THE CRUNCH
Intel chief executive Lip-Bu Tan said on Thursday that yields on the company's 18A process, a 1.8nm-class fabrication technology, have reached 70%, up from what he described as a horrible state when he took over in March 2025. Speaking at the TomorrowX conference, he put the figure at 70%, which the report characterised as more or less in line with expectations for an all-new manufacturing technology at this stage of its ramp. Intel began volume production of 18A compute tiles for its Core Ultra 300-series Panther Lake processors in the second half of 2025, and its Xeon 6+ Clearwater Forest server chips, with up to 288 cores, are in high-volume production on the same node.
The same event brought a commercial win: Microsoft has signed a multi-year agreement to secure Intel Xeon supply capacity, including the next-generation Diamond Rapids server processors, and will use Intel's EMIB advanced packaging technology for data centre solutions it develops in-house. EMIB, which stitches multiple chiplets together on a package, matters because high-end data centre designs increasingly use multi-die processors. Microsoft is the first company to publicly confirm it will use EMIB for its own silicon, a notable publicity result for Intel Foundry, although the announcement does not identify the processors involved, their production volumes or a deployment schedule.
There is some ambiguity in the yield claim. Lip-Bu Tan did not specify whether the 70% figure refers to functional yield, parametric yield or the share of sellable silicon, though it is reasonable to assume a semiconductor chief executive means functional die yield, the percentage of dies passing basic testing. Intel's economics also depend heavily on parametric yield and how many premium Panther Lake variants it can supply. As a separate, differently-scoped data point, a September report put the yield of the Panther Lake compute tile, a relatively small die of about 114 square millimetres, at a reported 80%; that figure covers one specific tile rather than the 18A process overall, and the measurement basis may differ from the CEO's claim.
The collaboration goes beyond buying chips. Intel and Microsoft are jointly developing infrastructure capabilities spanning processor architecture, heterogeneous integration, CXL memory expansion, QuickAssist Technology acceleration and Trust Domain Extensions security. Microsoft had already committed in 2024 to use 18A for its data centre silicon, though it is unclear whether the new EMIB agreement covers the same solution.


