Samsung Electronics is reportedly in talks to make a key portion of one of Google's most advanced future artificial intelligence chips. The deal would give the South Korean company a role in a supply chain as the industry is being forced to widen, with Taiwan Semiconductor Manufacturing Co. running out of capacity to meet demand.
Google is considering using Samsung's 2-nanometer process to produce a memory input-output die for its 10th-generation tensor processing unit, the in-house chip Google runs in its cloud data centers. The Information reported the talks, citing two people with direct knowledge of the project. The chip, code-named "Icefish," could reach mass production as early as 2028, though the plan is still in the design stage and could change.
Asked about the report, a Samsung Electronics official told The Korea Herald the company could not confirm anything.
The work would reportedly be divided between TSMC and Samsung. TSMC would make the computing engine, the most demanding part, on its 1.4nm technology. Samsung would make the I/O die that connects the processor to high bandwidth memory, a link The Information called critical because AI chips need a steady flow of data to keep their cores busy. The reason that particular job would land with Samsung is not spelled out in the reporting, but the logic is easy to follow. No company knows memory better than the world's largest maker of it.
The talks fit a broader pattern.
TSMC's most advanced production lines are fully booked, and CEO C.C. Wei told shareholders earlier this month that chip supply would not keep pace with AI demand for years. Google has separately ordered more than 3 million TPUs from Intel for delivery in 2028, The Information also reported earlier this month, and is working with Taiwan's MediaTek on the Icefish design.
Both Samsung and Google declined to comment to the outlet.
The timing helps Samsung, which has struggled to close the gap with TSMC since setting up its foundry unit in 2017. Its contract chipmaking division is expected to return to profit as early as the third quarter, according to industry estimates reported by Korean media. That would be its first profit since trillion won losses began in 2022, helped by better 2nm yields and rising orders.
The company also won a 22.8 trillion won ($15 billion) long-term contract from Tesla last year for next-generation AI chips, due to enter mass production at its new fab in Taylor, Texas, in the second half of this year. It is making Nvidia's new language processing unit as well.
TSMC's share in the global pure foundry market rose to 73 percent in the first quarter from 69 percent in late 2024, while Samsung's slipped slightly to about 7 percent, Counterpoint Research data show.
mjh@heraldcorp.com