AI frameworks prioritize FLOPs and TOPS. But @SamsungSemiUS President Paul Cho says the bottleneck has shifted. Memory bandwidth now determines AI inference performance and cost. He believes memory-centric compute could define AI’s next decade. Watch:

Heat is a known issue in HBM, which makes this slide interesting from @SamsungSemiUS as a potential way to handle the increased thermal challenges. I'm also intrigued on the notion that each cHBM base die will not be created equal and performance will likely vary.

AI inference storage requires a tradeoff between performance and cost. Multi-tier storage helps optimize AI inference performance and cost. Check out this recap of Session 1 of Supermicro Open Storage Summit 2026, where industry leaders from @Supermicro, @intel, @scality, @weka, @wd_corporation, and @SamsungSemiUS dive into this topic. #StorageSummit2026# #OpenStorageSummit# #SupermicroStorage2026# #theCUBEresearch# #SiliconANGLE#

Samsung Speeds Up First P5 Mass Production Line, Pulls Equipment Installation Forward to Q2 Next Year Samsung Electronics is moving faster to build the first mass production line (Ph1) at Pyeongtaek Campus 5 (P5). It was learned on the 28th that Samsung is discussing with major equipment suppliers a plan to pull the target for moving equipment into Ph1 forward from the third quarter of next year to the second quarter. The move is seen as reflecting Samsung's intent to secure capacity quickly as demand for high value DRAM and NAND for AI surges. P5 is Samsung's next generation chip production base, targeted to begin operations in 2028. Construction of the Ph1 cleanroom began in the third quarter of this year. A cleanroom is the infrastructure that controls fine particle contamination, temperature, and humidity inside a fab. Samsung originally planned to build the P5 Ph1 cleanroom early next year but pulled that schedule forward by about six months. Equipment installation for Ph1 was then set for around the third quarter of next year. Recently, however, Samsung's P5 capex has picked up even more. "Samsung has told us it will start moving equipment into P5 Ph1 around May or June next year instead of July or August as originally planned," one industry source said. "The schedule already changed once when the cleanroom was built early, and now investment is moving even faster." Another source said, "Samsung is going so far as to ask for equipment to be delivered even in the first quarter of next year, saying it will store the tools somewhere else. Its commitment to investing early in P5 is very strong." Discussions on P5 Ph2 investment are also taking shape. Ph1 is currently likely to be built as a DRAM and high bandwidth memory (HBM) line, and Ph2 as a leading edge NAND line, including 10th generation (V10) NAND. "No formal purchase order (PO) has been issued yet, but Samsung is discussing plans with suppliers to build P5 Ph2 as a NAND line," said an equipment industry source. "Equipment lead times, the time from order to delivery, are very long right now, so Samsung is asking suppliers to secure related components in advance." The main reason Samsung is moving up its capex timeline is the memory chip shortage. Global Big Tech companies are increasing orders for the high performance DRAM and NAND needed to build AI infrastructure, but memory makers including Samsung cannot meet all of that demand because of capacity limits. On its second quarter earnings call in July, Samsung said, "Based on the demand we have received, demand that goes unmet this year is being pushed into next year, adding further supply pressure." It added, "We expect next year's supply shortage to be more severe than this year's, and the shortage to continue into 2028."

Samsung just became the first to validate LPDDR6 on a Snapdragon. The 16GB chip hit 10.7Gbps on Qualcomm's Snapdragon 8 Elite Extreme Gen 6, with power efficiency up 21% over LPDDR5X. Samsung says sample supply and quality validation are done and it's ready to ship on customer schedules. Meanwhile China's CXMT announced LPDDR6 in mass production earlier this month at 12.8Gbps, faster than the 10.7Gbps baseline, and Xiaomi's 18 Fold with its own XRING O3 will be the first platform to use it. On the Galaxy side, Samsung may split memory tiers: testing Micron's 1 gamma LPDDR5X for the S27 and S27 Plus while keeping LPDDR6 for the Ultra and Pro