Nvidia is reportedly phasing out several Jetson developer platforms built around LPDDR4 memory, and the culprit is a familiar one: supply constraints. The affected Jetson Xavier and TX2 family is already old enough that a retirement notice would have made sense on its own; the memory shortage is simply speeding up the clock.
The models named in the notice include Jetson TX2 NX, TX2i, AGX Xavier with 32 GB in its industrial version, and Xavier NX in 8 GB and 16 GB variants. A Canadian distributor says all Xavier and TX2 modules are now in end-of-life status, with orders accepted through 1 July and final shipment scheduled for 15 July next year.
Jetson Xavier and TX2 are being retired
These Jetson boards first appeared between 2017 and 2021, which is a long enough run for embedded hardware to become somebody else’s maintenance headache. Nvidia’s cleaner path is to push customers toward Orin NX or AGX Orin, where the company can keep shipping newer silicon instead of fighting for memory that manufacturers would rather devote to DDR5 and HBM.
What customers need to change
The migration is not a pure drop-in swap. Users moving from Xavier or TX2 to Orin-based modules may need to adjust power delivery and thermal design, which is the polite way of saying the board may fit but the system probably will not behave exactly the same.
- Order cutoff: 1 July
- Last shipment: 15 July next year
- Affected families: Jetson Xavier and TX2
- Suggested replacements: Orin NX and AGX Orin
LPDDR4 is losing out to newer memory
The broader story is not Nvidia being unusually ruthless. LPDDR4 is simply less attractive for memory makers than DDR5 and HBM, so older embedded platforms that depend on it get squeezed first. Nvidia could have kept Jetson Xavier and TX2 around longer, but shortages have a way of turning ”sometime later” into ”right now.”
The obvious question is how many other aging developer kits are next. If the memory market keeps rewarding newer standards, expect more industrial and edge-computing hardware to age out faster than its compute silicon does.