Nvidia is not the only beneficiary, and AI training is also beneficial for storage chip manufacturers

On May 30th, it was reported that although the storage chip market is sluggish, there is a huge demand for artificial intelligence, which will benefit companies such as Samsung and SK Hynix.On May 24th, NVIDIA released its financial report, and its market value surged by $207 billion within two days

On May 30th, it was reported that although the storage chip market is sluggish, there is a huge demand for artificial intelligence, which will benefit companies such as Samsung and SK Hynix.

On May 24th, NVIDIA released its financial report, and its market value surged by $207 billion within two days. Previously, the semiconductor industry had been in a sluggish state, and this financial forecast gave people great confidence and hope.

If the field of artificial intelligence ushers in a moment of takeoff, traditional technology giants like Microsoft and startups like OpenAI will seek help from companies such as Samsung and SK Hynix.

Machine learning requires memory chips to process large amounts of data, analyze videos, audio, and text, and simulate human creativity. In fact, artificial intelligence companies may purchase more DRAM chips than before.

The reason for the high demand for memory chips is simple: NVIDIA's AI chips are different from standard CPUs, as they can read and process a large amount of data at once, and then immediately output the results. But to leverage this powerful advantage, computers need to receive data quickly and without delay. This is the function of storage chips.

The processor cannot directly read data from the hard drive - it is too slow and inefficient. The first choice is to save the data in the internal cache of the chip, but the space here is limited, and chip manufacturers are more willing to use valuable internal space to improve computing power. Therefore, the second best approach is to use DRAM.

When training complex chat robots, it may be necessary to process billions of pieces of information at once, requiring quick access to these data. If there is insufficient DRAM in the system, the speed of the computer will significantly slow down, which means that the best processor purchased for $10000 will not be able to realize its value. This means that for each high-end AI processor, it may be necessary to install up to 1TB of DRAM memory, which is 30 times more than for high-end laptops. Research firm TrendForce stated that this means that at some point this year, the sales of DRAM chips used for servers will exceed those used for smartphones.

Artificial intelligence systems also need to be able to quickly and massively save output data for fast reading and writing, therefore requiring the use of NAND chips, which are also used in smartphones and most laptops. Samsung is a global leader in this field, followed by the memory factory Armor, which was spun off from Toshiba in Japan, and SK Hynix in South Korea.

Last quarter, DRAM and NAND storage chips contributed $8.9 billion in revenue to Samsung, far exceeding Nvidia's $4.3 billion in data center business (which includes products for artificial intelligence). However, it should be noted that this is the worst quarter for Samsung's memory department in seven years, and memory sales related to artificial intelligence only account for a small portion of total revenue.

In the future, for every high-end artificial intelligence chip sold, more than a dozen DRAM chips will be shipped, which means that companies such as Samsung and SK Hynix will increase their revenue. These companies collectively control 95% of the DRAM market share. As NVIDIA grows and expands, they will also expand accordingly.

Undoubtedly, the artificial intelligence revolution has arrived, with cool chat robots, ubiquitous search engines, and high-performance processor manufacturers being the biggest winners. And companies that produce storage chips in large quantities will not be excluded. (Chen Chen)