Micron is discontinuing its 2GB (16Gb) GDDR7 memory chips, ending both the 28 Gbps and 32 Gbps variants, and reallocating fabrication capacity toward higher-density 3GB (24Gb) GDDR7 silicon aimed at AI GPUs, according to Tom's Hardware, which cites an X post by UNIKO's Hardware first flagging the change. Tom's Hardware reports it confirmed the shift itself: the module pages on Micron's site now return "No results found" for the discontinued parts.
The mechanics here are straightforward capacity reallocation, not a technology failure. Micron had only begun supplying 2GB GDDR7 to Nvidia earlier this year, per Tom's Hardware, making it a late entrant against Samsung and SK Hynix, which the outlet describes as the main GDDR7 suppliers. That meant Micron's 2GB modules shipped on a narrow slice of the market — Tom's Hardware names retail variants of the GeForce RTX 5060 as one example — rather than across the standard RTX 50 lineup broadly.
The pivot target is 3GB (24Gb) GDDR7, which Tom's Hardware describes as impractical for Nvidia's RTX 50 consumer lineup because the larger packages are more expensive and produce less conventional consumer memory capacities. Instead, the outlet reports these higher-density chips are used in Nvidia's RTX PRO Blackwell GPUs built for generative and agentic AI workloads — parts that command higher margins than gaming silicon.
On supply risk, Tom's Hardware's own assessment is that the immediate impact on gaming GPU availability should be limited: since Samsung and SK Hynix already cover the vast bulk of 2GB GDDR7 modules for the standard RTX 50 retail lineup, Micron's exit removes a minority supplier rather than a primary one. The piece frames Micron's position in the 2GB segment as "uncompetitive," which it says made a fast exit "logical" to free up fab capacity for the 3GB line.
No pricing, yield, or capacity figures accompany the report — Tom's Hardware does not disclose wafer allocation percentages, unit volumes, or margin deltas between the 2GB and 3GB parts, so procurement teams should treat this as a directional signal rather than a quantified forecast.
The pattern matters more than the single SKU. Tom's Hardware ties the move to a broader trend of memory suppliers prioritizing AI-facing products over consumer parts, noting Micron previously exited its Crucial consumer memory and SSD business, a decision the company had explicitly attributed to surging AI data-center demand pushing it to prioritize larger strategic customers in faster-growing segments. Two consumer-adjacent exits from the same supplier in succession is the more telling data point than either move alone.
For architects building GPU procurement roadmaps, the risk is not this quarter's RTX 5060 supply — Samsung and SK Hynix cover that. The risk is medium-term: as more memory capacity gets pulled toward AI-grade, higher-margin die, teams sourcing consumer or edge-inference hardware should expect the pool of suppliers for standard-density memory to keep shrinking, and should start qualifying second-source components now rather than after a shortage shows up in lead times.