Back in September, one of our customers scoped a data center refresh at $7.4 million.
Today, well, that same project is worth $27 million.
Nothing changed in the design. It had the same infrastructure, same requirements, and the same outcome. The components just changed in value. Most of that jump comes from memory, but storage device pricing is climbing too as availability tightens.
What the DDR4 dip means
Over the past few weeks, there have been early signs of softening in DDR4 pricing. Buyers have paused. Inventory is being released. Some segments are flattening. On the surface, it looks like relief, but it isn’t.
The growth of AI is creating a situation where demand is well exceeding supply. Manufacturers are reallocating capacity toward DDR5 and HBM for AI workloads, and legacy components like DDR4 are no longer a production priority. They’re still widely used across enterprise systems and embedded infrastructure, but nobody is making more of them.
DDR5 is the latest generation, so it’s not really entering the secondary market yet. DDR4 has been around for years, so theoretically there should be more of it coming out of decommissioned hardware. But because OEMs are directing new production toward AI, companies that previously bought DDR4 new are being squeezed into secondary markets to find it. DDR4 prices are going up even as the technology ages out.
Prices spike, stall, dip, and spike again. We’ve seen one product range where the price per chip went from $2 to $17. Exact same chip with the exact same spec. You can’t plan around that kind of movement.
The cost isn’t the hardest part
When prices rise steadily, companies adjust. They budget, forecast, and plan around it.
The problem is when prices move unpredictably, week to week, and sometimes even day to day. Today’s price is not tomorrow’s price, and today’s availability is not tomorrow’s availability.
We’re seeing more cases where customers don’t have time to wait. If allocation isn’t secured or stock isn’t reserved, it gets picked up by someone else almost immediately. Procurement ends up reacting instead of following a plan, engineering has to adjust to whatever’s available, timelines start tightening, and the original margin assumptions stop lining up with reality.
What really causes problems is how quickly control over the situation starts to slip.
Where the supply is coming from
If you can’t reliably buy what you need new, you have to find it somewhere else. And increasingly, that somewhere else is internal. Recovered components, chips extracted from retired hardware and put back into circulation.
We have a client right now who manufactures a high-volume consumer product. They need specific chips to keep their production line going, and they can’t get them new. So we’re sourcing those chips from completely different manufacturers and market sectors, recovering them, validating them, and delivering them into their supply chain. If they can’t keep that production flow going, their customers go somewhere else. There’s no room to wait.
There’s nothing controversial about reusing DDR in secondary markets. It’s a functional product, and reuse is standard.
Chips are recovered, reballed, integrated into new modules, and put through full validation before redeployment. In most cases, you wouldn’t know the difference between a recovered chip and a new one.
How this scales
We’ve been pulling DRAM and NAND chips off SSDs since around 2016, back when certain product ranges made it impossible to reconfigure the devices any other way.
Our capacity in just our Penang location is ramping from 100,000 toward 800,000 chips per month. We built up the technology, the tooling, and the initial capacity in less than six weeks. We’re now working to roll it out across other regions.
There’s more to it than pulling chips off boards. You have to identify the right chipset, find and procure the full unit, salvage the chip without damaging it, validate it against clear parameters, and get it into a state where it can be redeployed. Every step in that process has to work or the volume falls apart.
Storage is next
If demand for memory is rising, demand for storage follows. Memory doesn’t store anything on its own. So NAND and SSD availability is already tightening in certain segments, and the same pressures are starting to show up.
NAND recovered from SSDs doesn’t have to go back into an SSD. It can go into another lifecycle activity entirely, which gives it more flexibility than a lot of people realize.
We’re also seeing more clients ask for box-in-box solutions, where they want recovery capability deployed inside their own facilities. If you can’t get hold of a new drive, you need another source. Having that on-site means you can redeploy drives into second use without shipping them out and waiting.
The compliance part is simpler than people think
When it comes to recovered components, there are really three situations.
If it’s data-bearing, you need a clear methodology to erase, certify, and validate. That’s where platforms like Proteus and Rackwipe come in, which are controlled environments for secure testing and certified data erasure with full traceability.
If it’s IP-protected and proprietary to a specific client, and it’s not going back to that client, you destroy it. There are no exceptions here.
And then there’s the third category, which covers most cases: standard chip methodology. This is where tools like Maya and Bujang are used to test, validate, and qualify NAND and DRAM at the chip level before redeployment. Once validated, those components can be rebuilt into modules and returned to use within spec.
On DDR specifically, it’s not data-bearing. Chips in most cases aren’t a compliance issue at all.
Where things slow down is internal. Companies still have to work through their own compliance frameworks before they can act, and that’s where most programs stall.
The part nobody talks about
Most conversations about chip recovery focus on the technical side. Can you pull the chip cleanly? Can you validate it? Can you match it to spec? The answer to all of those is yes, and has been for years.
The bottleneck is organizational.
Getting internal approvals. Engineering sign-off. Quality verification. Legal review of reuse policies. Updating procurement frameworks that were written when “end of life” meant “shred it.” That process takes months. Sometimes over a year.
If supply tightens further in Q3 and your company decides in Q3 to start exploring recovery, you’re not recovering chips in Q4. You’re recovering chips in late 2027 at the earliest, after a year of internal process. And the whole time, you’re still exposed to a market where a $2 chip becomes a $17 chip overnight.
Almost everyone sees the value of recovery by now. The companies that benefit from it are the ones who started the internal approval process early enough that the capability was ready when the supply crunch hit. The technical side has been solved for years. The organizational side is what decides who has options.
The OEMs already see it
Companies that previously had no interest in hardware once it was out of warranty are now taking second life seriously. When OEMs start changing behavior around something they’ve ignored for decades, it tells you what they’re seeing in their own forecasts.
Ian Lovell, who runs our recovery operations, puts it simply: a sailor who sees a storm coming goes back to harbor.
The storm is visible. Current indicators suggest this runs well into 2027.
What to do about it
If your operations depend on DDR4, NAND, or SSD components, three things are worth doing now.
First, look at what you’re currently destroying. Run an audit of the components going through your end-of-life stream. A lot of companies are shredding material that has significant and growing market value, and they don’t have visibility into it because the destruction process was set up when that material was worthless.
Second, start the compliance conversation now. Talk to your engineering, legal, and quality teams about what it would take to approve recovered components for secondary use. You don’t need to commit to a program. You just need to understand how long the approval path is, so you’re not starting from scratch when you need to move fast.
This is something worth working through with Reconext directly. Our engineering teams are already helping customers map approval paths, identify constraints, and validate what’s actually feasible before they need to move.
Third, understand the storage timeline. If you’re feeling the memory pressure now, storage is coming. The same dynamics that drove DDR4 prices up are beginning to appear in NAND and SSD segments. Getting ahead of memory means you’re also getting ahead of storage.
The companies that get through this period won’t be the ones who called the market correctly. They’ll be the ones who had a secondary supply path ready before they needed it.
