Intel's long-awaited graphics card update is arriving for data centers and workstations with the introduction of "Crescent Island," now detailed. At Hot Chips 2026, the company revealed more about the upcoming graphics card. Based on the Xe3P GPU IP, the graphics card is designed as a 350 W, air-cooled, PCIe format compatible with any configuration. The GPU itself features 32 Xe3P cores, 256 XMX engines, 32 MB of unified L2 cache, and 160 GB of LPDDR5X memory. This memory configuration is exclusive to Intel-branded "Crescent Island" graphics cards, while the company allows its ODMs to equip the card with up to 480 GB of LPDDR5X memory for maximum VRAM capacity. Just a few of these in their maximum configuration could power frontier open-source models locally.

What is particularly interesting about Xe3P is the pipeline, as XMX engines have been redesigned from the four-deep systolic design on the regular Xe3 to 16-deep processing on Xe3P. Additional details include FP4 data format support alongside FP8, as well as massive register files of 1 MB per core and L1 cache of 512 KB per core. In total, there are 256 Vector engines and 256 of the new XMX engines, each split into groups of eight per Xe3P core. Additionally, in Xe3P, there are no graphics processing elements that would output any imagery, as the silicon is designed purely for AI inference workloads that include long context, agentic tasks.

AMD has updated its documentation page for the late-July launch of the Radeon RX 9050 graphics cards, which include two models: one with 4 GB and another with 8 GB of GDDR6 memory. Both cards feature 1,024 Streaming Processors across 16 Compute Units and can boost up to a 2.6 GHz clock speed. However, the lower-end GPU has only 4 GB of GDDR6 memory on a 64-bit bus, resulting in a memory bandwidth of 144 GB/s, which is significantly lower than the 8 GB version. The 8 GB version of the Radeon RX 9050 uses 8 GB of GDDR6 on a 128-bit bus, providing a memory bandwidth of 288 GB/s, doubling the throughput.

Another difference lies in the Infinity Cache within each GPU die. The 8 GB edition has 32 MB of Infinity Cache, while the 4 GB edition has only 16 MB. Initially, the listing for the 4 GB edition did not reveal the difference in Infinity Cache size, but AMD's documentation confirms it. Naturally, this will result in some performance differences, as larger cache buffers tend to improve performance significantly.

Micron Technology, Inc. (Nasdaq: MU) today unveiled Micron Research Labs, a U.S.-based long-horizon premier research institution headquartered in Boise and backed by a planned $10 billion investment over the next decade. Building on Micron's technology and manufacturing leadership, the new hub will bring together customers, academia, government and the broader semiconductor ecosystem to pursue breakthroughs beyond today's technology roadmaps and help define what's possible in the decades ahead. Key research areas include critical memory technologies, advanced memory and compute architectures, packaging and future semiconductor manufacturing.

As the first dedicated memory research hub of its kind in the U.S., Micron Research Labs will center on a flagship Boise campus designed to support advanced research across critical technology domains. The planned investment will also fund extensive university collaborations, global satellite labs and deep ecosystem partnerships, creating a connected research network focused beyond the next generation of technologies for the AI era.

Memory pricing is a hot topic these days as consumers struggle to purchase any, with little hope of prices dropping. Memory costs are reaching new all-time highs, with various trackers reporting that prices in August this year are up to five times more expensive than the same month last year for DRAM kits. According to data from PCPartPicker, DDR5 memory prices have increased anywhere from 372% for 2 x 16 GB 4,800 MT/s kits to 485% for 2 x 32 GB kits running at 5,600 MT/s. A 6,000 MT/s 64 GB kit, which used to cost $222, is now considered a luxury investment at an astounding $1,272. At the TechPowerUp editorial team, we remember when that was the entire budget for building a very decent high-end gaming PC.

PCPartPicker also shows that DDR4 kits have increased by 120% to 177%, which is still significant but mainly affects those using the older DDR4 technology. For those on a newer DDR5 platform, the change is much more pronounced. Note that PCPartPicker only tracks pricing in the United States, while German websites like ComputerBase have reported a 345% year-over-year increase for August in the Eurozone, which is much higher than anything seen before. With no signs of slowing down, we might see even higher year-over-year changes.

