Daily Alpha · X
· Premarket Alpha · by Buzzberg Research
Jukan relayed Chinese cloud providers seeking long-term memory deals, some through 2035, Zephyr said Foxconn's 2026 Rubin rack cuts were true, and bubbleboi said the S-1 points to a neocloud glut.
Themes on this desk
CoWoS pricing
Pequityresearch relayed reports of 10% to 20% CoWoS price increases, and liwei cited estimates of CoWoS capacity above 130,000 wafers a month in 2026.
Neocloud contract rates
Citizens, quoted by pequityresearch, says the newest neocloud contracts are being signed at higher rates, while Bernstein estimates CoreWeave needs $108 billion of incremental capex to reach 4.2 GW.
VR200 rack outlook
An outlook shared by pequityresearch projects 96,000 NVL72 racks in 2027, with SpaceX the largest VR200 buyer at 20,000 racks.
Ticker heat
?
Top voices by smart followers and alpha score
Market Radar →Chinese cloud providers reportedly seek long-term memory deals, some through 2035
Per Korean media relayed by Jukan, Alibaba, Tencent and ByteDance have requested long-term DDR5 and LPDDR5 supply agreements from Samsung Electronics and SK hynix, with some seeking 10-year contracts through 2035 rather than the typical three to five years.
KOREAN MEDIA: MAJOR CHINESE CLOUD SERVICE PROVIDERS (CSPS), INCLUDING ALIBABA, TENCENT, AND BYTEDANCE, HAVE REQUESTED LONG-TERM SUPPLY AGREEMENTS (LTAS) FOR
Contracts beyond the typical three-to-five-year term would lock in server memory supply for these buyers.
Watch Whether Samsung or SK hynix confirm agreements of this length.
Source →Zephyr says Foxconn's 2026 Rubin rack production cuts were true
Zephyr says not even 500,000 Rubins will be shipped in 2026 and that Elon looks set to bring online nearly as much compute in 2027 as Anthropic and OpenAI combined (10-12 GW).
So, Foxconn's Rubin rack production cuts for 2026 were true Not even 500k Rubins will be shipped in 2026 Looks
The post contrasts thin 2026 Rubin shipments with a large 2027 compute buildout by Elon.
Watch Foxconn's Rubin rack shipments into 2027.
Source →Bubbleboi says the S-1 points to a glut for neoclouds
Bubbleboi argues that if the company is already profitable with accelerating ARR and very long-term compute commitments, it is unclear who the marginal buyer is for all the data centers being built, while not necessarily believing the rosy picture.
The bigger takeaway that I think most people aren’t talking about from the S1 is it deff proves there is
If the S-1 numbers are believed, the post questions how they are bullish for neoclouds.
Watch Neocloud contract signings and pricing after the filing.
Source →TheValueist: Rambus expects Q4 product revenue growth to accelerate to 34%
TheValueist reads Rambus's Q3 call as evidence AI infrastructure spending is broadening into CPU-based servers and conventional memory modules, with product revenue growth expected to accelerate to 34% year over year in Q4 despite minimal MRDIMM contributions.
$RMBS KEY READ-THROUGHS FROM RAMBUS Q3 2026 EARNINGS CALL Rambus’s results strengthen the case that AI infrastructure spending is broadening
The strongest positives fall on conventional DDR5 interface suppliers, server processors and semiconductor IP, while customers are preserving module counts but reducing module capacity.
Watch Whether earlier ordering reflects end demand rather than downstream inventory.
Source →SK hynix's Q3 consensus implies a ~78% operating margin ahead of its October 28 report
mkfilko cites FnGuide consensus of about KRW98.2 trillion of revenue and KRW77 trillion of operating profit, after margins of 58.4% in Q4 last year, 71.5% in Q1 and 76.3% in Q2, and says consensus has already been cut ~5.6% on the stronger won.
Memory (SK hynix reports tomorrow). SK hynix reports Q3 tomorrow (Oct 28, 9am Seoul).
mkfilko argues whether the supply story holds matters more than a headline beat or miss, citing BofA's view that memory supply should be tighter in 2027 and 2028.
Watch SK hynix's operating-margin trend and progress on 2027 HBM volume and pricing talks.
Source →