Google reportedly places order with Intel for more than 3 million TPUs in 2028, boosting Intel Foundry hopes

Alphabet’s Google has reportedly ordered more than three million Tensor Processing Units, or TPUs, from Intel for production in 2028, according to The Information and people familiar with the discussions. TPUs are Google’s in-house AI accelerators designed to train and run large machine-learning models for Google Cloud and other systems. The report did not provide a contract value or details of the exact chip designs to be produced.

The reported order is expected to be manufactured through Intel’s contract chip manufacturing and advanced packaging capabilities, an area Intel has been emphasizing as it tries to rebuild its presence as a “systems foundry” for AI-era chips. Intel describes its foundry approach as pairing process nodes with assembly and test, and it highlights advanced packaging techniques aimed at high-yield production of multi-die “systems of chips.”

Multiple reports said Google’s decision followed months of testing Intel’s manufacturing and packaging capabilities. Mobile World Live said Google plans to use Intel to manufacture some of its TPUs after evaluating Intel’s ability to produce the chips, pointing to growing pressure on Taiwan Semiconductor Manufacturing Co. as AI customers look for additional supply paths.

In the Reuters report that circulated from The Information’s reporting, Intel declined to comment on the story, while Alphabet’s Google and Nvidia did not immediately respond to requests for comment. Reuters also stated it could not independently verify the report. That means the timing, scale, and whether the order is finalized or still subject to change are still unclear.

The backdrop for the reported shift is the stress on advanced semiconductor capacity. As AI demand has surged, the supply chain for leading-edge wafers and advanced packaging has become harder to scale quickly, according to reporting on the Intel order. That environment has encouraged some chip designers, including large AI-capable cloud operators, to explore alternative manufacturing partners rather than relying solely on TSMC.

For Intel, the potential order would represent a major proof point for its strategy to win business from chip designers that do not want to run their own factories. The Intel Foundry fact sheet frames the approach around full-stack support, including technology for chip design enablement and global assembly and test, and it describes its focus on advanced transistor architectures and packaging methods used for dense AI accelerators.

Watch points are straightforward but limited by what is not disclosed: the reported TPU order’s commercial terms, the specific TPU generations involved, which Intel process node(s) and packaging flow would be used, and whether production ramps are guaranteed. Until Google, Intel, or Nvidia provide more detail, the most immediate confirmation likely comes from future Intel guidance, customer announcements, or additional reporting that can independently corroborate the deal. If finalized, the order would also increase scrutiny of how quickly Intel can meet volume demand across wafer fabrication, advanced packaging, and test schedules for AI workloads in the late 2020s.