Google has committed to purchasing more than three million of its custom-designed Tensor Processing Units from Intel, with delivery stretching through 2028. The deal positions Intel as a major contract manufacturer for one of the world’s largest AI hardware buyers and signals a shift in how tech giants secure chip supply. For Intel, the agreement represents a high-profile validation of its foundry ambitions at a time when the company is fighting to regain relevance in advanced semiconductor manufacturing.
Why a three-million-unit TPU order reshapes chip supply chains
The sheer volume of the order sets it apart from routine foundry contracts. Three million TPUs is a massive production commitment that locks Intel’s fabrication capacity into Google’s roadmap for years. Google designs its own TPUs but relies on outside manufacturers to build them, and this deal suggests the company is actively diversifying away from sole dependence on Taiwan Semiconductor Manufacturing Company, which has historically produced the bulk of custom AI silicon for major cloud providers.
The timing matters because demand for AI training and inference chips has outstripped available supply across the industry. Companies like Google, Microsoft, Amazon, and Meta have all raced to secure fabrication slots, and the competition for cutting-edge manufacturing capacity has driven up costs and wait times. By placing a multi-year order with Intel, Google gains a second reliable source for its most important internal hardware, reducing the risk that any single supplier bottleneck could slow its AI infrastructure buildout.
Intel, for its part, has spent billions retooling its factories under a strategy to become a world-class contract chipmaker for outside customers. Landing Google as a high-volume client validates that strategy in a way that smaller or lower-profile contracts cannot. The deal also sends a competitive signal to TSMC and Samsung, the two other companies capable of manufacturing chips at the most advanced process nodes. If Intel can demonstrate that it can deliver complex AI accelerators on time and at scale, it strengthens its case for winning additional marquee customers.
Primary sources behind the Intel-Google TPU deal
The order was first reported by The Information and then confirmed through secondary reporting. Coverage from Reuters cited The Information, stating that Google placed an order with Intel to manufacture more than three million Tensor Processing Units for delivery through 2028. The same reporting noted that Nvidia has also been evaluating Intel’s manufacturing technology, though Nvidia had not finalized a comparable commitment at the time of the report.
Intel’s own regulatory filings provide additional context for the company’s foundry trajectory. The company’s quarterly disclosure in an SEC Form 10-Q offers a window into its financial position, capital spending, and manufacturing investments. A separate corporate certification filing documents governance and reporting changes tied to Intel’s evolving business structure, including its effort to operate the foundry arm with greater transparency.
These filings do not break out the Google contract by name, and the company has not publicly listed the order in its customer concentration disclosures. Nonetheless, the rising capital expenditures and references to large-scale foundry engagements in those documents are consistent with Intel preparing to support major external clients. Analysts tracking Intel’s shift from an integrated device manufacturer to a more open foundry model see the Google deal as the clearest evidence yet that its strategy is resonating with top-tier buyers.
Google’s TPUs are purpose-built processors designed to accelerate machine learning workloads. Unlike general-purpose GPUs sold by Nvidia, TPUs are optimized specifically for Google’s internal AI frameworks, including the infrastructure behind its Gemini family of large language models and a wide range of search, ads, and cloud services. Each new generation of TPU has grown in complexity and transistor count, which means fabricating them at scale requires access to advanced manufacturing nodes, the kind Intel has been investing heavily to develop.
Open questions around Intel’s capacity and Google’s chip strategy
Several significant unknowns remain. The financial terms of the deal have not been disclosed publicly, so the per-unit price Google is paying and the total contract value are not yet clear. Without those figures, it is difficult to assess how profitable the arrangement will be for Intel or how it compares to what Google pays TSMC for similar work. Investors will be watching whether Intel’s gross margins improve as foundry revenue grows, or whether aggressive pricing to win marquee customers weighs on profitability.
The specific process node Intel will use to manufacture the TPUs has also not been confirmed in public reporting. Intel has been developing successive generations of advanced manufacturing technology, and the node chosen for this contract will determine both the performance characteristics of the finished chips and the margins Intel can extract. If the TPUs require Intel’s most advanced nodes, the deal could serve as a proof point that outside customers trust Intel’s newest technology for mission-critical AI workloads. If the chips are built on older, more mature nodes, the strategic significance would be somewhat different, emphasizing capacity and diversification over absolute performance leadership.
There is also the question of whether Google intends to shift a larger share of its total TPU production to Intel over time or whether this order represents a fixed, bounded engagement. Google has historically kept its supply chain flexible, splitting orders across multiple foundries to hedge against geopolitical risk, manufacturing hiccups, and pricing shifts. The company could use this contract as a trial run before committing to deeper integration with Intel’s foundry services, or it could treat Intel as a permanent second source alongside TSMC.
Nvidia’s parallel evaluation of Intel’s manufacturing capabilities adds another layer of uncertainty. If Nvidia were to place its own large-scale order with Intel, it would represent a dramatic reshaping of the chip manufacturing market, where TSMC has long held a near-monopoly on producing the most advanced AI processors. Such a move could ease some of the capacity constraints that have plagued AI chip buyers, while also forcing foundries to compete more aggressively on price and technology roadmaps. But as of the most recent reporting, Nvidia had not finalized such a deal, and the company’s ultimate decision could depend on how well Intel executes on the Google contract and whether it can consistently hit yield and performance targets.
For investors and industry watchers, the next concrete milestone to track is Intel’s quarterly earnings and any updated guidance that reflects foundry revenue attributable to large external customers. Management commentary around capacity utilization, customer mix, and capital intensity will help clarify how central the Google order is to Intel’s business model. On Google’s side, disclosures about capital expenditures for technical infrastructure and comments on AI training and inference efficiency may offer indirect clues about how quickly Intel-manufactured TPUs are entering production and how they compare with existing TSMC-built parts.
Over the longer term, the deal underscores how AI is reshaping the semiconductor landscape. Cloud providers are no longer just buying standard processors; they are designing custom accelerators and using their purchasing power to influence where and how those chips are made. By aligning with Intel, Google is both securing critical capacity and helping nurture an alternative to the dominant Asian foundries. Whether this experiment delivers the performance, economics, and resilience both companies are betting on will be one of the defining questions for the next phase of the AI hardware race.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
