Intel will allegedly sink up to $14 billion into TSMC’s 3nm chip capacity. Mark Liu, the Foundry Board’s Chairman has informed key customers that Intel will order $4 billion worth of N3 (3nm) chip supply for its upcoming Lunar Lake processors. Previously, the American chipmaker used TSMC’s 6nm node capacity for its Arc A-series graphics cards, and soon the 4nm process for the tGPU of the 1st Gen Core Ultra Meteor Lake processors.
The 3nm order worth $4 billion will make Intel TSMC’s second-largest customer after Apple, beating AMD and NVIDIA to the punch. Additionally, 4nm and 3nm orders for the Meteor Lake/Arrow Lake tGPU and the 2nd Gen Arc Battlemage GPUs will further Team Blue’s reliance on the Taiwanese foundry. By 2025, Intel’s investment into TSMC’s 3nmm node is expected to total $14 billion.
The chiplet (or tiled) architecture gives Intel more flexibility over its supply chain. It can now utilize its flagship foundry node for the CPU tile (Intel 4 and 20A), outsourcing the tGPU and SoC tiles to TSMC. This reduces the load on its internal foundry, easing deadlines and leaving ample room for third-party foundry clients.
Intel’s five nodes in four years strategy won’t be possible without offloading a good chunk of the production capacity to third-party foundries. And so it has: The 1st Gen Arc Alchemist GPUs, Meteor Lake tGPU, Arrow Lake tGPU, and the entire Lakefield SoC. If the Arc discrete lineup isn’t downsized, we should see the 2nd Gen Battlemage family in 2024.
TSMC expects to fabricate 15,000 3nm wafers per month for Intel by the end of 2024 and scale it up to 30,000 the next year.