Remember when graphics cards were something only gamers cared about? There was a stretch in the late 2010s when the average person’s only brush with Nvidia was a teenager in the house complaining that a certain card was sold out or overpriced. The company made the thing that made video games look pretty. That was the whole story.
That story has changed so completely that Nvidia’s sales forecasts are now treated as economic weather reports. Bloomberg’s recent coverage carries the headline that Nvidia predicts its AI-fueled sales surge will extend into 2028. Not next quarter. Not next year. Multiple years out, which is an unusual thing for any hardware company to say out loud.
If you’re reading agent101.net, you’re probably not here for chip margins. You’re here because you want to understand AI agents, the software that does tasks on your behalf. So let me explain why a forecast about a chipmaker’s revenue is actually a forecast about the tools you’ll be using.
Why chip timelines matter to people who never touch chips
AI agents run on borrowed muscle. When you ask an agent a document, sort your inbox, or draft a reply, that request travels to a data center where specialized processors do the actual thinking. Nvidia makes most of those processors. So when Nvidia says demand stays strong through 2028, it is saying something about how much compute the world is building out and for how long.
For everyday users, that translates into a few practical things:
- More capacity generally means agents that respond faster and handle longer, messier tasks without timing out.
- More capacity also tends to mean cheaper per-request costs, which is how features move from expensive tiers into free ones.
- A multi-year buildout signals that the companies making your tools expect to keep investing rather than quietly winding things down.
That last point is the one I’d underline. If you’ve ever adopted a tool and watched it get shut down eight months later, you know that the scariest thing about new software isn’t whether it works. It’s whether it will still exist.
The other headline in the room
Here is where I want to be honest with you rather than cheerful. On the same news cycle, Tech Times reported that SanDisk fell 54% after Morningstar described the AI boom as finite, projecting a downturn before 2029. So we have two timelines sitting side by side. One company forecasts strength into 2028. One analyst outfit expects a turn by 2029.
Those two things are not actually contradictory, and I think that’s the useful insight. Both are describing a boom with an end date. They just disagree about the exact shape of it. A market that grows hard for several years and then cools is not a scandal. It’s what infrastructure cycles look like. Railroads did it. Fiber optic cable did it. Telecom did it loudly enough that people still write books about it.
JLL’s 2026 Global Data Center Outlook is another piece of this puzzle. Data centers are the physical buildings where all this happens, and the fact that firms are publishing multi-year outlooks on them tells you the buildout is being planned like real estate, not like a software launch. Buildings take years. That’s part of why forecasts stretch this far.
What I’d actually do with this information
Not much, honestly, and I mean that as reassurance. You don’t need to time an AI infrastructure cycle to use an AI agent well. But there are a couple of habits worth forming.
First, prefer tools that let you export your data. If a company’s economics get squeezed in a cooler market, the ones that made it easy to leave are the ones that earned your trust in the first place.
Second, treat “the boom continues” and “the boom ends” as the same instruction: learn the underlying skills, not just one product’s buttons. Knowing how to write a clear instruction for an agent, how to check its work, and how to spot when it’s confidently wrong will outlast any particular app.
Third, resist the urge to read every chip headline as a verdict on whether AI is real. Nvidia forecasting demand through 2028 is a statement about spending. It is not a promise that agents will be smart, useful, or worth your subscription. Those are separate questions, and they get answered by the software, not the silicon.
A boom with a calendar attached
What strikes me most about this news is how specific the years have become. Not “the AI era.” Not “the coming decade.” Just 2028, and a counterpoint at 2029. Numbers on a calendar are a sign that a technology has moved from speculation into planning, with all the ordinary business risk that comes with it.
For those of us learning to work alongside AI agents, that’s a reasonable place to be. The tools are getting steadier funding for a few more years. Use that window to get genuinely good at directing them, and you’ll be fine regardless of which forecast turns out to be right.
🕒 Published: