Dylan Patel reports NVIDIA downgraded Rubin Ultra GPU memory specs from 16-high to 12-high HBM stacks

LMAO Rubin Ultra's HBM literally got downgraded to 12 Hi, and they are not even using HB yet They won't use 16 Hi in 2027 Hybrid Bonding is very expensive for HBM rn (yields are abysmal) The big near-term upside for BESI is that TSMC's COUPE & Tower use Hybrid Bonding for EIC + PIC integration for CPO

Clients of SemiAnalysis Memory and Accelerator model knew about Rubin Ultra 16 Hi to 12 Hi cut in March :)

China will be huge for BESI as Huawei is adopting HB for Logic Folding U can even see it in their China exposure

Don't expect Hybrid Bonding in HBM before 2028

@bubbleboi no, they are not using ASMPT hybrid bonders for logic folding SJ Semi is the main partner they use Piotech & Besi

@zephyr_z9 how HBM works

@zephyr_z9 It’s more bullish Huawei or ASMPT

@zephyr_z9 Ok

@paurooteri 銘柄の情報は📝

@dylan522p I have no idea what this is but my guess is it's insider info inferred from semianalysis data

@zephyr_z9 My strategy plan. ....

🔻↩️↩️

@paurooteri

@volaryu @dylan522p Models they only offer to institutional subscribers. Kinda a bummer to not offer anything for not institutional subscribers. I’d pay whatever.

@paurooteri 🔻以下、詳細です

@zephyr_z9 My strategy plan. ....

🔻↩️↩️

DRAM is broadly too expensive now, making the yield-hits from 16-Hi unjustifiable for the cost. Since HBM core dies are so precious, the base case is now squeezing more juice out of single dies and stretching out existing tech.

@zephyr_z9 My strategy plan. ....

🔻↩️↩️

@zephyr_z9 @grok what are the implications for memory and nand stocks, cpo, optics ?

@dylan522p Details of my stock holdings are as follows

⬇️

@zephyr_z9 Do you think Dongshan/Source Photonics ramp projecting 22 million 100G/200G EML is a possible oversupply risk for current photonics players?