Meta starts building servers with old DDR4 memory amid DRAM shortage

Meta (the company's social networks are designated extremist and banned in Russia) has started building new servers from used DDR4 RAM modules recovered when old machines are decommissioned. The move partly addresses the shortage of server DRAM, CNews reports.

How the memory reuse works

The engineering problem isn't trivial. Old DDR4 modules aren't directly compatible, physically or electrically, with new server platforms designed for modern memory types such as DDR5. To bridge legacy and current hardware in a single system, Meta uses a custom adapter built around an application-specific chip (ASIC) called Vistara. The chip acts as a "translator" between memory protocols of different generations, letting a new server address and use the old modules as if they were its native RAM.

• Memory source: DDR4 modules recovered from Meta's decommissioned servers
• Compatibility: a custom adapter based on the Vistara ASIC
• Goal: partly offset the server DRAM shortage without buying new modules

Why the market is short on memory

The global RAM shortage of 2026 is largely driven by the AI boom. Memory chipmakers Samsung, SK Hynix, and Micron are shifting production capacity to HBM (high-bandwidth memory mounted next to AI accelerators such as Nvidia H100 and Blackwell), which sells at higher prices and is in greater demand than conventional DDR memory for ordinary servers and PCs. As a result, less conventional DRAM is physically being made, and its prices are rising for everyone — from hyperscalers running data centers to consumer PC builders and smartphone makers.

For companies like Meta, which are building giant AI data centers while still upgrading conventional server infrastructure for social networks and recommendation systems, this creates double pressure: new AI clusters need HBM, while ordinary servers need DDR4 or DDR5, which is physically getting scarcer on the open market. Reusing memory from decommissioned hardware is a practical engineering way to work around the shortage at least partly, without waiting for new shipments from manufacturers or overpaying for them as prices climb.

An ASIC (application-specific integrated circuit) is a chip designed for one specific task, unlike general-purpose processors. This narrowly specialized approach typically delivers better energy efficiency and lower cost than more expensive, general-purpose controllers — which matters especially when a solution is scaled across thousands of servers in a hyperscaler's data centers.

The approach also has a positive side effect: reusing still-working memory modules instead of scrapping them right away reduces data-center e-waste, a topic major tech companies increasingly raise in their sustainability reports. If the solution proves successful, other hyperscalers facing the same DRAM shortage amid the AI boom may adopt similar component-reuse practices.

What it means

The memory shortage triggered by AI infrastructure is pushing even the largest internet companies toward unconventional engineering fixes like reusing old hardware. It's an indirect but noticeable side effect of the global race for AI compute.