IMAGIS said Chief Executive Officer Lee Dong-seop, who was appointed in March, has resigned after serving for just four months.

The company disclosed on July 31 that its board approved the resignation at a meeting held the same day. Lee stepped down before completing his term, which was scheduled to run through March 2029. The company did not disclose the reason for his departure.

Chairman Kim Jung-chul will return as chief executive. Born in 1964, Kim has led IMAGIS since the 2000s and is currently the company's largest shareholder with an 18.43% stake.

In March, IMAGIS appointed Lee, a former SK hynix executive, as CEO. At the time, the company said it planned to expand into physical AI and robotics-based solutions, adding that the appointment was intended to drive revenue growth and strengthen management accountability.

Since then, the company has secured a major new order. On June 8, IMAGIS announced an automotive semiconductor supply contract worth approximately 5.8 billion won. The deal was equivalent to 43.71% of the company's revenue of about 13.2 billion won in the previous year and marked a significant win as the company seeks to recover from three consecutive years of losses since 2023.

IMAGIS has projected revenue of 21.0 billion won for 2026, which would represent its highest annual sales since 2017.

The leadership change comes as the company works to capitalize on growing opportunities in automotive semiconductors and next-generation AI-related applications.

Samsung Electronics' next-generation gallium nitride (GaN) power semiconductor foundry project has encountered a setback during trial production, raising concerns that commercial production may be delayed beyond this year. The issue emerged after some test chips reportedly failed to operate under extended high-temperature conditions.

According to multiple industry sources on July 31, Samsung has established a dedicated 8-inch GaN semiconductor foundry line at its Giheung campus in Yongin with monthly capacity of approximately 1,300 to 1,500 wafers. The company recently conducted trial production of chips designed by fabless semiconductor companies. During testing, some of the chips reportedly stopped functioning after being exposed to temperatures of 100 to 200 degrees Celsius for roughly 1,000 hours.

GaN semiconductors are commonly used in high-voltage applications of up to 1,200 volts and in environments exceeding 200 degrees Celsius. Their ability to operate reliably under high-temperature conditions has made them increasingly attractive for electric vehicles, robotics and data center applications. Industry sources said the reported reliability issues could limit the usability of chips manufactured using the current process.

Samsung is said to be investigating the root cause of the problem. Industry sources identified the wafer manufacturing process as a leading area of focus. According to those sources, process modifications introduced to achieve targeted device performance may have contributed to the reliability issue.

GaN devices are typically fabricated on epitaxial wafers, in which a GaN layer is grown on a silicon (Si) wafer through an epitaxy process. If the quality or structure of the GaN layer is not fully optimized, defects can become more pronounced under high-temperature and high-voltage operating conditions. Samsung manufactures its own epitaxial wafers using metal-organic chemical vapor deposition (MOCVD) equipment.

A person familiar with the matter said Samsung may have focused on achieving the target threshold voltage of 2 to 3 volts for power semiconductors while overlooking potential effects on device longevity. GaN semiconductors generally require a threshold voltage of around 2 volts. If the threshold voltage is too low, the device may respond to very small currents and become more susceptible to operational errors.

Samsung reorganized its LED business in 2023 and established the Compound Semiconductor Solutions (CSS) division to focus on GaN semiconductors. The division originally planned to launch a dedicated GaN foundry line in 2025. However, weaker-than-expected conditions in the power semiconductor market prompted Samsung to postpone the schedule to the end of 2026.

Industry sources now believe the latest reliability issue could push the timeline back by several additional months. One Samsung GaN foundry customer, speaking on condition of anonymity, said Samsung had already entered the market later than competitors because the company prioritized major businesses such as memory semiconductors. Companies including Infineon Technologies, STMicroelectronics, Texas Instruments and Innoscience have already been shipping GaN semiconductors commercially for one to two years.

Some fabless semiconductor companies are reportedly evaluating alternative foundry partners until Samsung resolves the issue. Potential alternatives include GlobalFoundries, Powerchip Semiconductor Manufacturing Corporation and X-FAB. Unlike silicon semiconductors, where mask costs are typically high enough to encourage production at a single foundry, compound semiconductor mask costs are relatively lower, making it less burdensome to work with multiple manufacturing partners.

A Samsung Electronics representative said the company is still in the pre-production stage and is working through normal development challenges.

