Intel Foundry Challenges TSMC Dominance With Massive Google AI Chip Order

Intel's foundry ambitions may have just received their biggest validation yet. According to a report from The Information, Google has placed an order with Intel to manufacture more than three million Tensor Processing Units (TPUs) in 2028 after months of testing Intel's advanced packaging technology. The deal, if accurate, would represent the largest external AI chip manufacturing win in Intel Foundry's history and could mark a significant shift in the semiconductor industry's efforts to reduce reliance on TSMC.

The reported order arrives as demand for AI hardware continues to outpace available manufacturing capacity. TSMC remains the dominant supplier of advanced chips for companies such as NVIDIA, Apple, and Google, but the AI boom has placed enormous pressure on both its leading-edge wafer production and especially its advanced packaging operations. TSMC executives have repeatedly acknowledged that demand for AI-related manufacturing capacity continues to exceed supply.


While Intel has spent years attempting to establish itself as a viable alternative leading-edge foundry, the company struggled to gain traction with major external customers. That narrative has begun to change in recent months as Intel ramps production on its 18A process technology and begins shipping products such as its Core Ultra Series 3 "Panther Lake" processors, helping convince customers that its next-generation manufacturing roadmap is no longer theoretical.

According to the report, Google selected Intel after extensive evaluation of the company's advanced packaging capabilities. Advanced packaging, which refers to things like TSMC's CoWoS and Intel's Foveros, among other technologies, has become one of the most critical parts of modern AI chip design as processors increasingly rely on multiple chiplets and stacks of high-bandwidth memory integrated into a single package. In this area, Intel's EMIB packaging technology competes directly with TSMC's widely used CoWoS platform.

The Information also reports that NVIDIA is evaluating Intel technology for future products, although no production orders have reportedly been placed. Sources cited by the publication claim NVIDIA is exploring whether Intel's packaging technology and manufacturing processes could be used in future multi-die GPU designs associated with the company's Feynman architecture, currently expected to arrive in 2028.


That possibility is not entirely new. Back in January, DigiTimes reported that Intel could play a limited role in manufacturing and packaging NVIDIA's Feynman-generation GPUs. According to that report, TSMC would continue producing the primary GPU compute die, while Intel could manufacture portions of the I/O die and handle as much as 25% of the advanced packaging volume through its EMIB technology. Neither NVIDIA nor Intel has publicly confirmed those plans, but the latest report suggests discussions between the companies remain active.

The broader significance of the purported Google order extends far beyond a single customer. Intel Foundry has spent years attempting to prove it can compete with TSMC not only in terms of manufacturing technology, but also on execution, yields, and volume production. Winning business from one of the world's largest AI infrastructure operators would provide a powerful endorsement of Intel's progress.