--- title: "SemiAnalysis Responds to Vera SOCAMM Controversy: Critics Did Not Visit SK Hynix Booth at Computex" type: "News" locale: "en" url: "https://longbridge.com/en/news/289015072.md" description: "NVIDIA's Vera Rubin platform reducing its SOCAMM memory to 28TB initially sparked market turmoil, causing Micron Tech shares to plummet 13% in a single day. However, SemiAnalysis directly pointed out that critics \"did not visit the Hynix booth at Computex\"—where SK Hynix had already displayed physical 192GB SOCAMM2 modules. More importantly, the capacity reduction may not signal cooling demand, but rather reflect the Crowding Out Effect of HBM production capacity" datetime: "2026-06-08T07:14:03.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/289015072.md) - [en](https://longbridge.com/en/news/289015072.md) - [zh-HK](https://longbridge.com/zh-HK/news/289015072.md) --- # SemiAnalysis Responds to Vera SOCAMM Controversy: Critics Did Not Visit SK Hynix Booth at Computex Following the intense market reaction to SemiAnalysis' report on SOCAMM memory for the NVIDIA Vera Rubin platform, the latest response has shifted the focus of the controversy from "whether AI memory demand is cooling" back to the product facts themselves: the firm stated that those accusing it of publishing "fake news" clearly did not visit the Hynix booth at Computex. SemiAnalysis posted on social media that its Vera SOCAMM report "caused some volatility," but "some people are drawing incorrect conclusions." The firm emphasized that critics labeling the content as fake news **"clearly did not visit the Hynix booth at Computex."** The report previously stated that **NVIDIA had reduced the modular memory capacity of its next-generation Vera Rubin server racks from 55TB to 28TB, which the market quickly interpreted as a signal of cooling AI memory demand.** However, the key to the controversy lies in the fact that SOCAMM DRAM and HBM are not the same type of product. SemiAnalysis founder Dylan Patel subsequently clarified that his analysis contained "no bearish implications." According to Barron's, the reduction in modular memory capacity for the Vera Rubin platform may instead reflect the ongoing Crowding Out Effect of HBM on traditional memory production capacity. ## Focus of Controversy: Market Interprets SOCAMM Capacity Reduction as Cooling Demand The direct trigger for this round of market volatility was SemiAnalysis' report on changes to the SOCAMM DRAM capacity of NVIDIA Vera Rubin server racks. The report stated that NVIDIA had reduced the modular memory capacity from 55TB to 28TB. The market quickly priced in this change as a signal of cooling demand for AI server memory. Micron Tech shares plunged 13% that day, the semiconductor sector came under pressure, and major stock indices all closed lower. Micron Tech's stock reaction also amplified external attention on the influence of SemiAnalysis' research. **Multiple market participants and online users questioned the quality of the firm's research and its method of information dissemination, arguing that it repeatedly created bearish narratives against Micron Tech in the form of "in-depth research," triggering market panic.** However, SemiAnalysis' response indicated that the firm does not accept the "fake news" allegations. Its latest statement shifted the focus to the on-site display at Computex, implying that the relevant SOCAMM products were not market speculation but had physical exhibits on display. ## SemiAnalysis Strikes Back: If You Didn't Visit the Hynix Booth, You Shouldn't Easily Deny It In its latest post, SemiAnalysis stated: "Our Vera SOCAMM note caused some volatility. As always, some people are drawing incorrect conclusions. Those saying this is fake news clearly did not visit the Hynix booth at Computex." The core of this response shifts the controversy from market inferences about demand to whether the product actually exists. SemiAnalysis did not deny that its report caused market volatility, but explicitly refuted the characterization of its content as "fake news." Semiconductor Insider also posted on social media in support of this point. It stated that SK Hynix had already displayed 192GB SOCAMM2 modules at Computex, designed specifically for the NVIDIA Vera Rubin platform, based on LPDDR5X, emphasizing power efficiency, and ready for mass production. The account