"It is part of the process of addressing trial-and-error issues before mass production," the representative said. "It is too early to determine whether these findings constitute a defect."

SK hynix said it was unable to disclose additional shareholder return plans because of restrictions related to its American depositary receipt (ADR) offering and US securities regulations. The company expects to provide more specific details after the ADR process is completed in early August.

During the company's second-quarter earnings conference call on July 29, Song Hyun-jong, president of SK hynix's Corporate Center, said regulatory requirements tied to the ADR offering limited the company's ability to disclose new material information.

"Due to regulations and procedures related to the ADR offering, there are restrictions on providing additional material information that was not disclosed during the offering process," Song said. "We are reviewing various shareholder return options and will share specific plans with the market once they are finalized."

The restrictions are understood to stem from communication rules under Section 5 of the US Securities Act of 1933, which limit communications that could be interpreted as an offer to investors during a registered securities offering.

The US Securities and Exchange Commission (SEC) broadly interprets communications that could stimulate public interest in an issuer or its securities as potential offers. Violations of these restrictions are commonly referred to as "gun jumping."

The term originates from track-and-field competitions, where an athlete starts running before the starting gun is fired. In securities markets, it refers to the disclosure of new material information not included in an offering prospectus before the registration process is completed, potentially influencing investor decisions or market sentiment.

Because shareholder return measures such as share buybacks or special dividends could affect investment decisions, SK hynix was limited in what it could disclose while the registered offering remained in progress.

The company is expected to provide more detailed shareholder return plans after the ADR-related procedures conclude in early August.

Under Rule 174 of US securities regulations, prospectus delivery requirements remain in effect for up to 25 days after a newly registered public offering. During that period, investors must continue to receive offering-related information while post-offering procedures are completed.

An SK hynix spokesperson said the company expects to announce more specific shareholder return measures once the ADR process is finalized. The remaining post-offering procedures are expected to be completed on August 4.

SK hynix reported record quarterly results in the second quarter, posting revenue of 79.3187 trillion won and operating profit of 60.5426 trillion won. Despite the strong earnings, the company's share price fell nearly 10% on the day of the announcement, as investors were disappointed by the absence of additional shareholder return plans.

At the company's annual shareholders meeting earlier this year, Kwak Noh-jung, president of SK hynix, said the company aims to strengthen its financial position by building net cash to 100 trillion won over the long term while continuing to consider dividends and share repurchases.

As of the end of June, SK hynix held 69.37 trillion won in net cash, up from 35.01 trillion won at the end of March. The company is expected to reach 100 trillion won in net cash by the end of the third quarter.

SK Group Chairman Chey Tae-won has made his first-ever direct open-market purchase of SK hynix shares, a move interpreted as a display of his commitment to responsible management following the company's record quarterly earnings.

According to a regulatory filing on DART, the Financial Supervisory Service's electronic disclosure system, Chey purchased 3,620 common shares of SK hynix on the open market on July 30. Based on the stock's closing price of 1,322,000 won that day, the purchase was valued at approximately 4.79 billion won.

The transaction increased Chey's direct ownership of SK hynix shares from zero to 3,620 shares. Total holdings by the company's largest shareholder and related parties increased from 146,124,531 shares to 146,128,151 shares, while the ownership stake remained unchanged at 20.00%.

The purchase marks the first time Chey has directly acquired SK hynix shares in the open market. The move is seen as a display of responsible management, reflecting confidence in the company's mid- to long-term growth potential and corporate value amid the recent decline in the share price.

SK hynix reported record quarterly results on July 29, posting revenue of 79.3187 trillion won and operating profit of 60.5426 trillion won for the second quarter, both all-time highs on a quarterly basis. After the announcement, however, the shares weakened as profit-taking coincided with a correction in the South Korean stock market.

The purchase has also drawn attention because it came after the Seoul High Court, in a remanded case on July 24, ordered Chey to pay 944 billion won in property division to Roh Soh-yeong, director of Art Center Nabi. Analysts said the move demonstrates Chey's willingness to invest in a key group affiliate despite the substantial financial burden from the ruling.

Intel posted results that beat market expectations on surging demand from artificial intelligence (AI) data centers. With its third-quarter outlook also exceeding expectations, the company's shares jumped about 9% in after-hours trading.

