[Tech Talk] Why Does the 'Bandochae Boritgogae' Occur?

Memory Semiconductors Sensitive to Economic Fluctuations
Like Rice, Wheat, and Copper... Characteristics of 'Raw Materials'

Domestic semiconductor companies are experiencing difficult times amid the prolonged 'semiconductor slump' lasting several months. With export volumes also declining, the shadow of recession has deepened over the domestic economy.

The fundamental cause of the slump lies in the price instability of semiconductors. In particular, prices of memory semiconductors such as DRAM and NAND flash, which are mainly produced domestically, have repeatedly plummeted over the past several months. So why is it that only memory prices are especially unstable?

Quartered in 4 Years... Memory Prices Uniquely Unstable

D-RAM Spot Price Fluctuation

According to market research firm 'DRAMeXchange,' as of January, the market transaction price of DRAM (based on DDR4 8Gb) was recorded at $1.81, dropping below $2 for the first time since 2016. In June 2018, DRAM was traded at $8.19. This means the price has fallen to less than a quarter in about four years.

Samsung Electronics and SK Hynix, which mainly sell DRAM and NAND, have inevitably taken a direct hit from this price drop. In the last quarter, Samsung Electronics' operating profit was only 4.31 trillion won, and Hynix turned to a loss. As a result, companies are forced to cut operating costs as much as possible and reconsider their investment plans.

Like other products, semiconductors are influenced by supply and demand in the market. When an overall IT industry downturn is expected, semiconductor prices also adjust accordingly. However, the price decline of memory is much more severe than that of microprocessors such as Nvidia's graphics processing units (GPUs) or Intel and AMD's central processing units (CPUs). Why does such a difference occur within the same semiconductor product group?

Memory, the 'Rice of Industry,' Has Price Determination Similar to Raw Materials

Rice and DRAM
[Photo by Yonhap News]

The reason memory is sensitive to recessions lies in its characteristics. Memory semiconductors are often referred to as the "rice of industry," meaning they are essential components in modern industry. Moreover, memory resembles actual rice not only metaphorically but also in terms of price determination.

We call grains or metal resources that are widely distributed and consumed in large quantities?such as rice, soybeans, corn, and copper?'commodities.' The biggest feature of commodities is that they have a 'cycle' based on supply and demand. When the economy is good and companies buy large quantities of certain grains and metals, commodity prices rise. Conversely, when a recession hits and inventories accumulate, prices fall. Prices also fluctuate due to production and distribution issues such as droughts or wars.

Similarly, memory prices adjust according to cycles. For example, 2018, when DRAM prices peaked, is remembered as a 'super cycle' when Samsung Electronics and Hynix posted record-high performances.

Memory is a Battle of 'Price Competitiveness'

Samsung Electronics Hwaseong Factory
Photo by Yonhap News

Jim Handy, a leading analyst and author in the U.S. semiconductor industry, explained these characteristics of memory in detail on his blog last year. According to him, memory undergoes cycles approximately every two years.

First, when prices rise, companies increase production, which causes supply to exceed demand about two years later, leading to inventory accumulation. Eventually, a slump arrives, and semiconductor companies refrain from increasing production until demand recovers.

Due to the nature of the industry, it is very difficult for memory to reduce production, making the cycles more distinct. The product gap among top-tier global semiconductor companies is not very large. Price determines competitiveness. In other words, large buyers (mainly IT companies) buy the cheaper DRAM, whether it is from Samsung Electronics or Hynix.

Therefore, memory companies must produce as much as possible to increase yield (the ratio of good products to input quantity) and lower unit costs. Reducing production or cutting investments risks losing market share to competitors and being driven out. For this reason, memory companies always keep production cuts as a 'last resort,' even if it means incurring losses.

This makes memory much closer to a commodity than other semiconductors. Companies making microprocessors such as Nvidia, AMD, and Intel segment their products into 'high-end' and 'entry-level' and compete on performance between products, strengthening brand value and making price defense easier.

On the other hand, memory has a faint brand presence and competes mainly on price competitiveness, making it more affected by cycles. This is similar to how it does not matter where wheat or corn is grown as long as the price is lower than competitors'.

Generative AI Could Become a Savior

How can the semiconductor slump be resolved? According to Handy, memory companies avoid production cuts as much as possible, leading to a 'chicken game' where they endure by cutting into their own flesh until demand catches up with supply again.

In fact, during the first chicken game in 2007, German DRAM company Qimonda, which was ranked second in market share, went bankrupt, resolving the oversupply issue. In 2010, Japanese and Taiwanese DRAM companies went bankrupt, and the 'Big 3 system' of Samsung Electronics, Hynix, and U.S. Micron was solidified, creating the current market structure.

It is unknown how severe the new slump continuing since the second half of last year will be or when it will end. Market research firm Gartner predicted that global semiconductor sales will decrease by 3.6% this year.

However, there is also a possibility of early demand recovery. With the generative artificial intelligence (AI) boom, including OpenAI's ChatGPT, many companies are investing in AI computers. AI programs are mainly processed through microprocessors such as GPUs, but significant capital must also be invested in memory devices to store vast amounts of data.