Rumors about the upcoming Sony PlayStation 6 have been flying around for months now, purporting to reveal everything from pricing and specs to an approximate release window. Now, the gaming giant's CEO, Hiroki Totoki, has addressed the PS6 launch date in a new interview with The Wall Street Journal.

In the interview, Totoki explains that the volatility of the current PC and console hardware market makes it difficult to pin down pricing or a release date. He notes that memory prices are still expected to be high in the 2027 financial year, thanks to an ongoing supply shortage, and, given those market conditions, Sony has to "think carefully" about its launch strategy for the new console.

Chinese memory giant CXMT has reportedly achieved a significant milestone in DDR5 production, with manufacturing yields now comfortably above 90%. This is comparable to what major manufacturers like Samsung, SK hynix, and Micron are achieving. According to reports, CXMT uses a 17 nm node for DDR5 chip production and has reached a 90% yield for these chips. This is only 2-3% lower than Samsung's yield on its 10 nm class node for DDR5 production, indicating that CXMT's production quality is matching that of the world's largest memory maker. This is a remarkable achievement for a company that excels in cleanroom construction time, taking only 12 months to build its fab's cleanroom compared to the two years others require. As the company plans to end the year with 350,000 DRAM wafers per month, high yields mean there is little waste or leftover.

Chinese MyDrivers also notes that PC OEMs have mixed reactions to adopting CXMT DRAM in their products. For instance, Dell has completely banned the use of CXMT in its machines, while HP only allows CXMT DRAM in its Chinese PCs. Major Taiwanese brands such as ASUS and Acer have increased adoption in some models, as they have more flexibility in supply chain procurement. However, both companies use CXMT memory for PCs sold in mainland China and some emerging markets, meaning the West is still far from receiving PCs with CXMT memory modules. PC OEMs are also cautious about procuring CXMT memory due to concerns about disrupting their relationships with the big three memory makers.

Kioxia Corporation, a subsidiary of Kioxia Holdings Corporation (TOKYO: 285A) and Sandisk Corporation (Nasdaq: SNDK) today unveiled their new 9th-generation, high-performance 2 Tb QLC 3D flash memory technology designed to support the growing storage demands of AI-driven infrastructure. The technology demonstrates how architectural innovation can help storage advance at the fast pace required by AI-driven workloads, while supporting more capital-efficient manufacturing.

Leveraging the companies' CMOS directly Bonded to Array (CBA) architecture, the new technology combines an advanced CMOS wafer with a proven memory-array platform, along with significant design and device innovation, to deliver performance improvements and enable faster deployment of advanced storage solutions for cloud and data-intensive applications. Compared with the previous 8th-generation 2 Tb QLC, the new technology delivers higher write and read bandwidth due to a 6-plane architecture and performance enhancements, and better write and read power efficiency. And it also delivers a NAND interface speed of 4.8 Gb/s, a 33% improvement over 8th-generation devices.

SK hynix Inc. announced today that it has decided to build new fabs in Yongin and Cheongju to respond to the continuously growing demand for memory in the AI era.
The company approved a total investment of approximately 54 trillion, 35.2 trillion won in Yongin "Y2" fab and 19.1 trillion won in the Cheongju "M17" fab, in a board meeting.

This decision represents execution of the mid-to-long-term investment strategy announced by the company last June. Under its master plan to invest 600 trillion won in the Yongin Semiconductor Cluster and 100 trillion won to expand its Cheongju production base, SK hynix is currently constructing its first fab in Yongin (Y1). With this latest decision, the company is embarking on the construction of subsequent fabs in both Yongin and Cheongju.

According to Counterpoint Research's Global Memory Tracker update for the second quarter of 2026, memory manufacturers are adjusting their share of the global DRAM revenue. The report highlights that South Korean memory makers remain dominant, with 65% of global DRAM revenue going to Samsung and SK hynix. Samsung holds the top position, capturing 39% of the global DRAM revenue. The South Korean giant has reportedly resolved its manufacturing issues, and its foundry has delivered strong performance, widening the gap with other manufacturers. Price increases in regular DRAM and specialty solutions like HBM have significantly boosted the company's revenue, securing its stable number one spot.

