Intel emerges as potential alternative for AI chip production
Google and Nvidia are turning to Intel as a possible backup chip manufacturer as market leader TSMC struggles to meet soaring demand for artificial intelligence processors. According to a report from The Information, Google has already placed an order with Intel to produce more than three million of its custom AI chips, known as TPUs, for delivery in 2028.
Nvidia, the dominant supplier of AI training and inference chips, is testing Intel's manufacturing technology for its upcoming Feynman GPU architecture. However, the company has not yet committed to a production order.
The developments mark a potential lifeline for Intel's foundry division, which has been losing money and missing manufacturing milestones for several years. Intel has invested billions in its foundry business, aiming to become a major contract chip manufacturer alongside TSMC and Samsung, but has struggled to win significant orders from outside customers.
TSMC's capacity constraints open a window
TSMC's inability to keep up with explosive demand for AI chips is creating an opening for Intel. TSMC CEO C.C. Wei said on Thursday that global chip supply will not be able to meet AI-driven demand for years. That assessment has sent customers scrambling to secure alternative production capacity.
Intel's stock price jumped more than ten percent after the report of interest from Google and Nvidia surfaced, reflecting investor optimism that Intel may finally gain traction in the foundry market.
SK Hynix tests Intel's packaging technology
Memory chip maker SK Hynix is also evaluating whether its products work with Intel's chip packaging technology. If SK Hynix signs off, it would strengthen Intel's credibility as a viable TSMC alternative among other chip designers.
Chip packaging is a critical part of advanced processor production, especially for AI accelerators that combine multiple chiplets into a single package. Intel has developed its own advanced packaging technologies, including EMIB and Foveros, which could make it an attractive partner.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
A rare opportunity for Intel's foundry ambitions
Intel's foundry business has been on a difficult path. The company reported significant operating losses in its foundry segment last year, and delays in its manufacturing process roadmap have shaken customer confidence. The Intel 18A node, which the company has positioned as a comeback technology, is expected to enter production this year but has not yet landed major external customers.
Google's TPU order, even if placed years in advance, signals that at least one major cloud provider sees Intel as a credible supplier. Nvidia's testing of Intel's process for Feynman could lead to a more lucrative win if the technology passes muster.
TSMC remains the dominant force in advanced chip manufacturing, producing chips for Apple, Nvidia, AMD, Qualcomm, and most other major chip designers. But its capacity constraints are so severe that even its largest customers are looking for alternatives.
Google's TPU is a custom tensor processing unit designed specifically for AI workloads and used extensively in Google Cloud and internal AI operations. Moving some TPU production to Intel would diversify Google's supply chain and reduce its reliance on TSMC.
Nvidia's Feynman architecture is expected to follow the current Blackwell generation and the upcoming Rubin architecture. Choosing Intel as a manufacturing partner for Feynman would be a major coup for Intel's foundry business and a signal that its technology has reached parity with TSMC's offerings.
The Information report did not provide details on which Intel manufacturing process would be used for the Google order or the Nvidia test. Intel currently offers several process nodes, including Intel 18A and Intel 3, to external foundry customers.
SK Hynix's involvement is significant because the company is one of the world's leading memory makers and a key supplier of high-bandwidth memory (HBM) used in AI accelerators. If Intel's packaging technology works well with SK Hynix's memory chips, it could make Intel more attractive as a one-stop shop for AI chip production.
The report represents a notable shift in the chip manufacturing landscape. For years, TSMC has been the undisputed leader in advanced nodes, while Intel has struggled to compete. Now, a combination of TSMC's capacity limits and Intel's renewed manufacturing efforts may finally give customers a viable alternative.