Clients offer direct funding.
This is not your father’s supply deal. Forget the usual song and dance of long-term contracts. Major tech players, drowning in a sea of AI-fueled demand and facing a memory capacity that’s basically a black hole right now, are now floating proposals that would make a venture capitalist blush. We’re talking about ponying up cash for entire production lines, not just signing on the dotted line for future shipments. And the pricetag? We’re not just talking about spare change; some are offering to bankroll the acquisition of ASML’s EUV lithography machines, each costing hundreds of millions of dollars. This is a level of desperation I haven’t seen since the early days of the smartphone boom, and frankly, it smells like panic.
Why This Memory Shortage is Different
Here’s the kicker: this direct funding model is old hat in the logic chip world. Companies building CPUs or specialized AI accelerators have long done this. But for DRAM and NAND? Never. Those chips? They were always made speculatively, a gamble on the open market. Now? The market’s so tight, there’s not a single wafer to spare for any one customer.
“Regardless of the type of offer, available capacity is essentially zero right now,” said another source to Reuters. “There isn’t even a small portion that can be designated for a specific customer.”
This shift is seismic. Just months ago, SK hynix and Samsung were playing hardball, shortening contract terms and scrambling to capture soaring spot prices. Now, customers are saying, “Fine, if you won’t give us supply, we’ll build it ourselves.” It’s a bold move, a desperate gambit to bypass the existing power structure.
The Price of AI’s Hunger
SK hynix isn’t just considering these offers; they’re actively reviewing them. They’ve even hinted at “structural alternatives that differ from conventional long-term agreements.” This isn’t just about securing a few extra gigabytes for a chatbot. This is about fueling the insatiable engine of artificial intelligence. The company’s stock price has apparently taken notice, soaring 154% this year. Makes sense. When everyone’s starving, the baker who can actually produce bread suddenly looks like a king.
Samsung and Micron are also in talks, but SK’s situation seems particularly acute. The offers go beyond mere multi-year deals. We’re hearing whispers of price-band agreements, setting floors and ceilings for annual pricing. Imagine pre-paying 30-40% of your contract value upfront. It’s a stark departure from the quarterly haggling that defined memory pricing for decades.
But SK hynix is playing it cool—or trying to. They’re concerned about becoming beholden to a single wealthy buyer. And honestly, who can blame them? Picking a favorite in this AI gold rush could backfire spectacularly if the winning technology shifts. Chipmakers don’t want to be caught holding the bag for the next big thing that fizzled.
A Long Road Ahead
The big question is, can SK hynix — or anyone else — actually scale up fast enough? SK Group’s chairman is predicting shortages through 2030. Samsung and SK hynix themselves are warning investors that supply constraints will persist at least until 2027. That’s a lot of years to be running on fumes.
This isn’t just a supply chain hiccup; it’s a fundamental shift. The AI arms race has officially emptied the chip cupboard. And now, the wealthiest clients are opening their wallets, hoping to literally buy their way to the front of the queue. It’s a fascinating, if terrifying, spectacle.
Can Customers Really Fund a Fab?
The proposals involve significant capital, potentially hundreds of millions of dollars for EUV machines or dedicated production lines. While unprecedented in memory, it reflects the extreme pressure customers face.
When Will the Shortage End?
Industry leaders like SK Group’s chairman are forecasting shortages until 2030. SK hynix and Samsung have also warned of constraints lasting through at least 2027.
Why Are Memory Chips So Scarce?
The surge in AI development has dramatically increased demand for high-bandwidth memory (HBM) and other memory components, outpacing existing production capacity.
