Memory Bottleneck Clouds Hybrid Bonding Outlook, Besi Stock Drops 14% This Week
nashnova research
Dutch semiconductor equipment maker Besi fell 14% this week after UBS downgraded the stock from buy straight to sell — memory supply shortages and cost concerns are delaying hybrid bonding adoption, and Nvidia's next-generation AI chip will actually use *fewer* memory layers, forcing the market to re-examine Besi's core growth thesis.
How bad is the sell-off, and what triggered it?
Besi dropped 14% this week and is now down roughly 45% since late June. It trades at about 29× forward earnings, more than half below its peak valuation four months ago.
UBS cut its rating from buy directly to sell, citing lower earnings expectations. Bank of America had already downgraded to neutral the day before.
This means → two major banks downgraded in the same week — not a sentiment blip, but a systematic re-rating of Besi's core business outlook.
What is hybrid bonding, and why does Besi's fate hinge on it?
Hybrid bonding — a process that joins multiple chips face-to-face or stacks them vertically — boosts computing power in a smaller footprint. It is seen as a key alternative path now that traditional chip-shrinking has hit physical limits.
Besi is the world's leading hybrid bonding equipment maker. Its growth story rested on one assumption: memory architectures would grow ever more complex, forcing chipmakers to stack memory dies into packages thinner than a credit card.
In plain terms = Besi bet on "more and more chip layers" as the industry's direction, with its equipment demand rising in lockstep.
Why does Nvidia's Rubin Ultra undermine that thesis?
According to independent research firm SemiAnalysis, Nvidia's next AI accelerator Rubin Ultra will use fewer memory layers than current products — a concession to tight memory supply, designed to ship more AI chips.
SemiAnalysis analyst Myron Xie noted that hybrid bonding requires expensive pre-processing — ensuring two chip surfaces are nearly perfectly flat — and for high-bandwidth memory, the cost "may not be worth the incremental gain."
This means → if even Nvidia is simplifying its memory stack, the core assumption behind Besi's growth story — that stacking complexity only goes up — is directly challenged.
Are there alternatives that bypass hybrid bonding altogether?
Xie pointed to two paths: spreading workloads across multiple memory units, and next-generation AI models that are themselves designed for lower memory consumption.
AMD was an early adopter, introducing hybrid bonding in its Ryzen 7 processor in 2022 — but uptake among memory chipmakers has lagged noticeably.
This reflects a hard reality: a technology being viable does not mean the supply chain is willing to pay for it.
What is the bull case?
Bernstein analyst David Dai said "the investment thesis still holds" — hybrid bonding remains a necessary technology in the long run. He expects memory supply may catch up with demand by 2028, at which point Besi's equipment will be called upon.
He cited South Korea's SK Hynix: the memory giant has not yet adopted hybrid bonding at scale, but publicly endorsed its value at an industry conference in August that drew close market attention.
In plain terms = the bulls argue the question is not *whether* but *when* — and the answer may be 2028.
Where does the market disagreement really sit?
Barclays analyst Simon Coles (equal-weight rating) noted that the factors governing hybrid bonding adoption are many, and "none of them are within Besi's control."
Over the past three to four years, he said, Besi's investment story has been almost entirely about hybrid bonding — "if adoption momentum is positive, the stock performs; if there's uncertainty, it comes under pressure, as we've seen this summer."
This means → the core question is whether Besi's technology remains irreplaceable if memory architectures trend toward simplification rather than complexity — and the answer may not arrive until supply-demand dynamics clarify around 2028.
市场有风险，内容仅供研究参考，不构成投资建议。