Intel will produce Microsoft's Maia AI processor using the 18A process technology.

New details about the cooperation have been revealed Intel Foundry and Microsoft: according to the latest reports, Microsoft's next-generation Maia AI processors will be built on Intel's 18A or 18A-P process technology.This is the first major order from an external customer at this stage of Intel Foundry's development and an important step toward competing with TSMC in the custom logic market for data centers.

Microsoft is actively developing its own line of chips., including Cobalt CPU, DPUs, and Maia AI accelerators used in Azure. The first Maia 100 chip is a gigantic 820mm crystal structure with 105 billion transistors, surpassing even NVIDIA's flagship solutions, such as the H100 and B200, in area. The next generation of Maia will likely continue to use crystals close to maximum EUV reticule size (858 mm²), which places special demands on the technological process, in particular, defects and yield.

If Microsoft did indeed choose 18A, it could indicate high quality and maturity of the technological process Intel. This choice is especially logical given TSMC's current capacity constraints, including in advanced packaging, and also considering that Manufacturing in the United States reduces geopolitical and logistical risks for Microsoft.

Interestingly, the possible integration of EMIB or Foveros technologies would allow Microsoft to split a large die into smaller chiplets, but this approach could lead to performance losses. Therefore, Intel and Microsoft are most likely following the traditional monolithic design path, using DTCO (Design-Technology Co-Optimization) and possibly embedding backup computing units to increase the yield of usable crystals - as NVIDIA does.