Component Shortages Push Hyperscalers to Refurbish, Not Retire

In less than a year, the market for refurbished data center components has flipped so completely that equipment once headed for resale channels is now being requested back by its original owners. That was the message from Lynn Jacobs of Sims Lifecycle Services, speaking at the E-Scrap Conference on Oct. 27 during a panel on circular data centers.

"If you asked us the same question probably 10 months ago, we'd have been in a totally different situation," Jacobs said.

Shortages of memory, storage and other hardware are making replacement harder to source, prompting Google and Microsoft to keep data center equipment in service longer. For ITAD companies and electronics recyclers, that shift could shrink the volume of gear flowing through conventional refresh cycles — while opening new demand for refurbishment support from hyperscalers.

How are the hyperscalers responding?

Google historically favored newer technology, refreshing infrastructure regularly and pushing older equipment into remarketing. Yara Barkopoulos of Google said that calculus has changed. Some older parts are no longer manufactured, making refurbishment necessary to keep existing systems running — a practice long common elsewhere in IT but rarely adopted at hyperscale.

"Now that we cannot acquire assets as easily as we had in the past, and we're expanding our business, we're really looking at ways how we can refurbish our inventory instead of retiring it," Barkopoulos said.

Microsoft works through a hierarchy, according to Sanyukta Saptasagar: first, whether retired components can be reused internally; then external buyers; recycling only when neither works. Jacobs said Sims is seeing hyperscale customers reclaim components previously destined for external markets — and recovered parts can go back to customers at lower cost than new hardware, especially when processing happens at scale.

Why is reuse harder than it sounds?

Refurbishment isn't like ordering new equipment to spec. Jacobs compared decommissioning to assembling Lego pieces: retired gear arrives in inconsistent quantities and configurations, often at times that don't match customer demand. Components must clear performance and compatibility requirements, and timing has to line up.

Custom hardware narrows options further. Barkopoulos said some data center components are built for individual companies, limiting resale potential. She suggested greater standardization would improve future reuse — though customization remains necessary when commercial alternatives can't meet technical requirements. Some devices still yield parts: Google recovers components and materials from inside solid-state drives unsuitable for reuse as whole units.

What about AI hardware specifically?

AI expansion is making lifecycles harder to predict. Saptasagar said newer AI technology may shorten initial deployments even though equipment stays useful for years — a GPU pulled from demanding training work could run inference or lighter workloads instead. Google sees similar opportunities to shift components between applications.

Jacobs pointed to neocloud providers as another demand source, using hardware that no longer meets the largest hyperscalers' requirements. Sims is fielding requests for complete racks as well as systems rebuilt around specific requirements. But most current AI hardware hasn't finished its first life: many GPUs Sims handles are returned to original customers because demand is strong. The industry has seen little of the eventual retirement volume from today's AI infrastructure.

Has the theft risk changed?

Yes, and materially. Five years ago, secondhand IT shipments attracted little attention from thieves. Today, some loads are attractive targets because component values have risen. Sims has increased shipment tracking and security, and Jacobs said insurance- and theft-related expenses have climbed — on top of higher freight costs already squeezing the business.

Where does data security fit in?

Rising storage values are prompting companies to reconsider mandatory physical destruction. Barkopoulos described Google's internally developed sanitization process: drives that pass can be evaluated for internal reuse or remarketing; those that can't be securely sanitized are shredded on-site. Jacobs said improved sanitization and reliable asset tracking are making customers comfortable with reuse — backed by auditable chains of custody. Microsoft, meanwhile, is working to scale recovery of valuable materials from shredded hard drives, including the rare-earth magnets inside them.

What's next?

Export controls remain an open question. Google prefers keeping recovered raw materials in the U.S. when possible; Saptasagar offered no prediction on future restrictions. Jacobs noted GPUs generally stay in their country of origin for now — but the industry won't know how restrictions bite until larger AI retirement volumes actually arrive. That timing, more than anything else, will decide what material reaches ITAD and recycling markets.