CPUs Are Back
Intel posted its fastest revenue growth in 15 years — 25% to 6.1B — driven by a 59% surge in data center and AI revenue. The GPU-centric AI narrative is being rewritten by inference demand.
On July 24, Intel reported second-quarter revenue of $16.1 billion — a 25 percent year-over-year increase, the company's fastest quarterly growth in more than 15 years Motley Fool, July 24, 2026. The Data Center and AI segment led the charge, surging 59 percent to $6.3 billion, accelerating from 22 percent growth in Q1. Adjusted gross margin hit 41.8 percent, up 12.1 percentage points from a year earlier. Adjusted operating margin swung from negative 3.9 percent to positive 17.2 percent. The business that produced an adjusted loss in the same quarter last year generated $2.2 billion of adjusted net income and $7.0 billion in operating cash flow Intel Press Release, Q2 2026.
The headline number is 25 percent revenue growth. The structural number is the 59 percent growth in data center and AI revenue — because that number tells a story about the AI buildout that the GPU-centric narrative has been missing.
For the last two years, the AI hardware story has been straightforward: GPUs are everything. NVIDIA's market capitalization, data center revenue, and supply allocation have dominated the semiconductor conversation. CPUs were cast as supporting infrastructure — necessary but not strategic. Intel, the primary server CPU supplier, was treated as a company that had missed the AI wave entirely, with investors questioning whether the foundry strategy or the core product line could generate growth in an AI-dominated compute environment SemiAnalysis, July 2026.
That analysis was correct until approximately six months ago. It is no longer correct.
What changed is the composition of AI compute demand. Model training — the phase that consumed the industry's attention and NVIDIA's GPU supply — is being joined at scale by inference, the deployment phase where trained models are actually used. Inference workloads are more CPU-intensive than training. They require orchestration, memory handling, and real-time processing that GPUs alone do not efficiently provide. The CPU-to-GPU deployment ratio in data centers has shifted from approximately 1:8 to 1:4 and is projected to reach 1:1 or beyond for agentic AI applications Tom's Hardware, July 2026. AI is projected to account for nearly 50 percent of all data center workloads by late 2026, with inference overtaking training as the dominant use case IDC, July 2026.
The supply response has not kept pace with demand. Intel has confirmed price increases on select server CPUs, with some Xeon processors now over $1,000 more expensive than their previous list prices Tom's Hardware, July 2026. Intel is reallocating manufacturing capacity from consumer chips to data center processors and raising capital expenditure to $20 billion in 2026, with further increases planned for 2027 Strait Times, July 2026. Chief Financial Officer Dave Zinsner described the demand as unforecastable at prior run rates: Intel's own April forecast topped out at $14.8 billion in revenue; the company delivered $16.1 billion Motley Fool, July 24, 2026.
CEO Lip-Bu Tan, in the earnings release, described the environment in terms that contrast sharply with the company's recent positioning: "AI is driving unprecedented demand for compute, and as we continue to execute, Intel is well-positioned to capture sustainable growth across our CPU franchise, ASICs, advanced packaging and vast wafer foundry network" Intel Press Release, Q2 2026.
The market's reaction to the earnings was revealing. Intel stock rose roughly 12 percent in after-hours trading Thursday, then fell below Thursday's close by Friday — an 8 percent swing that erased the gain. The reason was valuation: at a $480 billion market capitalization trading at roughly 60 times annualized adjusted earnings, the stock prices in the assumption that quarters like this one are the new baseline, not an anomaly Motley Fool, July 24, 2026. Investors want to believe the turnaround is real. They are not yet willing to pay for it at peak-cycle multiples.
The broader implication is about the economics underneath the AI model layer. Every narrative about AI infrastructure spending has emphasized GPU scarcity, training compute requirements, and the $30,000 price tag on a single H100. Those factors are real. But the inference transition means the hardware requirements of AI are diversifying — and diversification changes the economics. The CPU market is structurally more competitive than the GPU market, with Intel, AMD, and ARM-based alternatives all contesting the same socket. AMD has nearly doubled its server CPU total addressable market forecast to over $120 billion by 2030, driven by the agentic AI and inference demand that Intel's Q2 numbers are now confirming Seeking Alpha, July 2026. The pricing power Intel is demonstrating — $1,000-plus increases on Xeon processors — exists because demand exceeds supply across the entire silicon ecosystem, not just in GPUs.
Intel's Q2 is not proof that the company has permanently reversed its trajectory. It is proof that the AI buildout is large enough and complex enough to revive a company that the market had written off as irrelevant to the moment. That is a statement about the scale of the buildout, not about Intel's execution. But it is also a statement about the nature of the AI industry's second act: the transition from training frontier models to deploying them at scale requires hardware that the first act never had to buy.
