Google Orders 3 Million AI Chips from Intel Foundry to Scale TPU Production

Google is aggressively scaling its in-house AI hardware production to keep pace with the exploding demand for compute. The company has reportedly ordered more than 3 million next-generation Tensor Processing Units (TPUs) from Intel Foundry for production in 2028.

This massive order is part of a broader ramp-up in Google's silicon strategy. According to estimates from Morgan Stanley, Google is expected to produce over 6 million TPUs between 2027 and 2028. As part of this technical evolution, Google's TPU v8e—slated for release in the second half of 2027—is likely to incorporate Intel's Embedded Multi-die Interconnect Bridge (EMIB) technology for advanced packaging.

The Shift from TSMC

For years, the high-end AI accelerator market has been dominated by Nvidia, with the vast majority of these chips manufactured by TSMC. While Google has long developed its own TPUs to optimize for specific AI workloads and reduce operational costs, the industry is currently facing severe capacity constraints at TSMC. This bottleneck has prompted "hyperscalers" like Google to seek alternative foundries to ensure their hardware roadmaps are not stalled by a single point of failure in the supply chain.

Industry Implications

By scaling its internal chip production to these levels, Google significantly reduces its dependency on external vendors and lowers the long-term cost of running its massive AI models. The move allows the company to maintain a tighter loop between software optimization and hardware design, providing a competitive edge in the race for generative AI supremacy.

Furthermore, a commitment of this scale serves as a critical validation for Intel. As Intel attempts to regain semiconductor leadership through its Intel Foundry services—specifically its 18A process—securing a high-volume client like Google proves that it is a viable high-end alternative to TSMC. This shift could potentially alter the geopolitical and economic balance of global semiconductor manufacturing by introducing a second major player capable of producing cutting-edge AI silicon.

What to Watch

While the order signals a strategic pivot, the industry will be watching to see if Intel can meet the rigorous yield and performance requirements of Google's next-generation designs. Some analysts suggest Google's production volume could eventually rival Nvidia's sales, though such projections remain unconfirmed. The primary focus remains on whether the 2027-2028 production window will successfully break the current foundry monopoly.