www.igorslab.de, May. 26, 2025 –
In an interview published in the Korean tech medium Chosun Biz, AMD’s Senior Vice President took a clear stance on the company’s future manufacturing strategy: From AMD’s perspective, TSMC is currently the undisputed leader in 2nm process technology. This means that the Taiwanese contract manufacturer is once again the preferred partner in the current technology round – at least as far as AMD’s upcoming server processors are concerned.
Zen 6 on N2: EPYC Venice as technology carrier
A few weeks ago, AMD announced that it would be the first customer to use TSMC’s 2nm process (N2). It is now clear that this is the next generation of EPYC server processors, codenamed Venice. These are to be based on the new Zen 6 architecture and will be officially launched on the market in 2026. They are currently being developed and manufactured in close cooperation with TSMC – and, according to AMD, with a focus on efficiency per watt and maximum performance within the respective node. A small side note that is nevertheless not insignificant: Lisa Su, CEO of AMD, visited the TSMC plant in Arizona especially to accompany this milestone in a media-effective manner. Whether this was a symbolic closing of ranks or a tactical signal to potential second suppliers remains to be seen.
Figures and segment shifts
The development of the data center segment was also discussed in the same breath. According to AMD’s SVP, the company recorded growth of 57 percent in this area in the first quarter of 2025. These figures underpin the focus on scalable server solutions, which is reflected in the broader product range, among other things: With the EPYC-4005 series (“Grado”), the company now wants to make greater inroads into the mid-range and low-end segment – an area that has previously been rather neglected. Leaks about the technical data of the new Zen 6 family are particularly interesting here: up to eight CCDs (Core Complex Dies), 96 classic cores and 256 so-called “dense cores” are being discussed, flanked by 128 MB L3 cache per CCD. It remains to be seen whether this represents the final expansion stage or whether AMD is once again pursuing a variant strategy here, as we know it from the Ryzen and Instinct divisions.
No exclusivity – Samsung remains in discussion
Despite the clear preference for TSMC, AMD is leaving it open as to whether other foundry partners could also come into play. Samsung in particular is mentioned as a possible second supplier. According to the AMD manager, the company is always in talks with suppliers “who can offer the best services for our customers”. There is therefore no exclusive partnership with TSMC – at least not on paper. Samsung Foundry has repeatedly communicated progress with 2nm and underlying nodes (e.g. SF2 and SF1.4) in recent months, even if concrete design wins have so far been in short supply. Particularly in the server segment, where validation cycles are long and risk aversion is high, an actual switch from AMD to Samsung will probably only take place if the technical and economic pressure is correspondingly high – for example due to supply bottlenecks or geopolitical considerations.