Micron Technology (MU) has quietly pulled the plug on its 2GB GDDR7 memory chips, a move that removes one of three suppliers from NVIDIA’s GeForce RTX 50 series graphics card supply chain and underscores how aggressively memory makers are redirecting capacity toward higher-margin AI silicon.

The Boise, Idaho-based company has designated both the 28 Gbps and 32 Gbps variants of its 2GB (16Gb) GDDR7 chips as end-of-life, according to a post on X by hardware leaker UNIKO’s Hardware. Micron’s product catalog now lists the 28 Gbps D8FHL chip, model MT68A512M32DF-28, with EOL status, and module specification pages return no results, Tom’s Hardware confirmed.

Micron only began shipping GDDR7 to NVIDIA in meaningful volumes earlier this year, well behind Samsung Electronics and SK Hynix, which have dominated supply since the RTX 50 lineup debuted. The 2GB Micron modules appeared on a limited number of retail cards, including some GeForce RTX 5060 models. A GALAX RTX 5060 board torn down in February was found to carry four D8FHL chips, making Micron the third GDDR7 supplier for the series after Samsung and SK Hynix.

Because Micron’s share of the 2GB GDDR7 market is relatively small — estimated at around 10% — the immediate risk of shortages is limited. Samsung and SK Hynix continue to cover the vast majority of 2GB modules for the standard RTX 50 retail lineup. Still, the exit removes a source of supply at a time when the consumer graphics memory market is already tight, and even a small gap could add short-term pressure on allocation.

The decision fits a broader pattern across the memory industry. Samsung, SK Hynix, and Micron have all been prioritizing AI data-center demand, where high-bandwidth memory (HBM) and higher-density GDDR7 command far better margins than consumer graphics parts. Micron had already signaled this direction earlier when it exited its Crucial consumer memory and SSD business, explicitly citing surging AI data-center demand and a focus on larger strategic customers in faster-growing segments.

Micron’s remaining GDDR7 offering is its 3GB (24Gb) chip, which is now in mass production. The higher-density package allows board partners to increase total memory capacity without adding more chips — four chips yield 12GB, and eight yield 24GB. The 3GB parts are already used in NVIDIA’s RTX PRO Blackwell workstation GPUs and the GeForce RTX 5090 Laptop GPU, both aimed at generative and agentic AI workloads.

However, the 3GB chips are unlikely to find their way into the mainstream RTX 50 desktop lineup. Pricing is the main obstacle. Market reports put the cost of 3GB GDDR7 at roughly $60 to $70 per chip, about triple the $20 price of a standard 2GB module, despite offering only 50% more capacity. The cost premium has already been cited as a factor holding back NVIDIA’s planned RTX 50 Super refresh, and switching the current RTX 50 lineup to 3GB would require board redesigns and memory bus adjustments that make little financial sense for mainstream cards.

For Micron, the calculus is straightforward. The company’s position in the 2GB segment was never competitive, and freeing up fabrication capacity for higher-margin products aligns with its stated strategy. The memory maker’s product roadmap had already indicated that GDDR7 production would shift toward 24Gb parts between 2026 and 2027.

The timing of Micron’s exit could become more significant if NVIDIA’s next-generation gaming architecture faces delays. Fresh market chatter cited by VideoCardz suggests the GR20x family, expected to underpin the Rubin-based RTX 60 series, may have slipped from the second half of 2027 into 2028. If that happens, the RTX 50 lineup would need to remain in production longer, and supply concentration with just two GDDR7 vendors could become a more pressing concern.

NVIDIA has already raised bundled GPU and memory procurement prices multiple times this year, according to industry sources, contributing to higher retail graphics card prices. With memory suppliers favoring AI contracts and consumer GPU supply already strained, the loss of even a minor GDDR7 source adds another variable to an already complicated supply picture.