In second place, SK hynix holds about 26% of the global revenue share. This is a decrease from the 39% share the company held in Q2 2025, despite its revenue growing 214% year-over-year. This indicates that while the DRAM market is expanding, Samsung is capturing a larger share of the new market growth, leading to a higher overall revenue share. SK hynix was the first to implement long-term agreements with customers, where price volatility is predetermined. This strategy involves encouraging customers into multi-year agreements rather than capitalizing on short-term price fluctuations.

Samsung Electronics a global leader in advanced memory technology, today showcased its latest AI memory innovations at Future of Memory and Storage (FMS) 2026 in Santa Clara, California.

With broad expertise spanning DRAM, NAND, enterprise storage and relevant advanced semiconductor technologies, Samsung presented its latest AI memory portfolio and technology roadmap, including a preview of its zHBM and zNAND-O concept models, the industry-first introduction of its V10 BV-NAND with 400-plus layers and a comprehensive range of solutions designed to advance future AI infrastructure. Samsung is also featuring its AI cloud server-inspired booth at FMS 2026, where approximately 30 memory and storage technologies are on display.

NVM Express, Inc. today announced the release of all 11 of the NVM Express (NVMe) specifications, incorporating new features and engineering change notices (ECNs). These updates represent the next step in the evolution of NVMe technology, with security, management and sustainability extensions to address the growing demands of data-intensive applications in AI, cloud, enterprise and client environments.

The updated specifications include:

• NVMe 2.4 Base Specification
• Command Set Specifications (NVM Command Set 1.3, ZNS Command Set 1.5, Key Value Command Set 1.4, Subsystem Local Memory Command Set 1.3, Computational Programs Command Set 1.3)
• Transport Specifications (NVMe over PCIe Transport 1.4, NVMe over RDMA Transport 1.3, NVMe over TCP Transport 1.3)
• NVM Express Management Interface (NVMe-MI) 2.2 Specification
• NVMe Boot 1.4 Specification

According to DigiTimes, industry insiders are now claiming that the entire memory production capacity for 2027 has been fully booked, well ahead of the schedule the industry expected. All of the big three memory makers—Samsung, SK hynix, and Micron—have sold out their memory capacity for the upcoming year, leaving no room for additional supply for third parties to purchase more DRAM wafers. Interestingly, the report notes that everyone from OEM server makers and PC makers to AI labs knows that July and August are crucial months for securing allocations for the following year. However, some businesses have been kept in the dark because nobody has been public about this out of fear of losing their allocation.

Memory makers have been pushing customers into long-term agreements with contracts ranging from three to five years, using an advance payment deposit model where customers essentially pre-pay for some of the memory capacity. These deals are made on an individual customer level. All of this combined signals that 2027 will be the worst year in recent history regarding memory shortages, meaning that only those who managed to sign agreements for supply will benefit, while the rest are left without as there simply isn't enough available DRAM. Only about 60-70% of the estimated demand can be met in 2027, leaving a significant amount of demand unmet and waiting for the 2028 supply.

Kioxia Corporation, a world leader in memory solutions, today announced that it will exhibit the KIOXIA XL1 series, a Compute Express Link (CXL) compatible memory expansion module, at the Kioxia booth at FMS: the Future of Memory and Storage, taking place August 4 - 6 in Santa Clara, CA. The product features KIOXIA XL-FLASH, a low-latency and high-performance flash memory technology that enables flash memory, traditionally used for storage applications, to also be utilized as a memory expansion solution. In addition, it is being designed and developed to support the growing volume of memory-intensive AI applications. Evaluation samples are scheduled to be shipped to industry ecosystem collaborators in August 2026.

In recent years, the growing number of AI workloads has significantly increased memory demand in data centers, creating serious challenges such as capacity shortages and rising costs. In response to this important market need, Kioxia is targeting an innovative memory expansion solution for hyperscale environments. The KIOXIA XL1 series memory expansion module combines low-latency, high-performance XL-FLASH with a CXL interface and improves DRAM utilization efficiency by placing less frequently accessed data on the module. It bridges the performance and cost gap between conventional DRAM and SSDs, enabling memory expansion for demanding AI applications.