Intel said in second-quarter results released on July 23 local time that revenue rose 25% from a year earlier to $16.1 billion, well above the market forecast of $14.4 billion. Adjusted earnings per share came to $0.42, double the market forecast of $0.21.

Demand for server central processing units (CPUs) led Intel's second-quarter performance. Revenue from its data center and AI business was about $6.26 billion, above the market estimate of $5.37 billion. Intel said earlier this year that CPU orders had exceeded its production capacity.

Revenue from the notebook and desktop PC business was about $8.88 billion, against a market estimate of $7.89 billion. Intel said the share of low-cost PC chips was reduced while the share of high-specification chips expanded, and that although unit sales declined, the average selling price (ASP) rose.

The foundry business, long a sore point for the company, posted revenue of $5.77 billion, above the market estimate of $5.55 billion. It is still in the red, however, with a loss of about $2.089 billion.

Internal transactions in the second quarter came to about $5.5 billion. Internal transactions refer to dealings between business units within the company. They are counted as revenue for each business unit but are deducted when calculating consolidated revenue to prevent double counting.

The third-quarter outlook also exceeded market expectations. Intel guided for third-quarter revenue of $15.8 billion to $16.8 billion and adjusted earnings per share of 38 cents. The average revenue estimate among analysts compiled by LSEG was $15.1 billion, with adjusted earnings per share of 27 cents.

Intel has also secured long-term supply contracts that promise recurring revenue. Chief Financial Officer David Zinsner told Reuters that the company has signed three- to five-year contracts with customers covering server CPUs and XPUs. Some contracts include both volume and pricing, and can be renegotiated depending on market conditions.

Intel is also raising capital expenditures to match growing demand. "We have raised this year's capital spending outlook to $20 billion from $18 billion," Zinsner said. "We expect next year's investment to increase meaningfully as well."

"AI is driving an unprecedented level of demand for computing," Intel Chief Executive Officer Lip-Bu Tan said. "We can drive sustained growth in areas such as CPUs, application-specific integrated circuits (ASICs), advanced packaging and foundry." He added, "The second-quarter results mark the highest revenue growth rate in the past 15 years," saying that "fast execution, accountable management and a customer-focused strategy made it possible."

SK Group Chairman Chey Tae-won must pay 944 billion won in property division to Roh Soh-yeong, director of Art Center Nabi, a South Korean court ruled in a remanded divorce case. The ruling included Chey's holdings in SK Inc. as part of the assets subject to division. The amount was about 400 billion won less than the 1.3808 trillion won awarded by the appellate court, but far higher than the 66.5 billion won ordered by the first-instance court.

The Seoul High Court's Family Division 1, presided over by Judge Lee Sang-joo, ruled on July 24 that Chey must pay Roh 944 billion won in property division, along with annual default interest of 5% from the date the judgment becomes final until payment is completed. The court ordered each party to bear its own litigation costs.

The court recognized Chey's holdings in SK Inc. as assets subject to division, concluding that a substantial portion of the couple's joint assets had been formed or acquired during their marriage. It set April 16, 2024, the date when oral arguments concluded in the divorce proceedings, as the valuation date for the assets. SK Inc. shares were trading at around 160,000 won at that time.

The court did, however, take into account the subsequent sharp increase in the value of Chey's SK Inc. shares when determining the amount of the property division. SK Inc. shares that Chey had gifted to relatives to maintain control of the group were excluded from the assets subject to division.

The court recognized Roh's contribution to the formation of the assets at 33.3%, or one-third. It found that Roh contributed to Chey's business activities and to the appreciation of SK Inc. shares by handling household affairs and child-rearing and through her external activities during the marriage. The contribution ratio was slightly lower than the 35% recognized by the appellate court.

The court did not take into account the 30 billion won in slush funds linked to former President Roh Tae-woo in calculating the property division ratio.

For the method of division, the court chose compensatory division, under which Chey will continue to hold the shares and pay Roh the shortfall in her portion in cash. The decision took into account the defense of management control and the stability of the group's governance structure.

The divorce proceedings began in 2015, when Chey publicly disclosed his intention to divorce and the existence of a child born outside the marriage. The legal dispute escalated when he filed for divorce mediation in July 2017.

