JAKARTA - The big wave of artificial intelligence investment continues to advance, but there is one component that suddenly becomes a fight like a K-pop concert: memory chip. The tech industry is now facing an acute supply shortage that forces AI giants and electronics manufacturers to compete hard to get components that have rarely been in the spotlight, but are vital to store and stream data.
In Japan, electronics stores are starting to limit the number of hard disks that can be purchased. In China, smartphone manufacturers warn of device price increases. Behind the scenes, giant players like Microsoft, Google, and ByteDance are fighting for supply rations from Micron, Samsung Electronics, and SK Hynix, according to sources familiar with the talks.
Lacks hit almost all types of memory from NAND flash for phones to high-bandwidth memory (HBM), an expensive chip that fuels Nvidia GPU in the AI data center. TrendForce noted that the prices of several memory segments have now jumped more than double compared to February.
SEE ALSO:
The impact could spread far beyond the technology sector. Economists and executives warn that a prolonged crisis has the potential to delay the construction of a hundreds of billions worth of dollars in AI data centers and slow the potential for a surge in AI-based productivity. Inflation can also be pushed up, as many countries are struggling to tame price increases and face new US trade rates.
Sanchit Vir Gogia, CEO of Greyhound Research, describes this condition as a threat that goes beyond technical issues. According to him, AI demand is colliding with supply chains that are not physically ready.
Reuters reports based on interviews with nearly 40 sources, including 17 chip executives, suggest that the acceleration of the industry towards AI chips has trapped memory manufacturers in a double dilemma. They were unable to meet HBM's demand for AI, but excessive focus on premium chips has also made traditional memory supplies for phones, PCs, and home appliances as well. Some companies are now trying to make a U-turn.
In the midst of this tense situation, Meta is reportedly discussing disbursing billions of dollars to buy chips made by Google starting in 2027. Other exclusive information from electronic retailers in China and Japan shows how fast chip prices have risen in recent months.
TrendForce noted that the main memory for computers and cell phones fell sharply from 13'17 weeks at the end of 2024 to just two to four weeks in October.
When some analysts begin to wonder if AI euphoria has caused a bubble, industry executives warn that future data center projects are likely to be delayed. Construction of new facilities takes two years or more, while chipmakers are afraid of overcapacity if the AI trend suddenly slows down.
Samsung and SK Hynix have announced new capacity investments, although they have not detailed how much is allocated for HBM chips or conventional memory.
SK Hynix even warned that this shortfall could last until the end of 2027.
Tensions escalated after OpenAI signed an initial deal with Samsung and SK Hynix for the Stargate project a data center supercomplex estimated to require up to 900.000 wafers per month by 2029, double the current global HBM production.
The push for the AI industry starts from the release of ChatGPT in 2022. Since then, memory factories have shifted most of their capacities to HBM to fulfill orders from Nvidia and large technology companies. However, this shift coincided with the increasing global demand for smartphones, PCs, and traditional data centers.
Samsung even had time to plan to stop production of DDR4 in 2024 before finally withdrawing the decision. Micron also notified customers that they would end deliveries of DDR4 and LPDDR4 in a matter of months.
As a result, the memory price of Samsung's servers jumped by 60 percent last month. Giants like Google, Amazon, Microsoft, and Meta asked Micron to provide chips with no limit on the amount of no matter how much. Alibaba, ByteDance, and Tencent sent special delegates to South Korea to secure supply allocations.
One source describes the situation straightforwardly: Everyone is begging for supplies.
SK Hynix said all the chips for 2026 had been ordered out, while Samsung said all HBM production next year had been sold. A new factory for new conventional memory will be active in 2027"2028.
Micron, Samsung, and SK Hynix share prices soared as memory requests were unstoppable. Consultant Counterpoint Research estimates that the memory price of the new and old generation will rise 30 percent until the fourth quarter and could rise again by 20 percent at the beginning of next year.
This crisis has not shown any signs of easing. The industry is moving fast, but the silicon infrastructure has unforced physical limits. Meanwhile, AI demand continues to accelerate like a super fast train, and the question is now no longer who is leading the AI race but who has managed to get enough memory chips to stay afloat
The English, Chinese, Japanese, Arabic, and French versions are automatically generated by the AI. So there may still be inaccuracies in translating, please always see Indonesian as our main language. (system supported by DigitalSiber.id)