Meta’s In-House Vistara Chip Reuses Old Memory to Cut AI Hardware Costs

Meta developed Vistara, a Compute Express Link (CXL) memory-expansion chip, to reduce the cost of upgrading its AI data centers. The chip allows new servers using DDR5 to reuse DDR4 modules from retired equipment. The design can reduce waste from hardware replacement, making it particularly important for expanding AI inference infrastructure.

Meta has deployed Vistara on its MemServer platform to support demanding workloads, including disaggregated machine-learning inference. CXL connects capacity from older DDR4 memory to next-generation DDR5 hosts. Available information does not disclose a formal release date, the scale of deployment or the amount saved, but the technology has moved from chip development into operational data-center use.