SK Hynix Q3 26 – Spotty Leopard

The memory leopard will not change its spots.

• SK Hynix has reported excellent Q3 26 results and again pronounced the end of memory cyclicality, which is a statement that I think is going to come back and haunt the company.  
• Q3 26 revenues / EPS were KRW104.6tn / KRW86,300 just ahead of expectations of KRW98.2tn / KRW85,420, driven by the ongoing demand for memory and the capacity shortfall that has driven prices up by 25%+ QoQ.
• Demand for DRAM in 2027 is expected to increase by more than the industry’s ability to supply it, with SK Hynix forecasting that it will be able to meet demand growth of 20%+ YoY.
• This does not take into account the effect of price, which is still expected to increase in 2027, although at a slower rate than in 2026.
• This is a factor of demand for memory growing much more quickly than SK Hynix, Samsung, Micron or anyone else’s ability to build the fabs required to make it.
• This will once again be the limiting factor in 2027, which means that the memory industry already has a very good idea of what it will be able to achieve next year.
• Hence, the scene is set for further strong growth in both revenues and profits in 2027, and given Nvidia’s 70% growth forecast, I suspect that in the short term, SK Hynix is lowballing the estimates.
• SK Hynix also stated that it was now sold out for 2027 and that it expects that demand will now outpace its ability to supply it until 2030.
• This is where my opinion and the opinion of the company diverge, as while the memory industry is calling the end of memory cyclicality, I have a suspicion that things may begin to change in 2028.
• The cure for high prices is always high prices, and with 86% gross margins being enjoyed across the memory industry and the suffering of consumer electronics, prices are in nosebleed territory.
• The industry has also used its current market power to migrate its customers to long-term agreements where prices and volumes are fairly fixed, which are hoped to smooth performance across the cycle.
• SK Hynix has secured 50% of its 2027 and 2028 production on these agreements, which I think will hold for as long as the price of DRAM continues to be flat or stable.
• I suspect that the minute that industry moves into a situation of oversupply (which it will), someone will break ranks, offer cheaper memory, which will result in customers breaking their contracts.
• Herein lies the problem for the memory makers, as the market is fairly concentrated, and the last thing one wants to do in a commodity market is to start suing one’s customers for breach of contract.
• Hence, I think that these contracts are going to be largely ineffective at preventing a price crash when the market moves into oversupply.
• However, this remains some time away, as AI models that “think” for longer to improve results and agents all require substantially more memory than the use cases that came before them,
• Hence, memory demand is likely to continue to outpace any other component in the industry, meaning that the 12-month outlook remains excellent.
• I still have a position in Samsung Electronics that I remain comfortable with, although it does regularly test my patience.