also noted that SOCAMM2 is a pluggable module rather than soldered memory, with bandwidth up to twice that of older server DRAM, and better energy efficiency, helping to reduce heat dissipation and cost pressures for AI servers running around the clock. ## SOCAMM2 Positioning: Vera Rubin Uses Both HBM4 and System Memory Another key point in the controversy is the market conflating changes in SOCAMM with overall AI memory demand. According to Semiconductor Insider, the Vera Rubin Superchip uses two types of memory simultaneously: HBM4 and SOCAMM2. The former is high-speed stacked memory, while the latter serves as system memory. SK Hynix displayed the complete memory configuration required for the combination of Rubin GPUs and Vera CPUs at Computex. This means that SOCAMM2 is not a replacement for HBM, but another part of the Vera Rubin platform's memory architecture. Its role is more reflected in terms of capacity, power consumption, and system efficiency. Semiconductor Insider also stated that AI training and inference require large amounts of high-speed memory, and SOCAMM2 helps alleviate power and capacity bottlenecks, noting that Vera Rubin is expected to achieve a 3.5x to 5x performance improvement over Blackwell. The account concluded that SemiAnalysis' statement regarding the booth display was accurate, and AI infrastructure construction is continuing. ## HBM Crowding Out Effect: Capacity Reduction Does Not Necessarily Mean Weakening Demand For investors, the more important question is whether the reduction in SOCAMM capacity for Vera Rubin signifies weakening AI memory demand. SemiAnalysis founder Dylan Patel has clarified that the firm's analysis contains "no bearish implications." This contrasts with the market's initial reaction. According to Barron's, NVIDIA's reduction in modular memory capacity may actually reflect the occurrence of the HBM Crowding Out Effect. The reason is that HBM production consumes significant wafer capacity, with each unit of HBM requiring approximately three times the wafer capacity of standard memory. As HBM supply expands, production capacity for other memory types will be compressed, potentially driving up prices across the entire memory industry. Trendforce analysts also pointed out that as HBM technology generations continue to evolve through 2027, chip sizes are expanding and demand is rising synchronously, so the Crowding Out Effect on traditional DRAM capacity is expected to intensify further. This will provide suppliers with reasons to raise HBM prices and strengthen their pricing power in next year's HBM negotiations. In other words, the adjustment in SOCAMM capacity is not necessarily a signal of weakening AI memory demand, but may also be the result of prioritizing high-value HBM production capacity. ### Related Stocks - [MU.US](https://longbridge.com/en/quote/MU.US.md) - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [NVDY.US](https://longbridge.com/en/quote/NVDY.US.md) - [NVDQ.US](https://longbridge.com/en/quote/NVDQ.US.md) - [NVDD.US](https://longbridge.com/en/quote/NVDD.US.md) - [MUD.US](https://longbridge.com/en/quote/MUD.US.md) - [NVDX.US](https://longbridge.com/en/quote/NVDX.US.md) - [NVDL.US](https://longbridge.com/en/quote/NVDL.US.md) - [MUU.US](https://longbridge.com/en/quote/MUU.US.md) - [NVDU.US](https://longbridge.com/en/quote/NVDU.US.md) - [07709.HK](https://longbridge.com/en/quote/07709.HK.md) - [NVD.US](https://longbridge.com/en/quote/NVD.US.md) - [MULL.US](https://longbridge.com/en/quote/MULL.US.md) - [07788.HK](https://longbridge.com/en/quote/07788.HK.md) - [NVDS.US](https://longbridge.com/en/quote/NVDS.US.md) - [07388.HK](https://longbridge.com/en/quote/07388.HK.md) - [SOXL.US](https://longbridge.com/en/quote/SOXL.US.md) ## Related News & Research - [Should You Buy Micron Stock Before June 24? History Has a Clear Answer.](https://longbridge.com/en/news/288613413.md) - [Micron Stock Could Still Have Nearly 70% Upside Potential Left in Its Tank](https://longbridge.com/en/news/288468369.md) - [Micron's AI Shift Drives Trillion-Dollar Rise](https://longbridge.com/en/news/288467141.md) - [If You Had Bought $5,000 of This Tech Stock 5 Years Ago, Here's What You'd Have Today](https://longbridge.com/en/news/288283585.md) - [SK Hynix plans to double wafer capacity in next five years, group chairman says](https://longbridge.com/en/news/288379830.md)