SK Hynix Is Doubling Its Chip Capacity. The Memory Shortage Driving That Decision Will Last Until 2030.

Most supply chain teams have spent the past year tracking memory chip availability as a quarterly procurement problem. This week, the chairman of SK Group put a timeline on it that should change how procurement teams think about it entirely.

SK Hynix plans to double its memory chip capacity over the coming half-decade, responding to an endemic deficit of storage chips that could last until 2030. SK Group chairman Chey Tae-won confirmed the expansion at Computex in Taipei, telling reporters that "until 2030 there's still some shortage" and that the company would do whatever it takes to fund the wafer capacity buildout.

The scale of demand driving that decision is hard to overstate. Memory chips have become one of the biggest bottlenecks to AI development. Arm Holdings CEO Rene Haas said that memory remained the biggest bottleneck for the global AI industry, in part because the sector leaders had pulled back on expansion during the post-COVID downturn. Some industry executives have declared a super-cycle of exponential demand for memory, ranging from the high-bandwidth memory required to train AI models to conventional DRAM and NAND flash employed across data centres.

Why this is not just a good news story for chip buyers

A commitment to double capacity sounds like a supply chain fix. The timeline behind it tells a different story. Chey said that the lead time for setting up a new site was long, with a greenfield project potentially requiring more than five years. SK Hynix's capital expenditure this year will rise significantly from last year's 30.2 trillion won, but the company has not elaborated on the specific amount, noting that it is hard to quantify because of volatile prices for a range of resources including land, equipment and electricity. 

The structural constraint is upstream of the fabs themselves. Advanced memory production requires extreme ultraviolet lithography equipment, ultra-pure process chemicals, and specialist gases including helium, the same helium whose supply was severely disrupted by the Strait of Hormuz closure and the Qatar LNG outage covered in earlier editions of this newsletter. SK Hynix, which together with Samsung Electronics and Micron Technology dominates the global market for memory chips, is one of the biggest beneficiaries of a global data centre buildout, with trillions of dollars in projected spending by hyperscalers like Meta having sent SK Hynix's and Micron's market valuations past the US$1 trillion mark for the first time last week. 

The valuation milestone matters here. Companies being asked to fund five-year capacity doubling programmes while managing volatile input costs, a constrained helium supply, and a geopolitical environment that includes an active labour dispute at Samsung, their nearest competitor, are not operating in a low-risk capital allocation environment. Every dollar committed to a greenfield fab is a dollar committed to a five-year payback cycle on a supply chain that has already demonstrated it can be disrupted from multiple directions simultaneously.

The demand side is not slowing down

The memory industry remains wary of over-expanding and precipitating another market collapse, having broken a decades-long cycle of boom-and-bust. That caution, combined with the structural AI demand surge, is helping sustain the shortage dynamic that Chey described. The companies that would normally race to fill a capacity gap are moving deliberately because they have been burned by overcapacity before. The result is a shortage that self-perpetuates even as investment announcements accumulate.

For any company building AI infrastructure, running data centres, manufacturing electronics, or supplying components to the automotive sector, the message from Computex this week is consistent with what Samsung's labour situation, the helium supply crunch, and the Airbus engine story all point to: the supply chains that the digital economy runs on are being stretched from multiple directions at once, and the fixes being announced are measured in years.

The exposure for European and Asian companies

For procurement teams sourcing DRAM, NAND, or high-bandwidth memory, the SK Hynix announcement confirms what lead time data has been signalling for months. The shortage is structural, not cyclical, and it will not resolve before 2030 regardless of how many investment announcements are made between now and then.

The disruption does not arrive as a production stoppage or a force majeure notice. It arrives as an allocation email, a lead time extension, and a five-year capacity buildout that cannot deliver product until the greenfield site is operational.