Google Orders 3 Million AI Chips From Intel for 2028, Nvidia Tests Foundry for Feynman
- Google ordered more than 3 million TPUs from Intel Foundry for 2028 production, the largest known customer commitment for Intel's contract manufacturing business [1]
- Nvidia is running early trials on Intel's 18A process for its Feynman GPU architecture but has not placed an order [2]
- Intel shares surged roughly 12% on the news, trading near $110.81 with a market cap of approximately $557 billion [4]
- SK Hynix is testing whether its high-bandwidth memory works reliably with Intel's advanced packaging [3]
- TSMC capacity constraints — particularly in advanced packaging — are pushing AI chip designers toward Intel as a second source [3]
Google has ordered more than three million of its next-generation tensor processing units from Intel for 2028 production, according to a report by The Information citing people with direct knowledge of the discussions [1]. The deal represents the largest publicly known customer commitment for Intel's foundry business, which has been struggling to attract outside clients for years.
Separately, Nvidia is running early-stage trials on Intel's 18A process node, testing whether Intel can manufacture a processor that fuses four graphics chips into a single package — a design tied to Nvidia's Feynman GPU architecture slated for 2028 [2]. Nvidia has not placed a production order with Intel.
Intel shares jumped approximately 12% on Monday to around $110.81, pushing the chipmaker's market cap to roughly $557 billion [4]. The stock has more than tripled over the past year, rising 441% from its 52-week low of $18.97 [4]. The rally reflects growing investor confidence in CEO Lip-Bu Tan's foundry turnaround strategy, which has now attracted validation from two of the world's most important AI chip buyers.
The Google Deal
Google's order covers more than three million TPUs — the company's custom AI accelerators used to train and run large language models across its data centers. Intel will act as a contract manufacturer, fabricating Google's chip designs rather than selling Intel-branded processors [1]. Morgan Stanley estimates Google's total TPU buildout across 2027 and 2028 exceeds six million units, suggesting the Intel order covers a substantial share of that pipeline [3].
The commitment follows months of testing by Google of Intel's advanced packaging capabilities, which stitch together multiple chip dies and high-bandwidth memory into a single module [3]. Advanced packaging has become the critical bottleneck in AI chip supply, with TSMC's packaging lines strained beyond capacity.
Nvidia's Foundry Tests
Nvidia's engagement with Intel remains exploratory. The GPU maker is running multiproject wafer trials on Intel's 18A process node, evaluating whether it can serve as a manufacturing backup for future products [2]. The test work is tied to Nvidia's Feynman architecture, a next-generation GPU design expected in 2028 that aims to combine four graphics chips into a unified processor [3].
No production order has been placed, and Nvidia continues to rely on TSMC for its current Blackwell and Rubin GPU lines. But the fact that Nvidia is actively evaluating Intel's most advanced process node signals that TSMC's near-monopoly on leading-edge AI chip manufacturing is facing real pressure [1].
TSMC's Capacity Crunch
The moves by Google and Nvidia stem from a fundamental supply problem: TSMC, which manufactures virtually every leading-edge AI chip in production today, cannot keep up with demand [3]. TSMC CEO C.C. Wei has said publicly that global chip supply will not meet AI-driven demand for years [3].
The worst bottleneck is in advanced packaging — the process of integrating multiple chip dies, memory stacks, and interconnects into a single package. This packaging step has become the chokepoint for AI accelerators from Nvidia, Google, Amazon, and others. The concentration of this capability at a single Taiwanese supplier is increasingly viewed as a strategic liability by major customers [3].
Intel's Foundry Turnaround
The Google order marks a turning point for Intel Foundry, which CEO Lip-Bu Tan has positioned as the cornerstone of the company's comeback strategy. Unlike prior announcements involving equity investments or memoranda of understanding, Google's commitment represents actual customer orders with volume attached [3].
Intel is not the only beneficiary of TSMC diversification pressure. Tesla CEO Elon Musk said in April that the electric vehicle maker plans to use Intel's next-generation 14A manufacturing process for chips at its Terafab advanced AI chip complex in Austin [2]. SK Hynix, the leading maker of high-bandwidth memory, is separately testing whether its HBM products work reliably with Intel's packaging technology — a secondary validation that strengthens Intel's credibility as a full-service foundry [3].
What's Next
The 2028 production timeline gives Intel roughly two years to scale its 18A process to high-volume manufacturing and prove it can match TSMC's yield and reliability standards. SK Hynix's packaging compatibility tests will serve as an important external benchmark [3].
For investors, the question is whether Google's order is the beginning of a broader shift toward Intel as a second-source foundry, or an isolated hedge against TSMC supply risk. Nvidia's decision on whether to move beyond trials to a production commitment will be the next major signal [1].
Further sources
The stories that matter, in one email. Free — unsubscribe anytime.