Lexar has introduced a new THOR RGB DDR5 Gen 2 32 GB memory kit, consisting of two 16 GB DDR5 modules, specifically for the Chinese market. The kit features DDR5 memory running at 7,200 MT/s, with the actual DDR5 modules produced by an unidentified Chinese DRAM manufacturer, likely CXMT. The latest THOR RGB 32 GB kit offers memory with 38-48-48-86 CAS latency timings at 1.40 V and supports Intel's XMP 3.0 and AMD EXPO profiles to achieve maximum frequency and overclocking potential. While the exact overclocking range of these modules is unknown,

Lexar has not explicitly disclosed which Chinese DRAM maker is used for this DDR5 kit, but it is assumed that CXMT DRAM is housed under the sandblasted aluminium heat spreaders. The company tested a variety of ASUS and MSI motherboards on both AMD and Intel platforms, even mentioning a 32 GB kit with 7,600 MT/s speeds, as opposed to this newly listed 7,200 MT/s kit. It is likely that more motherboards and vendors will be compatible without issues, but only those mentioned in the list below were tested. Lexar's official JD store lists the price for the THOR RGB DDR5-7200 32 GB kit at 3,999 RMB, which is approximately $592 at the time of writing, based on the current exchange rate. This kit is meant for the mainland China, with no mention of global availability.

Following large-scale dissatisfaction with the overall Windows experience, Microsoft in March 2026 issued a statement explaining that it would be working to deliver "smoother navigation and more reliable performance for everyday tasks." Now, over four months later, Pavan Davaluri, President of Windows + Devices, has provided an update on the progress of this initiative in a new X post.

In this X post, one of the more notable improvements Davuluri claims Microsoft has made during recent months—along with improvements to the Taskbar and Start—is improved memory efficiency. He explains that Microsoft has implemented "substantial improvements in memory efficiency and latency," in order to "accelerate the modernization of Windows 11," that it has reduced memory overhead across apps and components, and that the company has tuned WinUI 3 to make built-in apps use less memory. Microsoft also outlined its plans for the future of Windows 11 memory efficiency improvements, specifically stating that the Windows team will be working on "memory optimization for 8 GB and above."

Montage Technology today announced the industry's first trial production of its CXL 3.2 Memory eXpander Controller (MXC) chip, designed to address escalating demands for memory expansion and resource pooling in AI, cloud computing, and data center environments. Leveraging CXL technology, the MXC chip enables customers to build memory infrastructure with higher capacity, improved resource utilization, and greater flexibility.

The CXL 3.2 MXC complies with CXL Type 3 specifications, including both CXL.mem and CXL.io protocols. It is designed with PCIe 6.x and CXL 3.2 protocol support, delivering data transfer rates of up to 64 GT/s. The chip integrates dual DDR5 memory controllers that support up to 8000 MT/s DDR5 memory. By converting host-side CXL memory requests into DDR commands in real time, it enables efficient data exchange between the host and backend memory, effectively breaking through the memory capacity limits of traditional servers.

Late last month, reports indicated that AMD was preparing to increase the prices of its graphics cards by about 10% in July due to ongoing memory shortages. However, according to the latest update from Chinese Board Channels, AMD will implement at least a 10% price hike in August, delaying the increase by about a month. This development affects AMD, which secures GPU dies from its manufacturer TSMC and makes agreements with GDDR6 memory makers like SK hynix, Micron, and Samsung. AMD negotiates rates and quantities for GDDR6 memory and bundles it with a GPU die to form a GPU kit, which is later shipped to its add-in board (AIB) partners.

AMD has now notified its AIB partners that the price increase will be at least 10%, indicating a baseline double-digit percentage increase, although the higher end of the estimate is unknown. Logically, SKUs with more VRAM are expected to carry higher premiums exceeding 10%, such as AMD's current best RDNA 4-based Radeon RX 9070 XT 16 GB. For SKUs with less VRAM, especially entry-level, OEM-exclusive ones like the Radeon RX 9050 4 GB edition, the price increase will likely be only 10%, as these graphics cards require fewer GDDR6 chips.

