For three years, the artificial intelligence boom has had one obvious winner: Nvidia, whose graphics processing units, or GPUs, do the heavy mathematical lifting behind systems like ChatGPT. Older corners of the technology industry — the makers of conventional central processing units (CPUs), storage drives, and enterprise software — have looked stranded, their products eclipsed by a flashier new stack. But the stock market has begun telling a different story, and Wall Street is now scrambling to understand how big a slice of AI spending the older technologies might actually command.
The evidence is hard to ignore. AMD said this past week that it expects compound annual growth of 35 percent in CPU sales over the next few years, roughly double the forecast it issued only six months earlier. ARM also reported surprisingly strong demand. Intel's share price has climbed 400 percent since the United States government took a stake in the company last summer. Even Seagate, a maker of disc drives long viewed as a relic, has jumped 60 percent in a month. The rebound rests on a simple insight: GPUs may train AI models, but CPUs are needed to load those models, manage queries against them — a process called inference — and stitch the output back into real applications.
The rise of so-called AI agents could deepen the trend. Agents are software programs that perform multi-step tasks autonomously rather than answering one prompt at a time, and tech executives including Nvidia chief executive Jensen Huang argue they will run on the same digital plumbing that already powers everyday work. Intel's chief financial officer noted recently that training a model requires only about one CPU for every eight GPUs, but agent workloads will demand CPUs and GPUs in roughly equal numbers. If accurate, that ratio rewires the economics of every data center on the planet.
The harder question is how the resulting value gets divided. If Old IT companies sit quietly in the background, they will collect modest margins. But if they can position themselves as central to the agent era, they may win back pricing power — and that question is sharpest in software, the most beaten-down corner of the old guard. Last month Salesforce became the first major company to launch a 'headless' product, software with no human-facing interface, designed to be operated directly by AI agents. The strategic logic is that years of accumulated customer data will let Salesforce orchestrate those agents intelligently. The risk is that, like most cloud software firms, Salesforce has always charged based on the number of human users — and headless software, by definition, has none.
The deeper bet beneath all of this is that probabilistic AI models, which by nature produce slightly different outputs each time, will not on their own deliver the reliable performance that real businesses require. Turning raw AI horsepower into a dependable computing resource is the next great battleground in technology, and it is one in which orchestration software, fast CPUs, and cheap storage all matter. Wall Street is not yet ready to declare Old IT a winner. But the days when investors could safely treat it as collateral damage from the AI revolution appear to be ending.
For three years, AI hype has been all about Nvidia's GPUs. But suddenly, the unsexy chips and dusty software companies of 'Old IT' are roaring back — and Wall Street is scrambling to figure out why.
Since ChatGPT launched, investors have crowned a small group of AI winners — Nvidia above all — while older tech names like Intel, AMD, Seagate and Salesforce have looked like yesterday's story. That narrative is shifting. As AI moves from research labs into actual business workflows, the boring infrastructure underneath it (CPUs, storage drives, enterprise software) is turning out to be more important than people assumed.
The trigger is a wave of stronger-than-expected demand for central processing units (CPUs) — the general-purpose chips that have powered computing for decades. AMD now expects 35% compound annual growth in CPU sales over the next few years, roughly double its forecast from six months ago. Intel's stock has bounced 400% since the US government took a stake last summer, and disc-drive maker Seagate has jumped 60% in a month.
GPUs do the flashy AI math, but they don't run alone. Think of the AI stack like a Formula 1 race team:
If you're considering computer science, finance, or any business career, this is the central question of the next decade: who actually captures the money in an AI economy? It's not always the company with the coolest technology — it's the one that controls a chokepoint customers can't avoid. Index funds, your future 401(k), and probably your first employer's stock comp all depend on getting this right.
We've seen this movie before. When the internet exploded in the late 1990s, everyone bet on the flashy dot-coms — but the durable winners turned out to be the 'picks and shovels' companies: Cisco's routers, Oracle's databases, Microsoft's operating systems. The current bounce in Old IT suggests Wall Street is starting to ask whether AI will follow the same pattern. Watch two things next: whether agent adoption actually drives the predicted CPU surge, and whether software companies like Salesforce can reinvent their per-seat pricing for a world where the 'seats' are bots.