Micron Technology Inc. (NASDAQ:MU) saw NAND revenue surge 526% year over year to a record $14.1 billion in its fiscal fourth quarter as AI data centers demanded more storage alongside the high-speed memory used by GPUs.

The surge reflects how much more information AI systems now need to store and retrieve, MS Hwang, research director at Counterpoint Research, told Benzinga.

Why AI Needs NAND

As AI systems remember more information, keeping all of it in expensive high-bandwidth memory, or HBM, becomes impractical.

Hwang pointed to the KV cache, which stores information an AI model has already processed and may need again.

"If it cannot all fit into HBM, it needs to be spilled over to DRAM, and from there, down to NAND," Hwang said.

HBM is the ultra-fast memory used alongside AI accelerators such as Nvidia Corp. (NASDAQ:NVDA) GPUs. DRAM provides broader working memory, while NAND is slower and cheaper storage used in solid-state drives.

The industry calls this storage tiering: keeping the most urgent data in fast memory while moving less urgent information into cheaper storage.

AI Workloads Are Getting Much Bigger

Hwang said AI has moved beyond text into images, video and more complex tasks, sharply increasing the amount of data systems need to process and retain.

He compared the shift with moving from simple computer games in the early 1990s to 3D games later in the decade, adding that today’s increase in data is "even greater."

Micron said longer context windows and larger models are increasing memory and storage requirements. Much of that storage sits in SSDs, which use NAND memory.

Micron’s data-center SSD revenue approached $10 billion last quarter, more than 10 times the year-earlier level and more than two-thirds of its total NAND revenue.

Cloud Providers Are Buying More, and Earlier

AI demand is not the only force pushing NAND purchases higher.

Hwang said memory investment remains concentrated in DRAM, leaving less new capacity for NAND and prompting cloud providers to bring SSD orders forward ahead of tighter supply and higher prices.

Micron’s NAND prices rose about 30% sequentially last quarter.

Customers are already trying to reduce memory requirements as costs rise, including by reusing older DRAM, Hwang said.

But he said that does not necessarily mean fewer purchases. Even if customers put less memory in each system, “the total volume of required chips increases even faster,” Hwang said, calling the cutbacks “a completely different story from an actual drop in purchase volumes.”

Hwang also said Micron’s ability to make both DRAM and NAND leaves it more flexible than pure NAND suppliers if DRAM moves to 3D designs, which could expand capacity at lower cost.

The surge in memory demand comes as investors debate how long the broader AI infrastructure boom can last. Polymarket puts the chance of the AI bubble bursting by June 30, 2027 at about 18%, with roughly $3 million traded across the overall market.

Micron shares finished Thursday about 3% higher after erasing early post-earnings losses.

Image: Shutterstock