According to Taiwan's "Business Times", AMD has started supply chain preparations for its next-generation x86 CPU architecture Zen 7 (codenamed Grimlock) in advance. The new product will use TSMC's cutting-edge A14 process, which is the 1.4nm process node, and is targeted for mass production in 2028. TSMC's A14 will compete head-to-head with Intel's 14A process during the same period, and the Zen 7 series is seen as AMD's key product line at this node.
The report pointed out that the current Zen 6 architecture has not yet officially entered the mainstream server and consumer markets, but AMD has launched mass production on TSMC’s 2nm process and is simultaneously deploying more advanced nodes, requiring supply chain partners to prepare production capacity and technology reserves in advance for the Zen 7 era. TSMC’s Fab 25 P1 factory in the Dazhong Science and Technology Park is expected to enter the trial production stage in 2027 and enter mass production in 2028, which will provide the basis for the introduction of the A14 process in the field of high-performance computing.
The news also stated that AMD CEO Su Zifeng recently visited a number of supply chain and industry partners, including Powertech, during his recent visit to Taiwan. The visit to Powertech is considered to be related to the allocation of advanced packaging production capacity. AMD is evaluating the use of LiCheng's FOPLP (fan-out panel-level packaging) solution to meet the design requirements of the new generation Zen 7 platform in terms of bandwidth, power consumption and stacked cache.
Supply chain sources revealed that Zen 7's core chip (CCD) - the Grimlock die - will be manufactured based on TSMC's A14 process and integrate next-generation 3D V-Cache technology to further amplify the cache advantage. On this basis, the Zen 7 CCD design is rumored to support up to 16 cores, and the L3 cache capacity of a single 3D V-Cache CCD can be up to 224MB, which means that the total capacity of conventional L3 and stacked caches will be further improved compared to existing products.
In the server market, the Zen 7 architecture will also upgrade the capabilities of the MATRIX engine and expand the supported AI data formats to better adapt to the current rapidly evolving artificial intelligence workloads. Against the background of the continued extension of the so-called "AI super cycle", the demand for CPUs in data centers and AI infrastructure is believed to remain high for a long time, prompting AMD to further increase its data center business layout like its main competitors.
Industry analysts believe that whether TSMC can advance the A14 process as planned in 2028 will have a direct impact on AMD Zen 7's launch pace and performance competitiveness. As Intel has recently received more customer support in its foundry business, including Apple and TeraFab, which have confirmed the adoption of its 18A-P and 14A processes, this makes the collaboration between TSMC and AMD on high-end HPC and AI nodes particularly critical.
In AMD's existing CPU core roadmap, Zen 4 and Zen 4c use 5nm/4nm processes, and gradually improve performance and AI capabilities through Zen 5 and Zen 6. Zen 7 will serve as a new generation node to further enhance computing and AI support. With the advancement of Zen 7, AMD is expected to enter into more fierce head-to-head competition with competitors in the global CPU market of approximately US$200 billion in the next few years.