The trend of investing heavily in AI infrastructure has caused high-end HDDs to sell out, forcing many technology companies to switch to SSDs.
The race towards AGI (artificial general intelligence) is pushing stakeholders to invest and build data centers at a rate far beyond current production capacity. Manufacturers are struggling to meet the needs of the AI industry, leading to a prolonged shortage of DRAM, pushing the price of this memory to more than double in the past few months. According to DigiTimes, not only DRAM, the storage sector is also being severely affected as the delivery time of HDDs for businesses is delayed by up to two years. This means that if a company wants to buy a large capacity hard drive, they have to wait in line for 24 months.
Unable to wait for the AI fever to cool down, hyperscaler computing providers are turning to SSDs with QLC NAND memory chips to avoid backlogs.
However, the continued hoarding of QLC NAND has caused a new shortage, as most cloud service providers in the US and China are racing to buy. This could cause global SSD prices to surge, experts say. In fact, DigiTimes reports that some NAND makers’ QLC production capacity is fully booked through 2026.
Photo: Western Digital.
There are two NAND memory chip technologies equipped on SSDs: QLC and TLC. Of which, TLC stores 3 bits of data, while QLC stores 4 bits of data on each memory cell. QLC stores more data on the same area, thereby reducing production costs, but has slower write speeds and lower durability than TLC.
In the context of HDD shortages and AI data centers needing to expand rapidly, many large enterprises have to choose SSDs with QLC NAND, which is less durable but enough to meet the needs of “cold” storage, meaning data that is rarely accessed, mainly used for long-term storage. QLC NAND is predicted to surpass TLC in popularity in early 2027.
According to Tom’s Hardware, the above scarcity is considered unprecedented in the field of memory and storage. Currently, DRAM and NAND manufacturers are prioritizing sales to AI customers who are willing to pay high prices. They used to have two to three months of spare capacity, but now it’s down to just a few weeks. This has led to the best results in years for memory chipmakers, but ordinary consumers are caught in a new shortage of electronic components.
Admin (by Tom’s Hardware, DigiTimes)
