For most of the last decade, memory was the one line on the BOM nobody thought about twice. That’s changing. DRAM and NAND supply is being redirected toward AI infrastructure at a pace that’s leaving conventional parts short and allocation increasingly vendor-specific. NAND is expected to loosen first, with new capacity easing that side of the shortage sometime in the second half of 2027. DRAM is the harder problem: most manufacturer and analyst forecasts now put continued tight supply through 2028, and some run considerably longer than that.

For hardware teams, that turns a stable line item into a design risk overnight. A board locked to one vendor’s package, or a product built on a single-source module, inherits that vendor’s allocation problem the moment supply tightens. A large forward buy can hedge against that for a while, but it ties up capital, adds inventory risk if the design changes, and doesn’t help once a vendor is rationing allocation and won’t fill a multi-year order at any price. The design itself is still the constraint. That risk shows up as a phone call from procurement telling you the part you’ve built six months of production around isn’t available, on any timeline anyone will commit to.

We’ve seen this movie before. During the COVID-era component shortage, the crunch wasn’t memory, it was microcontrollers and connectivity modules. Lead times on common MCUs pushed past a year at the peak, and clients who’d built products around a specific part suddenly needed that firmware running on hardware they could actually buy. We ported firmware twice in that window for a physical security manufacturer: once moving a lock controller from an NXP Kinetis MCU to a Microchip SAM D51, once moving Linux gateway firmware from one Digi ConnectCard variant to another after the client could no longer purchase the original module. Different products, different architectures, different toolchains, both under real schedule pressure.

That’s harder than what a memory shortage typically demands. Swapping an MCU means re-porting firmware, re-validating peripherals, and re-testing behavior that may depend on timing quirks specific to the old part. Swapping memory is closer to a footprint and signal integrity problem: get the pinout compatible, get the power budget right, and the firmware mostly doesn’t care what’s underneath it. If we can do the harder version, the memory version is well within reach.

“Both of those projects started the same way: a client who’d built a product around one part finding out the lead time on it had stretched past a year. Porting firmware across MCU families under a schedule you didn’t choose is not a small job. It’s the same underlying problem we’re seeing now with memory, just harder, because with an MCU the whole software stack rides on the part you’re replacing.”

— Luke Johnston, Chief Operating Officer

• Second-source / pin-compatible redesign. When a board is locked to one memory vendor’s specific package and pinout, an allocation problem becomes a production stoppage. We re-layout for footprint-compatible alternates across DRAM and NAND vendors so procurement has more than one supplier to call.
• SOM decoupling. A design pinned to a single system-on-module inherits that vendor’s allocation problem wholesale. Where it fits, we design the carrier board to an open standard like SMARC (Smart Mobility ARChitecture), which multiple vendors build to, instead of one proprietary module. That’s not a guarantee of true plug-and-play, since individual vendors’ pinouts can still vary within the spec, but it gives procurement more than one supplier to call, without giving up the reliability and support of a dedicated SOM vendor.
• Memory type substitution. LPDDR vs. DDR, eMMC/UFS vs. raw NAND, and for smaller storage needs, MRAM or FRAM as an escape hatch from the NAND crunch entirely. These aren’t drop-in swaps. They change power budget, PCB routing, and sometimes the SoC choice, but they’re real options depending on what the product actually needs to store.
• Firmware and architecture rightsizing. A lot of designs carry more RAM and flash than they use, because it was cheap and available when the board was spec’d. Shrinking the real memory requirement opens up parts tiers that aren’t as squeezed, and it’s a service a component distributor or EMS shop can’t offer.
• Proactive BOM risk audit. Reviewing a BOM now for parts exposed to allocation or end-of-life risk, before it’s a fire drill, instead of after.

The shortage moved from micros to memory this time. The pattern underneath it, a supply chain concentrating around fewer sources while the design stays locked to one of them, is the same problem we’ve solved before. If your product depends on a memory part you can no longer count on, that’s a design constraint, not just a procurement problem, and it’s one we can help you get out from under before it becomes a production stoppage.