Intel shares skyrocket as AI data centers rediscover the power of the CPU
The "Nvidia-only" narrative of the AI era just took a major hit. Intel shares surged nearly 20% in after-hours trading Thursday after the chipmaker delivered a blowout Q1 earnings report and a Q2 forecast that caught Wall Street completely off guard. The catalyst? A massive, unexpected demand for CPUs in AI data centers.
In a surprising show of "disciplined execution," Intel CFO David Zinsner revealed that the company was even able to sell through chips that had been previously written off as reserved inventory. This clearance, combined with a 22% year-over-year jump in Data Center and AI (DCAI) revenue, suggests that cloud providers are shifting their focus from simply training massive models to the more practical (and CPU-heavy) stage of deploying them.
Intel Q2 2026 Guidance
"The AI CPU recovery story is no longer a wish; it's evident in our demand profile." — Lip-Bu Tan, CEO
Google’s Dual-Chip Gamble
While Intel is reclaiming the data center, Google is trying to rewrite the rules of the AI chip itself. At its Cloud Next conference, the search giant unveiled its eighth-generation Tensor Processing Units: the TPU 8t and TPU 8i. For the first time, Google is splitting its silicon architecture in two.
The TPU 8t is the heavy lifter, designed specifically for training "frontier" models with 2.8x the performance of the previous generation. However, it's the TPU 8i that is turning heads. Designed specifically for inference (the act of running AI agents), this chip packs a staggering 384MB of static RAM (SRAM)—triple the amount of its predecessor.
By packing more memory directly onto the chip, Google aims to crush the "memory wall" that currently gives Nvidia's H100s their edge. This architectural shift is a direct challenge to Nvidia's dominance, offering Google Cloud customers a custom-built alternative that is 80% faster and significantly more power-efficient for the age of autonomous AI agents.
The Battle for the Data Center
The message from the market this week is clear: the hardware bottleneck is loosening, and the competition is getting specialized. As Intel revives the server CPU and Google optimizes for the agentic future, the AI infrastructure layer is becoming more diverse, more efficient, and—for the first time in years—truly competitive.
