Google Orders 3 Million TPUs From Intel Foundry in 2028

Three million tensor processing units — that's the volume Google has committed to Intel's foundry for 2028 delivery, according to The Information. Nvidia is also poking around Intel's 18A process and advanced packaging for upcoming chips. This isn't a trial run; it's a real bet that Intel can compete with TSMC at scale.

Why Intel's 18A Process Matters

Intel's 18A is their next-generation manufacturing node, targeting performance and power efficiency comparable to TSMC's N2. Google and Nvidia haven't publicly benchmarked it yet, but putting three million TPUs on that line signals confidence. TPUs are custom ASICs for AI inference and training — not simple parts. If Intel can yield those at volume, the foundry business suddenly has a credible second source.

Nvidia's evaluation is even more telling. Their packaging and process requirements are among the toughest in the industry. If Nvidia qualifies Intel for future products, TSMC loses its exclusivity on high-end AI silicon.

What This Means for TSMC's Monopoly

TSMC currently fabricates the vast majority of AI accelerators from both companies. Intel's 18A order from Google represents capacity diversification — a hedge against TSMC's Taiwan concentration and geopolitical risk. Three million TPUs is roughly a year's worth of a data-center GPU line. It won't dent TSMC's revenue tomorrow, but it cracks the lock.

Intel's last foundry effort fizzled. This time they have a concrete anchor customer, a real process, and a deadline. If Intel delivers on 18A, the chip manufacturing duopoly between TSMC and Samsung could become a three-horse race by 2028.

Source: Google, Nvidia consider Intel as backup chip manufacturer: Report
Domain: economictimes.indiatimes.com