
SK Hynix and Nvidia announced a more than $500 billion(約80兆円) partnership for memory supply and an AI data center set to launch in 2027, while Samsung and Broadcom committed to $200 billion(約32兆円) in memory and foundry services through 2030 at a South Korean government-hosted AI Summit in San Francisco. Bernstein analysts view these deals as evidence of memory's critical role in AI infrastructure and recommend buying memory stocks on recent weakness, rating Samsung, SK Hynix, and Micron as Outperform.
Summaries like this, in your inbox every morning.
Sign up free →What happened
SK Hynix and Nvidia announced a partnership worth more than $500 billion(約80兆円) covering memory supply and a 2GW Vera Rubin DSX AI Factory planned to come online for SK Telecom in 2027. Samsung and Broadcom signed a memorandum of understanding covering $200 billion(約32兆円) of memory and foundry services through 2030, including high-bandwidth memory (HBM) and 2-nanometer and below process technology. The agreements were unveiled Friday at an AI Summit hosted by the South Korean government in San Francisco.
Why it matters
Bernstein analysts say these announcements reinforce memory's growing importance to AI infrastructure and indicate that Nvidia and Broadcom need to secure stable memory supply. The deals suggest memory chips are becoming as critical as logic semiconductors for building AI systems. Bernstein views the memory sector's recent pullback as 'a good entry point' and remains constructive on memory makers.
What to watch
SK Group is pursuing an additional $250 billion(約40兆円) in memory supply collaborations with other global tech companies over the next five years. Bernstein rates Samsung, SK Hynix, and Micron Outperform, though the firm flagged a longer-term competitive threat from China in NAND and gave Kioxia an Underperform rating.
At an AI Summit hosted by the South Korean government in San Francisco on Friday, four major semiconductor companies unveiled partnerships totaling hundreds of billions of dollars in committed memory supply and manufacturing capacity. SK Hynix and Nvidia led the announcements with a partnership worth more than $500 billion(約80兆円). The deal covers memory supply to Nvidia and includes a 2GW Vera Rubin DSX AI Factory planned to come online for SK Telecom in 2027. Bernstein analysts, led by Mark Li, noted that the memory partnership includes "long-term technical developments & stable supply of next-gen AI memory" designed to help SK Hynix "expand the foundation for growth."
Separately, Samsung and Broadcom signed a memorandum of understanding worth $200 billion(約32兆円) covering memory (including high-bandwidth memory) and foundry services through 2030. The foundry component will focus on 2-nanometer and below process technology alongside advanced packaging techniques that Bernstein compared to TSMC's Chip-on-Wafer-on-Substrate (CoWoS) offerings. SK Group is also pursuing an additional $250 billion(約40兆円) in memory supply collaborations with other global tech companies over the next five years, expanding the total commitment beyond these two flagship partnerships.
Bernstein's bullish interpretation centers on memory's role in AI infrastructure. The analysts wrote that "the announced amounts are more for memory and indicate the need for NVIDIA & Broadcom to secure memory supply," suggesting that both companies view long-term supply agreements as essential to their AI growth plans. For context, the analysts noted that consensus expectations stand at roughly $1.3 trillion(約210兆円) in annual memory revenue in both 2027 and 2028, though they acknowledged they "can't precisely quantify the impact of the announcements." The partnerships also help ease supply concerns for Nvidia and Broadcom, both of which have guided toward substantial AI-related growth.
Bernstein rates Samsung, SK Hynix, and Micron as Outperform and views the memory sector's recent pullback as "a good entry point." The firm argues that memory is more important to AI than logic semiconductors and remains constructive on the sector overall. However, Bernstein flagged a longer-term competitive threat from China in NAND, resulting in an Underperform rating on Kioxia; the firm notes that this risk is far smaller in DRAM, where China will likely struggle to compete without access to EUV lithography. Regarding TSMC's position, Bernstein said any impact "should be negligible," since it remains unclear whether Broadcom will actually produce AI ASICs at Samsung and given the existing queue of demand for TSMC's capacity.
The announcements at Friday's AI Summit represent a significant pivot in how AI infrastructure builders secure critical components. Historically, memory supply has been treated as a commodity, but the scale of these partnerships—more than $500 billion(約80兆円) from Nvidia and SK Hynix alone—signals that memory is now viewed as strategically as foundry capacity or logic chips. Bernstein's note that "the announced amounts are more for memory" suggests both Nvidia and Broadcom recognize they cannot rely on open-market purchases to meet their AI growth targets and instead need long-term contractual guarantees.
The timing reflects the sector's recent volatility. Memory stocks have pulled back in recent months, but Bernstein interprets this as a buying opportunity rather than a structural concern. The firm's constructive stance rests on the belief that consensus expectations of roughly $1.3 trillion(約210兆円) in annual memory revenue in both 2027 and 2028 will materialize as these partnerships take effect. Bernstein also downplays risk to TSMC, arguing that Broadcom's willingness to use Samsung's foundry for AI ASICs remains unproven and that existing demand for TSMC capacity is too large to be materially displaced.
The one shadow on the outlook is China. Bernstein flags competitive risk in NAND from Chinese suppliers but sees DRAM as better protected, since China lacks access to the extreme ultraviolet (EUV) lithography tools needed to compete in that segment without export-control relief. This asymmetry explains why the firm rates Kioxia (exposed to NAND competition) as Underperform while maintaining Outperform ratings on Samsung, SK Hynix, and Micron, all of which have substantial DRAM exposure.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
No comments yet. Be the first to share your thoughts!
Log in to join the discussion





Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime