AI Memory

Volantis

2026-10-03 12:09:32

Volantis says it raised $88 million in Series A to pursue optical interconnects for AI memory limits

Volantis, an optical interconnect chip company, said on X that it has completed an $88 million Series A round. The company said it is using an optical approach to address AI memory bottlenecks by equipping chips with large amounts of high-speed memory and lifting per-chip memory bandwidth and capacity by several orders of magnitude. Volantis said the design is aimed at high-speed inference for models with more than 10 trillion parameters, with throughput of up to 10,000 tokens per second for a single user. In its initial stage, the company said the technology will be used to accelerate agents, cutting programming-related tasks from hours to minutes or seconds. Volantis also said optical systems can improve memory performance by several orders of magnitude across generations instead of roughly doubling each year, and added that its next-generation product has taped out.

60

Applied Mater

2026-10-01 04:13:13

Applied Materials adds KIOXIA to EPIC Center for AI memory work as AMAT climbs 90.2% in 2026

Applied Materials said KIOXIA, the Japanese flash memory and solid-state drive maker, will join its EPIC Center in Silicon Valley as an innovation partner to co-develop next-generation memory for AI computing. The collaboration spans four areas: new memory cells and structures, multi-chip stacking, advanced packaging for stacked memory, and materials engineering designed with future high-volume manufacturing in mind. KIOXIA advanced memory development general manager Minori Kajimoto said the two teams will work side by side at the center to speed the path from research to commercialization. The announcement also adds another memory partner to Applied Materials’ EPIC Center, a collaborative R&D site unveiled in May 2023 and expected to open this year. The company had originally planned to invest as much as $4 billion over seven years, but now says capital spending will gradually rise to about $5 billion as customer projects move forward. Samsung Electronics and SK Hynix had already joined, along with Broadcom, SCREEN and the University of California, Berkeley. On the stock side, AMAT has risen from $268.87 at the start of 2026 to a latest close of $511.38, up 90.2%. Morgan Stanley cut its price target to $563 from $642 on Sept. 28 while keeping an Equal-weight rating, even as it raised its 2027 revenue forecast to $50.7 billion and EPS estimate to $20.92.

70

KIOXIA

2026-09-29 20:21:41

KIOXIA Said to Join Applied Materials’ $5 Billion AI Memory EPIC Center

Storage chip maker KIOXIA is said to be joining Applied Materials’ $5 billion EPIC, or Electronic and Photonic Integrated Center, according to a Techub news brief that cited Crypto Briefing. The center is focused on innovation and collaboration in AI memory technology. The report said EPIC is intended to speed up advances in semiconductor technology, with particular attention on memory solutions built for artificial intelligence applications. If KIOXIA joins the initiative as reported, its participation is expected to strengthen collaborative innovation at the center in the AI memory segment. No further details were provided in the brief about the timing of the move or the specific scope of KIOXIA’s role within the center.

180

Tencent

2026-09-21 10:37:37

Tencent open-sources T-Mem, a long-term memory system accepted by EMNLP 2026

Tencent’s PCG team has open-sourced T-Mem, a long-term memory system whose paper has been accepted to the main conference of EMNLP 2026. The project focuses on a specific weakness in conventional memory systems: retrieval breaks down when a new query shares few or no keywords with earlier stored information. T-Mem addresses that by recording not just a fact, but also the kinds of future situations where that fact may become useful. The example cited in the release is a user saying a colleague is allergic to seafood, then later asking where the team should go for a group meal. A standard memory system may fail to connect the two because the wording barely overlaps. T-Mem is designed to link "seafood allergy" with the scenario of choosing a restaurant for a gathering, allowing the earlier memory to be recalled later. Evaluation results said ordinary memory systems saw performance drop sharply in low-keyword-overlap settings, while T-Mem was less affected. The report also said results declined noticeably after removing Horizon Trigger, the component used to predict future scenarios, identifying it as central to T-Mem’s associative recall.

310

Samsung

2026-09-07 05:24:37

Samsung's HBM Shipment Expansion to Offset General Storage Slowdown in 2027: Mirae Asset

Mirae Asset Securities analyst Kim Young-gun projects that Samsung's HBM shipments for AI will offset the slowdown in general memory chip growth in 2027. HBM capacity and ASP are expected to surge 57% and 49% respectively, outpacing general storage. Profit may be weighed by won strength but remain solid.

780

Federal Reser

2026-09-04 22:50:57

Fed's September Rate Decision Could Impact Micron's AI Memory Demand

The Federal Reserve's interest rate decision on September 15-16 could significantly impact Micron Technology's quarterly earnings, with market-implied rate hike probabilities ranging from 35% to 66%. The decision comes two weeks before Micron's September 30 earnings report and will influence AI data center financing costs, thereby affecting demand for Micron's AI-driven memory products.

780

Google Cloud

2026-09-02 02:17:35

Google Cloud: AI Is Memory-Bound Now, HBM Tops 75% of AI Server BOM

At SEMICON Taiwan 2026, Google Cloud senior director Nikhil Cherian said AI workloads are now memory-bound, with high-performance memory exceeding 75% of AI server hardware BOM cost. Google unveiled a two-pronged approach: TPU 8i (288GB HBM, 384MiB on-chip SRAM, zero off-chip latency) and TPU 8t (9,600 chips, 2PB pooled HBM) to split inference and training, plus a training-free lossless quantization algorithm called TurboQuant that compresses KV cache from 32-bit to 3-bit, cutting memory footprint sixfold and accelerating attention computation 8x. The algorithm also integrates older DRAM technology to extend component lifecycle.

790

SK Hynix

2026-08-27 03:20:55

SK Hynix Falls to a Key Midpoint as Traders Focus on Support Levels After a 45% Pullback

SK Hynix, which only two months ago ranked as the most valuable listed company in South Korea, is now trading roughly 45% below its record high of KRW 2.987 million set on June 25. On Aug. 25, the company’s production workers’ union narrowly rejected a previously agreed wage deal by 25 votes, sending the stock sharply lower intraday before it recovered much of the drop and closed down about 1.5% at KRW 1.646 million, while the KOSPI rose 0.2%. The current price sits just above the 50% retracement of the stock’s 52-week range from KRW 253,000 to KRW 2.987 million, placing KRW 1.62 million at the center of the technical debate. Traders are also watching the Aug. 25 intraday low of KRW 1.557 million, the KRW 1.943 million 38.2% retracement, and the KRW 2 million psychological level for signs of whether the recent downtrend can stabilize or extend. Beyond chart levels, investors are weighing a KRW 40 trillion share buyback announced on Aug. 19, the restart of labor negotiations after the union vote, Nvidia’s Aug. 26 earnings as an external demand signal for AI memory, and competitive pressure from Samsung Electronics in next-generation HBM products. Analyst consensus remains uniformly bullish, though target prices range from KRW 1.2 million to KRW 5.3 million, highlighting wide disagreement over the memory cycle.

1850