Goldman Sachs found North American investors more optimistic about memory pricing than Asian peers. For 2026 Q3, DRAM and NAND prices are expected to rise ~20% quarter-over-quarter, though expectations have been revised downward. For 2027 HBM pricing, Goldman Sachs expects ~100% year-over-year increase, with conservative estimates at 50% and optimistic views exceeding 100%.
Micron announced a 512GB DDR5 server module enabling up to 12TB memory in two-socket servers, with volume production planned for H2 2027. The module uses 60% less power than four smaller modules, offering a way to increase server memory without exceeding power budgets. This represents a potential long-term growth opportunity for Micron beyond HBM as AI inference demands expand.
A social media discussion questions whether a memory chip downcycle is expected in 2028, sparking engagement on X.
Apple secured Samsung's first-quarter 2027 memory pricing at $2.00 per Gb for DRAM and $0.33 per Gb for NAND, setting the market floor amid tight supply. Memory prices have surged 30–40% from Q3 2026, with 2027 capacity largely allocated to cloud and AI buyers under long-term contracts, leaving phone makers constrained. UBS projects a severe memory supply deficit of –13.6% in 2027, the widest in 30 years, as HBM and server demand outpaces production.
Edge AI applications in autonomous driving and robotics are showing genuine interest in HBM memory as a solution to DRAM bandwidth bottlenecks. SK hynix has already supplied HBM2E to Waymo's autonomous driving platform with automotive qualification, though these segments remain early-stage with limited near-term demand contribution.
A discussion about High Bandwidth Memory (HBM) system architecture was shared on X, gaining engagement with 138 likes and 21 retweets from silicon design community members.
AMD acquired Toronto-based Taalas, which has developed specialized silicon that etches AI models directly into chips rather than loading weights from memory. The HC1 chip runs Meta's Llama 3.1 8B at 17,000 tokens per second with minimal power draw, enabling high-speed AI inference on desktop or mobile devices without reliance on centralized data centers.
Andreas Steno Larsen posted a response to Dario Amodei titled 'We Must Pace the Prophecies' on X, discussing HBM demand in a trending conversation.
A social media post argues that Dario Amodei's essay on AI safety pacing does not reduce semiconductor demand, as it calls for third-party evaluators rather than training halts. The author contends that hyperscalers, Anthropic, OpenAI, and other frontier labs will continue purchasing chips for training and safety work, making the market's negative reaction to AI stocks a misreading of the policy proposal.
A discussion on X critiques HBM capacity as an overblown bottleneck in AI inference, arguing that recent optimizations like V4.1 Flash have reduced memory requirements per token from 3.5KB to 890 bytes through techniques such as CSA2 and SWA Bounded Replay. The author contends that bandwidth, not capacity, is the true limiting factor in decode operations, and that moving from 12-Hi to 8-Hi HBM stacks reflects this reality rather than supply constraints.
Patrick Moorhead shares a detailed AI analysis critical of Dario Amodei's proposal for industry-wide AI pacing, identifying eight structural flaws including unmeasurable commitments, unsupported forecasts, and potential cartel dynamics that could disadvantage open-source development and empower autocracies.
Citi forecasts optical interconnects will reach $92 billion by 2028 with 65% annual growth, as silicon photonics expands from 29% to 60% market share and continuous-wave laser demand grows 114% yearly. Analysts compare lasers to HBM as the next critical bottleneck in AI data center supply chains, with companies like Sive positioned to benefit from supply scarcity.
SK Hynix, Samsung, and Micron are ramping HBM wafer production toward year-end 2026 targets of 200,000, 250,000, and 100,000 units respectively. Samsung's aggressive capacity expansion could shift market share leadership in 2027, though SK Hynix's current yield advantage and NVIDIA relationship remain competitive factors. Separately, optical interconnect demand is projected to reach $92B by 2028 with 65% annual growth, positioning silicon photonics and lasers as emerging bottlenecks.
A user discusses how DeepSeek V4.1-Flash's reduction in KV Cache usage validates their thesis that memory optimization, not raw compute, will be the key competitive advantage for AI agents. They argue that as AI becomes cheaper, demand increases and total inference grows, making efficient memory hierarchies and infrastructure the next frontier for investment.
An investor discusses customized HBM (High Bandwidth Memory) as a growth opportunity, highlighting Samsung's manufacturing advantages over SK Hynix and Micron due to in-house foundry capabilities. The analysis emphasizes HBM's high barriers to entry, customer lock-in potential, and lower price sensitivity compared to DRAM.
Industry analysts discuss strong AI infrastructure demand driving hyperscaler capital expenditure, with particular focus on HBM and high-end memory chips. DRAM supply is expected to remain structurally tight through 2028, while NAND faces a shortage in early 2026 before recovery, with Chinese capacity additions being the key variable determining market balance.
A post discusses Meta's valuation, suggesting the market underestimates the company's value.
Memory chip pricing drove $306 billion of a $453 billion semiconductor sales increase through July 2026, representing 70% of industry growth. DRAM and NAND prices surged 258% and 344% year-over-year respectively, with memory's market share expanding from 29% to over 53%, fueled by AI demand for high-bandwidth memory and long-term supply agreements.
Teradyne CEO Greg Smith highlighted that AI is driving demand for semiconductor testing equipment beyond GPU production, with advanced nodes, chiplets, HBM, and CPO technology requiring significantly more testing per chip. The commentary suggests a structural shift toward higher test equipment intensity across semiconductor manufacturing, with potential TAM growth from $100M to $700M by 2028 in CPO testing alone, benefiting multiple equipment suppliers beyond Teradyne.
Huawei raised its Ascend 950DT AI chip price 60% to $37,300 as demand for Chinese AI compute surges, with DeepSeek planning to deploy 160,000 units. Supply constraints from limited HBM availability are pushing up costs across the industry, with Cambricon also raising prices by up to 30%.