Daily Alpha · X
· Premarket Alpha · by Buzzberg Research
Posts flagged Nvidia's lower HBM4 pin speed and NVLink bandwidth and SK hynix's reported $38-45 billion return plan, while Serenity questioned whether neocloud financing has entirely shut off.
Themes on this desk
Neocloud credit
Serenity (@aleabitoreddit) questioned whether neocloud financing has entirely shut off, citing Nebius's $3.5 billion Citibank facility, while Paradis Labs said Musk's and Chamath's comments should be taken seriously and named SPCX, NBIS and
Nvidia specs
Jukan reported Nvidia lowered its HBM4 pin speed, and Zephyr said NVLink bandwidth per GPU fell from 1.8 TB/s to 1.5 TB/s.
Brazil
Brazil's EWZ plunged nearly 20% in overnight trading after Lula won a fourth term.
Ticker heat
?
Top voices by smart followers and alpha score
Market Radar →Lam's CEO says each point of SLC NAND share needs about 2% more wafer capacity
Citing Lam Research CEO Tim Archer's October 21 call, Jukan says moving from TLC to SLC-class NAND cuts bits per cell to roughly one-third, so each 1-point rise in SLC's share of NAND bits needs about 2% more wafer capacity.
Arguably the most valuable insight from the past few days of memory industry news came from Lam Research CEO Tim
Jukan says NAND customers, including Nvidia and the hyperscalers, are developing next-generation products based on SLC.
Watch SLC's share of total NAND bits.
Source →SK hynix is reportedly preparing $38 billion to $45 billion in additional shareholder returns
Korean media report that most of the additional capital would be returned through special dividends, with SK hynix reportedly also considering more share repurchases and cancellations after calls from institutional and foreign investors.
Korean media report that SK hynix is preparing to announce $38 billion–$45 billion in additional shareholder returns.
The company has already carried out substantial share buybacks.
Watch SK hynix's announcement of the return plan.
Source →Zephyr says Nvidia cut NVLink bandwidth per GPU to 1.5 TB/s, leaving AMD ahead
Zephyr says Nvidia reduced NVLink bandwidth per GPU from 1.8 TB/s to 1.5 TB/s and that AMD now has higher HBM bandwidth at 23.3 TB/s and higher scale-up bandwidth at 1.8 TB/s.
Nvidia also reduced the NVLink bandwidth per GPU from 1.8 TB/s (3.6TB/s bidi) to 1.5 TB/s (3 TB/s bidi) AMD
The post follows Jukan's report that Nvidia lowered its HBM4 pin speed.
Source →Serenity says Nebius's $3.5 billion Citi facility undercuts claims neocloud financing has shut
Serenity cited Nebius's $3.5 billion Citibank facility at SOFR plus 2.75%, announced three days earlier, as having better rates than GPU financing from $2T+ companies. Serenity still expects only a few neoclouds to survive in a few years.
Not actively. I've been pointing out issues with Neocloud/Colo financing since 2025 from $SLNH to $CRWV debt interest.
Serenity sees some truth in Musk's comments but says they are framed in extremes.
Watch Neocloud moats against hyperscalers.
Source →Michael Dell expects the memory shortage to get worse in 2027
Dell said a new memory factory takes about four years to build and that Micron factories under construction will start producing in 2029.
CEO of Dell, Michael Dell, sees memory shortage getting worse in 2027 and not easing anytime soon:
Dell, a big Micron customer, says agents have changed how people use the technology.
Watch New memory factory output, which Dell expects from 2029.
Source →