SK Hynix has become Nvidia's primary supplier of high-bandwidth memory (HBM), the component that gets packaged alongside Nvidia's GPUs to keep them from sitting idle waiting for data. The two companies have formed a supply and co-development partnership under which SK Hynix will supply Nvidia with an estimated $500 billion worth of memory over the next several years.
Nvidia's edge in the AI chip race increasingly runs through one of the biggest bottlenecks in AI infrastructure: high-bandwidth memory. SK Hynix will supply Nvidia with an estimated $500 billion worth of memory over the next several years under a newly formed supply and co-development partnership, locking in access to a scarce component in AI infrastructure.
HBM gets packaged with GPUs like Nvidia's to reduce latency and optimize performance. Without it, GPUs would spend much of their time idle waiting for data, and power consumption would skyrocket. That makes HBM a necessary component stacked alongside the GPU die within the same package, not an optional add-on.
Why HBM supply stays tight
Only three companies make HBM today: SK Hynix, Samsung, and Micron. Each needs clean-room space that takes years to build, and HBM itself consumes upwards of three times the wafer capacity of conventional DRAM, which tightens the squeeze further. On top of that, all three memory makers compete with chipmakers for extreme ultraviolet lithography machines, which only ASML produces, and ASML can only expand its own capacity by so much.
Most investors have played the HBM bottleneck through Micron, but the source positions this as the wrong long-term move. Micron lags SK Hynix and Samsung in HBM even as it has benefited from rising conventional DRAM and NAND prices during the broader memory supercycle.
SK Hynix leads where it counts
As of the second quarter, SK Hynix held a 25% share of the DRAM market but a 50% share of HBM. Micron, by comparison, holds 24% of DRAM but only 18% of HBM, while Samsung sits at 38% of DRAM and 33% of HBM. SK Hynix draws the largest share of its revenue from HBM among the three memory makers as a result.
What the Nvidia deal means for both sides
The anchor contract gives SK Hynix visibility to expand capacity in a market that has historically swung through sharp cycles, now backed by demand it can count on from the world's largest AI infrastructure company. SK Hynix trades at a forward P/E of 5.5 times 2027 analyst estimates, which the source frames as room to run if the business proves less cyclical than in the past.
For Nvidia, securing scarce HBM supply reinforces one of its sneakier advantages in a compute-starved market where AI chips need HBM packaged alongside them to function. Nvidia trades at a forward P/E below 15 times fiscal 2028 estimates, a level the source's analyst describes as reason the semiconductor stock is a buy.
Source: The Motley Fool
Trading involves risk.
Risk Disclaimer
