Intel was a patient on a monitor, barely breathing. I remember watching the numbers and feeling the room tilt toward doom. Then a quarterly call arrived that made me sit up.
You and I both saw the story: for years Intel trailed AMD and Nvidia as GPUs became the shorthand for AI. But something small and stubborn changed this quarter, and it matters more than the press release headlines.
Intel reported a 7.2% revenue increase this quarter. That single line shifted the tone from salvage to momentum.
I read the earnings call the way a detective reads a witness statement. The number itself is modest. The story behind it is richer: demand for CPUs rose, Intel signing big deals with Tesla and SpaceX, and Washington taking a roughly 10% stake. CEO Lip-Bu Tan framed it plainly — CPUs are coming back as the orchestration and control layer for the new kinds of AI systems customers are buying.
Customers are buying more CPUs alongside GPUs. That shift is less about replacing GPUs and more about adding a control layer.
You should note this: the CPU-to-GPU ratio Intel used to see was about one-to-eight; it’s moving toward one-to-four. I asked why on the call and the answer landed like a fact you can bank on: agentic AI — systems that plan, delegate, and manage tasks across models and services — needs coordination. That coordination is what CPUs do best.
Why are CPUs important for AI now?
Let me be practical. GPUs still do the heavy lifting for matrix math and model inference. But when an AI system must orchestrate multiple models, route data, and manage state, GPUs are inefficient for that control work. CPUs handle I/O, concurrency, and context management without burning through costly accelerators.
Analysts and rivals are saying the same thing in public. When Morgan Stanley and Nvidia echo a trend, it stops being a rumor.
Morgan Stanley flagged the same movement: the bottleneck is shifting from raw GPU compute to the control and coordination layers CPUs provide. Nvidia’s Jensen Huang has even tied agentic AI to enormous market opportunity — he mentioned $1 trillion in potential revenue (≈ €925 billion) — and started talking about CPUs alongside GPUs. When competitors and financiers align, you have to pay attention.
Will GPUs become obsolete?
No. I want you to treat that question like a red herring. GPUs remain essential where large-scale parallel compute is required. What changes is the architecture of systems: more CPUs running orchestration, more GPUs doing acceleration, and more software gluing both together. The debate isn’t replacement; it’s about balance.
Intel’s deals matter. Real contracts with Tesla and SpaceX change the calculus.
Contracts with companies tied to Elon Musk give Intel not just revenue but credibility. Those agreements aren’t small pilot projects; they signal manufacturers and hyperscalers that Intel’s road map is being tested in high-stakes environments. That’s how perception shifts into pipeline and then into predictable revenue.
I’ve watched companies resurrect reputations before. Two forces make this one worth following: first, a macro shift in how AI systems are built; second, real commercial wins that turn theory into cash flow. Intel is showing both.
One metaphor is unavoidable: CPUs are the conductors of the AI orchestra. The second metaphor I’ll use is this: Intel’s quarter was a first steady breath for a company we thought might not recover.
If you want signals instead of hot takes, watch three things: the CPU-to-GPU procurement ratios at cloud providers, follow comments by Lip-Bu Tan and Jensen Huang on orchestration roles, and track deal announcements with major OEMs and hyperscalers. Tools and platforms to watch include Nvidia’s announcements, Anthropic’s product roadmap (including Claude Code), and enterprise buys from Tesla and SpaceX.
Intel’s comeback isn’t magic. It’s a market correction that favors orchestration and systems design as much as raw horsepower — and if you’ve been betting only on GPUs, you might be missing half the architecture.
So tell me: are you ready to reweight your assumptions about where AI value will actually accrue?