Just before last weekend kicked off, Nvidia (NASDAQ: NVDA) and SK Hynix (NASDAQ: SKHY) finished some business, signing the largest memory deal in history. As part of the partnership, SK Hynix's subsidiary, SK Telecom, will build a 2-gigawatt AI cloud data center in Korea using Nvidia's Vera Rubin Platform. Meanwhile, SK Hynix will supply Nvidia with high bandwidth memory (HBM) going forward, while the two companies will work together to co-develop future generations of AI memory.
One of the biggest bottlenecks in the AI infrastructure build-out right now is memory, especially HBM. To reduce latency and improve power efficiency, graphics processing units (GPUs) and other AI chips need to be packaged with HBM. Meanwhile, the need for HBM is only growing as inference increasingly focuses on fast memory access rather than raw compute power.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
However, increasing HBM capacity is a challenge. It requires access to the same EUV (extreme ultraviolet lithography) machines that manufacture GPUs and other advanced logic chips, and the supply for these machines is limited, since the only company that makes them is ASML. At the same time, HBM needs upward of 3 times the wafer capacity of regular DRAM, adding another challenge if one wants to rapidly increase supply. HBM also must be manufactured in massive clean rooms, which can take years to build.
My prediction is that this deal will play an important role in both stocks being big winners over the next few years.
#supply
One of the biggest bottlenecks in the AI infrastructure build-out right now is memory, especially HBM. To reduce latency and improve power efficiency, graphics processing units (GPUs) and other AI chips need to be packaged with HBM. Meanwhile, the need for HBM is only growing as inference increasingly focuses on fast memory access rather than raw compute power.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
However, increasing HBM capacity is a challenge. It requires access to the same EUV (extreme ultraviolet lithography) machines that manufacture GPUs and other advanced logic chips, and the supply for these machines is limited, since the only company that makes them is ASML. At the same time, HBM needs upward of 3 times the wafer capacity of regular DRAM, adding another challenge if one wants to rapidly increase supply. HBM also must be manufactured in massive clean rooms, which can take years to build.
My prediction is that this deal will play an important role in both stocks being big winners over the next few years.
#supply
2 months ago
