Google Recycles DDR4 Memory from Retired Servers to Address Memory Shortage

ChainCatcher reports that Nikhil Cherian, Senior Director of Supply Chain Infrastructure at Google, revealed that to overcome memory bottlenecks, Google is developing both software and hardware solutions, and is dismantling decommissioned servers to recycle DDR4 components, establishing an internal recycling supply chain. Google has designed specialized hardware adapters that enable the connection of legacy DDR4 memory to next-generation AI servers, while also importing decommissioned servers to extract and recycle their DDR4 modules. Cherian noted that the AI industry has rapidly shifted from being compute-constrained to memory-constrained, with high-performance memory accounting for approximately 75% of the BOM cost of a given AI server. Goldman Sachs reports indicate that memory prices will continue to rise in the third quarter, with PC DRAM prices expected to increase by 18% to 23% and server DRAM prices by 13% to 18%. Trendforce data shows that in August, DDR4 8GB and DDR5 8GB spot market prices rose to $142 and $133, respectively. Google’s two newly introduced TPU ASICs this year have been optimized for memory efficiency, aiming to reduce memory requirements to one-sixth of previous levels. The TPU8i chip features a dedicated hierarchical memory design, relying on high-speed DDR5 memory architecture to perform host-level tasks, with each chip equipped with 288GB of HBM3e high-bandwidth memory.