gotten carried away from the facts.

If you look back at computing, you see the cycle of expensive and cheap, as maturity cycles work themselves out, as early capacity bets get snagged by lagging demand, followed by time-sharing efforts to help cover some of those costs, it’s clear that computing isn’t and will never be immune to business cycles. But it all follows a fairly basic pattern.

So if you think RAM has gotten expensive, and that will cause the only way to compute to become via the cloud, I’ll try to push back.

It starts with the thin client. You still need some kind of minimal computer to rent remote compute, unless you’re doing some weird thing like compute-by-mail where they deliver you a stack of punchcards for your data and code, and you mail that in, and they mail you back more punch cards or a printout.

So you have your mobile. Those are getting more expensive because of the RAM crunch. Can it get thinner? How much thinner? How many websites want to serve to underpowered clients? How many apps do? The tendency has long been the other direction. And even if mobiles go thin, how much real work is done on them? Tablets, laptops, and desktops could go thin. Some use cases are already cloud-based.

So assume an extended demand-driven hike on RAM and GPUs, long enough to drive many onto thin clients. We would expect that to happen first with consoles. Why would Microsoft and Sony and Nintendo sell ever-more-expensive hardware and still treat it as a thin client? At the very least they would sell the thin-client-consoles for people who can’t afford the full machines. And as people who can afford them opted for thin clients, you’d see them pull back on high-end hardware entirely.

We haven’t seen that, despite the fact they don’t make most of their money on the consoles themselves, but on the games. But maybe it’s just waiting to happen.

And after that, more games would offer cloud-gaming clients, and future PCs would thin out to smaller RAM, lighter compute.

And in ten years, everyone’s on thin clients, cloud-based-everything, and then what happens? The hardware makers would suddenly have a supply glut as demand has filled out. Massive data centers power all the services, but they are already built. Demand tapers off. Hardware makers’ prices drop.

As prices drop, consumers that want real hardware ditch their thin clients or switch to dual-use, and the cycle begins anew.

Nobody likes supply chain disruptions, especially when caused by war or economic speculation. But they aren’t a good reason to invent fantasies that economic systems don’t work the way they always have. The above scenario is less plausible than the bubble popping. But even if the bubble grows flat as part of a new equilibrium, it still ends with hardware getting cheaper (relative to the value of money). If it pops, at least some current-generation hardware becomes significantly cheaper as orders cancel and firms try to make some profit off of sunk costs.

That said, sometimes there are benefits in the long-run. High costs of traditional hardware can open the door to alternatives that would otherwise be ignored, for example. Or new form-factors or designs can be developed to help account for price volatility (so that, e.g., a piece of hardware ships in one configuration if RAM is cheap or another if it is more expensive). It can be hard to guess what those possible silver linings look like, but we can dependably hope they will be there.