About half of the HBM memory chip market is now controlled by South Korean SK hynix. At the end of February this year, Micron Technology announced that from the second quarter it will begin supplying 8-tier HBM3E stacks for the needs of Nvidia, which will install them in H200 accelerators. This week it became known that Micron has begun supplying similar memory to a mysterious second major client.
Image source: Micron Technology
Under this vague definition, AMD may be hiding, which also equips its Instinct family of computing accelerators with HBM3E-type memory, but one cannot speak of an unambiguous correspondence. Moreover, at its quarterly earnings conference call this week, Micron Technology executives announced that the company will begin supplying its HBM memory chips to a third major customer in the first quarter of next year.
In September of this year, Micron introduced 12-tier HBM3E stacks, thereby demonstrating the elimination of the backlog from SK hynix. The third player in this market, the South Korean company Samsung Electronics, has been trying for several months in a row to certify its HBM3E memory to meet Nvidia’s needs, but fails over and over again, despite regular assurances that success is close. At the quarterly conference, Micron management emphasized that customers were very satisfied with its 12-tier HBM3E memory.
Micron estimates the market capacity for HBM-type chips by 2025 at more than $30 billion against the previously mentioned $25 billion. By 2028, the market capacity will grow to $64 billion, and by the end of 2030 it will exceed $100 billion. This year this value will not exceed $16 billion, according to Micron. What is typical is that all orders for HBM production for the next year have already been distributed to the company, and prices are fixed in contracts.