Custom silicon is getting serious competitive attention. Google and Intel announced a deepened partnership to co-develop custom chips, with both companies citing surging AI-driven CPU demand and a tightening global supply as the catalyst.
The shortage isn't abstract. Data centers running large language models burn through processors at rates that general-purpose CPU manufacturing wasn't designed to meet. Purpose-built chips - designed for specific AI workloads rather than general computing - can cut power consumption and reduce latency compared to standard server hardware when running inference (generating AI responses) or training (teaching models on data).
Google already designs its own Tensor Processing Units (TPUs), specialized chips optimized for AI workloads used internally for years. Intel brings manufacturing scale and deep CPU architecture experience. The partnership suggests each company believes they'll reach better outcomes together than separately.
The competitive backdrop is hard to ignore. Nvidia currently dominates AI chip revenue through its H100 and A100 GPU lines. Google has an incentive to lower its own infrastructure costs; Intel needs to recapture relevance in a market where it's been losing ground. A successful custom chip collaboration gives both companies leverage they don't have individually.
Custom chip development cycles typically run two to four years from design to production deployment. Any output from this partnership won't reach data centers until 2028 at the earliest. For now, it signals that both companies are treating AI compute constraints as a structural, long-term problem worth solving at the design level - not something that gets fixed by ordering more off-the-shelf hardware.