Apple has reported record quarterly results for the second quarter, ending with $109.4 billion in revenue, highlighting the resilience of its business in recent months. The company is also one of the largest supply chain operators globally, manufacturing electronics across mobile, laptop, and desktop computing. When asked about recent memory inflation during an earnings call, Apple's CEO Tim Cook explained that the company has benefited from stockpiling memory inventory and reduced costs for some other non-memory components in its bill of materials (BOM) for each device. Additionally, Apple sources memory from three suppliers: SK hynix, Micron, and Samsung, while also exploring other options. This suggests that the rumored evaluations of YMTC for NAND Flash and CXMT for DRAM are accurate, and Apple is likely undergoing some form of regulatory approval.You can see the statement from CEO Tim Cook below.

Renesas Electronics Corporation (TSE: 6723), a premier supplier of advanced semiconductor solutions, today announced its third-generation (Gen 3) DDR5 Multiplexed Rank Dual In-Line Memory Module (MRDIMM) chipset solutions, delivering DDR5 server-class MRDIMM speeds up to 16,000 mega transfers per second (MT/s). The new solutions are designed to address the growing demand for higher memory bandwidth driven by artificial intelligence (AI) data centers, cloud infrastructure and accelerated compute workloads.

Renesas will showcase memory interface components for Gen 3 MRDIMMs at FMS 2026 (Future of Memory and Storage Conference), Santa Clara, California, August 4-6, 2026. (Booth #807) They include third-generation Multiplexed Registering Clock Driver (MRCD, RRG5013) and Multiplexed Data Buffer (MDB, RRG5103) devices that support next-generation MRDIMM performance.

Innodisk, a leading global AI solution provider, introduces the DDR5 12800 MT/s MRDIMM to support the rapid growth of generative AI, large language models (LLMs), and robotics applications. With an industry-leading data rate of 12800 MT/s, this advanced memory module delivers exceptional bandwidth efficiency and scalability to meet the growing processing demands of AI-driven innovations.

Breaking the Memory Wall with Advanced Architecture
The growing complexity of AI and HPC workloads has exposed the limitations of traditional memory architectures, often referred to as the "memory wall"—a bottleneck where memory bandwidth fails to keep up with processing speeds, leading to inefficiencies. The DDR5 12800 MT/s MRDIMM overcomes this challenge with its innovative Multiplexed Registering Clock Driver (MRCD) and Multiplexed Data Buffer (MDB) architecture. By enabling simultaneous two-rank access, this breakthrough design increases bandwidth by 60% compared to traditional DDR5 8000 MT/s RDIMMs, significantly improving data transfer speeds and computational efficiency. By reducing the load on the memory controller and lowering latency, this module enhances AI model training, large language models (LLMs), and robotic automation applications.

TrendForce's latest research on the memory industry reveals that AI will remain the primary driver of memory demand in 2027. However, market dynamics are expected to diverge between DRAM and NAND Flash.

On the DRAM side, continued HBM capacity allocation, robust AI server demand, and growing procurement of CPU memory and HBM are expected to keep supply constrained and prices on an upward trajectory. By contrast, NAND Flash is projected to enter a looser supply environment in 2H27 as new production capacity comes online while consumer electronics demand remains weak, increasing downward pressure on prices.

Chinese memory maker CXMT has debuted on the Shanghai A-Class STAR stock market with one of the biggest public offerings in China. Today, CXMT's Initial Public Offering (IPO) saw the price of CXM jump from CNY 8.66 ($1.28). At the time of writing, the stock, under the ticker "688825.SS," is trading at CNY 49.00 ($7.24), indicating a 465.82% gain on its first trading day. Initial projections estimated CXMT's market capitalization at CNY 579 billion, or $78.11 billion, but the valuation has increased significantly, now reaching $513.75 billion. This makes it the world's 24th most valuable company by market capitalization, surpassing Tencent, Mastercard, and even the technology giant Intel.

