Google taps Intel to manufacture three million AI chips by 2028

This partnership represents a structural shift in the AI infrastructure market, moving from a TSMC-centric manufacturing model to a multi-source strategy. For streaming and cloud providers, this supply diversification reduces the risk of hardware bottlenecks that have historically delayed the deployment of generative AI features. If Intel achieves the necessary yields on its 18A process, Google gains a secure, U.S.-based pipeline for the custom silicon powering its Gemini models. Expect other hyperscalers to evaluate similar secondary foundry options as TSMC's 2nm and 3nm capacity remains oversubscribed through 2027.

The order is a pivotal win for Intel Foundry, which has struggled to secure large-scale external customers despite an aggressive 'five nodes in four years' roadmap. Per Wedbush and FinancialContent reports from early 2026, Intel's 18A process node entered high-volume manufacturing in January 2026, featuring RibbonFET and PowerVia technologies. While TSMC remains the dominant producer with a 73% market share as of Q1 2026 (per Counterpoint Research), its executives admitted in June 2026 that customer demand is reaching the absolute limits of its current manufacturing capacity.

Google's shift toward Intel manufacturing follows a period of rapid internal iteration on its TPU hardware. In late 2024, Google launched its sixth-generation 'Trillium' TPU, which provided a 4.7x compute increase over the previous generation. By 2025, the seventh-generation 'Ironwood' TPU reached general availability, specifically optimized for the inference requirements of Gemini 3. The reported 2028 Intel order likely targets eighth-generation or subsequent designs, which analysts expect will utilize Intel's 14A process node and EMIB packaging to scale beyond current 9,216-chip superpods.

Geopolitical and domestic policy incentives are also driving this transition. According to D.A. Davidson analysts in June 2026, Google and Nvidia are under increasing pressure to align with U.S. administration goals for domestic semiconductor manufacturing. Intel has already secured Tesla as a client for its 14A process for the Terafab AI hub in Austin and is reportedly in active testing with Nvidia for its upcoming Feynman architecture. These moves suggest that the concentration of advanced logic manufacturing in Taiwan is finally being diluted by credible Western foundry alternatives.