In 2019, Google announced their Stadia platform. I remember watching the presentation with a sense of excitement- if I couldn’t get the new graphics card, I could still play new games at high graphics for a monthly fee.
The more I thought about it though, the more the concept bothered me. I could play the game, but I wouldn’t actually have the system it was running on. I wouldn’t own a physical copy of the game, either. A monthly subscription would offer a window to the computational power offered by the tech giant, but I would never be able to tweak the game’s content or run experimental projects on the remote server that did the work of rendering the game’s graphics.
Easy, I thought: I just won’t use it- I’ll pay more for a little less power, and I’ll own it myself.
It’s been over 7 years since Stadia was announced, and 3 years since it was shut down. But the figurative digital clouds have formed a heavy ecosystem above those who rely on it.
In 2026, the clouds are more full than ever. We use remote compute for phone calls, shopping, and a growing percentage of applications require an internet connection. This year, Sony announced that it is discontinuing disc production in 2028. Companies like Google, Meta, Amazon, and Oracle together process dozens of exabytes of data annually. In the LLM era, the entire process is off-device. You send a query to the cloud, and a host of hyper-powerful GPUs use every bit of information they can to answer your request and train on the interaction. Cloud computing defines the current era, and this ability to access more compute than you physically own has helped in immeasurable ways. It has become the norm.
There are downsides, too. Consider a hypothetical situation where a new car is released, and anyone can get it for $100 a month, but it requires an internet connection to run. Let’s pretend internet has close to zero dead zones, as satellites fill the sky with signal. The car doesn’t start unless you pay your bill. The terms and conditions by which you can use your car can be updated at the whim of the provider, and you must agree to the terms to operate the car. The engine itself isn’t part of the machine, as the power is provided on the server side of the interaction. The entire utility of the thing has changed to be at the whim of the provider, and the tool itself will change its function to suit the provider’s bottom line.
This is the future I see for the use of any quantity of compute, funneled through skeletal laptop or touchscreen devices that exist to connect to the internet, rather than personally-controlled machines using local code and processing power as they have done in the past. It is not a future I want, but it seems to be where we are headed.
In 2000, about 50% of US households owned a computer. In 2026, this has grown to over 80%, and over 90% own smartphones in 2026. We already accept monthly fees and seemingly endless terms and condition for incremental software updates, each subtly adjusted for maximum returns.
In 2023, Meta admitted to training its AI with user data from Facebook and Instagram. Why wouldn’t they use it? Companies like Google, Meta, Amazon, and Oracle are giants that move based on capital and user-generated data, not on personal idealism. Profits must grow, and therefore more user data must be analyzed and tagged for advertising and surveillance to keep the perceived cost to the consumer low.
As more and more user data is required without increasing costs to the consumer, these trends converge on a future where our “personal computers” are little more than portals to “real computers” that do all the processing and user data analysis on the remote side. We pay a low cost for a touchscreen that does whatever makes the provider the most money, while retaining some semblance of functionality through an internet connection. Televisions have already undergone this transformation, with services like Telly offering the screen for free in exchange for user data and ad placement. Memory costs are inflating and the world’s financial behemoths are amassing all the compute they can afford (or more) for AI buildouts.
It seems like the end to this road is that tech companies will use machine learning systems to use software through user interaction data, and to then offer users a frictionless interaction with their devices (at the cost of handing over all interaction and data to an external cloud service).
Apple’s approach to AI is different in a significant way. While many requests will still be kicked to external services, Apple has stressed the importance of localized compute on its devices. If more companies are steered this way by financial motivation, we have a chance to encourage these giants to favor local, user-owned compute rather than surrendering to a financially tempting cloud alternative.
The devices we buy should work for us, not for their manufacturer.
The next time you make a hardware or software purchase, consider what is subsidized financially by your personal data, and what compute is outsourced to a remote machine that you will never see, own, or know the location of.