Bernstein: Memory Pullback Is a Buying Opportunity, Maintains Outperform on Samsung/SK Hynix
N.R. Finch
SK Hynix and Nvidia signed a $500 billion+ cooperation LOI while Samsung and Broadcom inked a $200 billion strategic MOU, locking memory capacity through 2030. Bernstein reiterated Outperform on Samsung, SK Hynix, and Micron, calling the recent pullback a "good entry point."
What do these two mega-deals actually cover?
The SK Hynix–Nvidia LOI spans long-term technology development and stable supply of next-gen AI memory. SK Group plans to seek another $250 billion in global memory partnerships over the next five years.
The Samsung–Broadcom MOU covers memory (including HBM) and foundry. The foundry portion targets 2-nanometer and beyond, using advanced packaging similar to TSMC's CoWoS — a technique that bundles multiple chips into one package. The deal runs through 2030.
This means → two AI-compute giants are pre-booking future memory capacity with ultra-long-term contracts — essentially hedging supply-chain risk.
Why does Bernstein call this pullback a buying opportunity?
Analyst Mark Li's team kept Samsung, SK Hynix, and Micron at Outperform, explicitly flagging the recent sector pullback as an entry point.
Core logic: the disclosed deal values point mainly to memory, reflecting Nvidia's and Broadcom's urgency to "secure memory supply." This means → memory chips matter more to AI than logic semiconductors, in Bernstein's view.
Reference point: consensus forecasts peg 2027 and 2028 global memory revenue at roughly $1.3 trillion each year. In plain terms = these two deals sit in the same order of magnitude as the entire industry's annual revenue — a measure of memory's weight in AI infrastructure.
What about TSMC and Kioxia?
TSMC: Bernstein sees the impact as "negligible." Whether Broadcom will actually produce AI ASICs — chips custom-designed for specific tasks — at Samsung remains uncertain, and TSMC's own capacity queue is still full.
NAND: Kioxia receives an Underperform rating, driven by long-term competitive pressure from China.
This reflects a layered call: Bernstein is bullish on DRAM/HBM leaders whose capacity is locked in, but cautious on NAND where the competitive landscape is deteriorating.
What should investors watch next?
These deals are letters of intent and memoranda, not signed purchase orders — Bernstein itself admits it "cannot precisely quantify the impact."
Whether the step-change in revenue visibility for memory giants shows up in upcoming earnings will be the market's key test of these agreements' real substance.
In plain terms = the deals sketch a very large promise; now the market needs to see it baked into actual numbers.
Content is for reference only, not financial advice.