Alphabet looks to Intel for millions of future TPUs, indicating a potential shift in AI chip supply chains

Alphabet’s Google is reported to have placed an order with Intel to manufacture more than three million Tensor Processing Units, or TPUs, for delivery in 2028, a sign that Google is working to diversify the manufacturing of its custom AI chips. The report, cited by Reuters on June 8, said the deal would support Intel’s expanding contract chip manufacturing business and potentially reduce Google’s reliance on TSMC, which has struggled to keep up with AI demand.

According to the Reuters report, Intel declined to comment, while Alphabet and Nvidia did not immediately respond to requests for comment. Reuters also said it could not independently verify the report. Still, Intel shares rose more than 9% in early trading on Monday, extending a dramatic rally tied to signs that its manufacturing turnaround under CEO Lip-Bu Tan is gaining traction.

The underlying market dynamic is supply constraint. Reuters said soaring AI chip demand has left TSMC struggling to bring in adequate supply and that the capacity crunch has prompted major AI chip designers to look toward Intel. One technology analyst cited in the report described the news as evidence that the largest players in AI are racing to diversify a supply chain still heavily concentrated in TSMC.

TPUs are Google-designed AI accelerators used to train and run machine learning models across Google’s own products and Google Cloud. Google Cloud describes TPUs as custom chips purpose-built to accelerate machine-learning workloads, including for inference and training at scale. Google has also been positioning TPUs as part of its cloud growth strategy, including efforts to make it practical for enterprise customers to obtain TPUs directly for their own data centers.

Those choices matter for investor expectations because custom silicon is tied to both compute availability and monetization. In its June 2026 investor presentation, Alphabet said it is expanding beyond hosted cloud infrastructure to deliver TPUs directly to select enterprise customers, describing it as a new addressable market. The same presentation pointed to a large Google Cloud backlog and indicated that a substantial portion of that backlog is expected to be recognized as revenue over the following two years.

Reuters also noted that Nvidia is evaluating whether Intel’s technology can be used to make a processor that combines four graphics chips into a single unit, though Nvidia has not placed an order yet. That detail suggests Intel’s foundry ambitions are beginning to attract interest not only from hyperscalers seeking backup capacity, but also from chip ecosystem partners planning next-generation architectures.

Notably, the report did not disclose whether the arrangement involves full wafer fabrication, advanced packaging, or both, nor did it provide pricing, production locations, yield expectations, or the specific TPU generation targeted for 2028. With Alphabet and Nvidia declining to comment and Reuters stating it could not independently verify the information, the magnitude and operational impact of any “Intel TPU” ramp remains uncertain.