Google and Nvidia are considering Intel as an alternative to TSMC for AI chip production.

      For years, the boom in artificial intelligence has permeated several manufacturing floors across Taiwan. This reliance is starting to seem precarious enough that both Nvidia and Google are seeking alternatives, with Intel emerging as an unexpected contender.

      Google has placed a substantial order with Intel for over three million of its proprietary tensor processing units (TPUs) to be delivered in 2028, following months of evaluating Intel's packaging capabilities. Additionally, Nvidia is assessing Intel’s advanced packaging technology and its latest 18A process for future chip production, as reported on Monday by The Information, which cited four sources familiar with the discussions. Consequently, Intel's stock surged nearly 12 percent following the announcement.

      The driving force behind these decisions is scarcity. TSMC, the Taiwanese powerhouse responsible for manufacturing nearly all cutting-edge AI chips, is struggling to keep up with rising demand, particularly in the advanced-packaging sectors that integrate chips and memory. For firms creating the most desirable silicon products, dependence on a single supplier in one country has morphed into a strategic vulnerability.

      The two companies' strategies differ in intensity. Google’s order is definitive, with a commitment to over three million TPUs in 2028 after thorough testing of Intel’s packaging. This is part of an expansion project that Morgan Stanley estimates will exceed six million TPUs across 2027 and 2028. Conversely, Nvidia has yet to make a commitment; it is conducting preliminary tests, including multiproject wafer runs on the 18A process, and evaluating if Intel can produce a processor that combines four graphics chips into one, linked to its Feynman GPU architecture set for release in 2028.

      For Intel, even this cautious engagement represents progress. Once a leader in American chip manufacturing, Intel has spent years attempting to transform its contract-manufacturing division into a viable competitor to TSMC, albeit with limited success and significant financial losses. The company has been vying for Apple as a customer, and both the US government and Nvidia have invested in it.

      What has been missing from Intel's recovery effort is actual orders instead of mere goodwill. The immediate opportunity exists in packaging rather than in high-end chip manufacturing, where Intel still lags behind. Reports indicate that SK Hynix is testing the compatibility of its high-bandwidth memory with Intel’s packaging as well. A key unresolved question is whether Intel's 18A process can achieve yields similar to TSMC, which has previously been a stumbling block for Intel.

      This situation highlights how far Google has gone to assert control over its silicon supply. Its tensor processing units are now shipped in significant numbers, and it has been diversifying orders among various partners to diminish its dependency on both Nvidia and TSMC. Integrating Intel into its supply chain serves as an additional safeguard.

      While none of this diminishes TSMC’s vast lead in cutting-edge manufacturing, it signals that the AI sector's most influential clients are less inclined to place all their resources in one Taiwanese supplier. For Intel—previously counted out—there may still be an opportunity to contribute to their supply needs. For a company eager to showcase a revival, securing Google as a customer represents a significant achievement.