This isn't a surprise, as CXMT will quickly generate revenue that will far exceed its IPO price, hence the valuation just got stronger. Some analysts, including those from SemiAnalysis, estimate that CXMT will generate about $50 billion for the entirety of 2026. In the first quarter of 2026 alone, CXMT reported $7.3 billion, marking approximately 700% growth compared to the same period a year ago, with operating margins around 70%. Massive quarterly growth, new capacity coming online, and an unprecedented demand for DRAM have pushed CXMT's valuation nearly five times higher than originally planned.

Samsung Electronics and Broadcom today announced the signing of a memorandum of understanding (MOU) to expand their strategic collaboration across memory and foundry technologies, supporting the next generation of AI infrastructure. The MOU was announced during the AI Summit, held at The Midway in San Francisco. Attendees included Jinman Han, President and Head of Samsung Electronics' Foundry Business, Hock Tan, President and CEO of Broadcom, senior executives from both companies and representatives of the Korean government. The companies expect the collaboration to be estimated at more than $200 billion across memory and foundry over the next five years through 2030.

On the memory front, Samsung and Broadcom plan to pursue a strategic collaboration for the supply of industry-leading memory solutions, including High Bandwidth Memory (HBM), supporting Broadcom's next-generation AI accelerators. In foundry, the companies' collaboration focuses on Samsung's 2-nanometer (nm) and below process technologies for Broadcom's products, including Wireless Broadband Communications (WBC) solutions. The collaboration is expected to extend to advanced packaging technologies built on Samsung's 2 nm process, including 2.3D and 2.5D integration, to enable higher-performance and more power-efficient AI and networking silicon.

CXMT, the world's fourth-largest DRAM maker, is now reportedly charging more for its 64 GB DDR5 server RDIMM modules than Samsung, which is already asking $1,240 per unit. The numbers come from Reuters and flip the script on what most people assumed about CXMT, that it would undercut Samsung, SK Hynix, and Micron on price. As we reported previously, CXMT's edge was never really about price. It was about supply availability at a time when demand was consuming everything that Samsung, SK Hynix, and Micron could provide. That demand has since surged roughly tenfold, and CXMT is now under the same pricing pressure as everyone else. The situation was so serious that, reportedly, this year, some Chinese electronics firms filed complaints with China's Ministry of Industry and Information Technology. The companies complained over price hikes from both CXMT and YMTC, claiming the increases were delaying product launches.

CXMT is also building two new plants in Shanghai and Hefei, and is involved with ongoing talks about a third fab. When those two fabs start production, CXMT's monthly wafer output will pass 600,000. That is more than doubling current capacity and the Chinese company could match Micron this year and possibly overtake it by 2030 if plans stay on track. Financially, CXMT reported Q1 2026 revenue of about RMB 50.8 billion (~$7.5B), up 719% year over year, with net profit rising over 1,200%. The company is forecasting first-half 2026 revenue of RMB 110-120 billion ($16-$17.7 billion), representing growth of over 600% compared to the same period last year. A historic IPO is also in the works alongside the fab expansion push. CXMT currently holds around 8-10% of the global DRAM market, behind Micron at 24%, SK Hynix at 29%, and Samsung at 36%.

Gigabyte Technology Co. Ltd, a leading manufacturer of motherboards, graphics cards, and hardware solutions, today announced that its full lineup of DDR5 motherboards now supports CXSH (CXMT) chip-based memory, giving consumers more options and greater availability amid an ongoing industry-wide memory shortage. The BIOS optimization enables AM5 800/ 600 series motherboards to run CXSH (CXMT) chip-based memory stably at 8200 MT/s, while Intel 800/ 700 series platforms support CXSH (CXMT) chip-based memory out of the box.

Following AMD's release of AGESA 1.3.0.1c and Gigabyte's own memory optimization technology, AM5 motherboards across both the 800 series and 600 series now support CXSH (CXMT) chip-based memory, including 16 Gb and 24 Gb densities. With this update, Gigabyte AM5 motherboards can now stably run these DIMMs at speeds of up to 8200 MT/s, breaking the previous AM5 platform limitation and giving builders more performance headroom on AMD's platform. Meanwhile, Gigabyte's Intel 800 and 700 series motherboards already support CXSH (CXMT) chip-based memory through their current shipping BIOS, with no additional update required.