In December 2022, the first-instance court ruled that Chey's SK Inc. shares were his separate property and excluded them from the assets subject to division, awarding Roh 66.5 billion won in property division and 100 million won in alimony. In May 2024, the appellate court determined that the shares were the couple's joint assets, recognized Roh's contribution at 35%, and ordered Chey to pay 1.3808 trillion won in property division and 2 billion won in alimony.

In October 2025, the Supreme Court found that the 30 billion won from the side of former President Roh Tae-woo appeared to be illicit funds raised during his presidency, and that it could not be taken into account as Roh Soh-yeong's contribution to the formation of the assets even if the money had flowed into SK Group. The court overturned the property division portion and sent the case back to the Seoul High Court. The divorce itself and the 2 billion won alimony payment were confirmed.

The business community is watching the impact the ruling will have on SK Group's governance structure. With Chey required to raise close to 1 trillion won for the property division payment, attention is turning to how he will secure the funds. Some observers have raised the possibility that he could use his SK Inc. stake as collateral or sell some assets, with SK Siltron cited as a prime example.

Immediately after the ruling, Chey's legal counsel issued a statement saying, "The divorce was confirmed by last year's Supreme Court ruling in a process of dissolving a marriage spanning nearly 20 years, and today's remanded ruling on property division followed. We feel apologetic for causing concern to many people." The statement added, "We respect the court's judgment, but we will closely review the details of the ruling before deciding our future response, including whether to appeal to the Supreme Court."

SK Hynix is accelerating construction of its Cheongju P&T7 advanced semiconductor packaging facility by bringing forward the opening of its first cleanroom by about three months.

According to industry sources on July 23, SK Hynix now expects to open the first cleanroom at P&T7 as early as July 2027. The company had previously announced an October 2027 target.

“SK Hynix recently began bidding for utility and infrastructure work at P&T7 earlier than expected,” an industry source said. “Internally, the company expects the cleanroom to open roughly one quarter earlier than originally planned.” Some equipment-related tenders have failed to attract bidders and are currently being reissued. Industry sources expect all utility infrastructure bidding to be completed by around August at the latest.

Several months after the bidding process is completed, SK Hynix will issue formal purchase orders (POs) to contractors. The contractors will then fabricate the utility infrastructure, deliver it to the site, and install it in the sub-fab. After commissioning is completed, the cleanroom will undergo final cleaning to establish a contamination-free positive-pressure environment. Although the timeline varies depending on the size of the facility, the entire process typically takes 11 to 12 months before a cleanroom is ready for operation.

On July 22, SK Hynix disclosed that it had shortened the cleanroom opening schedule for P&T7, although it did not specify the revised timeline. As investment spending is being brought forward, the initial construction budget has increased to 7.0931 trillion won. The company said the increase reflects the earlier recognition of expenditures that had originally been scheduled for later phases, while the total planned investment of 19 trillion won, announced in January, remains unchanged.

“Initial investment increased as we accelerated the cleanroom opening schedule to respond to rising production demand at P&T7,” SK Hynix said. “The board approved the revised investment amount on July 22, and the disclosure was made in accordance with regulatory requirements.”

P&T7 is SK Hynix’s dedicated advanced packaging and testing facility for AI memory products, including high-bandwidth memory (HBM). The plant is being built on a 230,000-square-meter site in the Cheongju Technopolis Industrial Complex and will include wafer test (WT) and wafer-level packaging (WLP) production lines. Total cleanroom space will reach approximately 150,000 square meters.

SK Hynix held a groundbreaking ceremony for P&T7 in April. At the time, the company planned to complete the 60,000-square-meter WT production line by October 2027, followed by the 90,000-square-meter WLP line in phases through February 2028. Completion of a semiconductor fab refers to the point at which the cleanroom has been opened in a contamination-controlled condition and is ready to receive manufacturing equipment.

Once operational, the P&T facility will package semiconductor chips produced at front-end fabs into final products and conduct quality verification. The plant will support production of HBM through 2.5D packaging technology, including through-silicon via (TSV) processing and vertical chip stacking for DRAM.

SK Hynix has recently accelerated large-scale capital investment to meet robust demand for AI memory. The company is also bringing forward the operational timeline for its Cheongju M15X fab and accelerating cleanroom construction at the first fab in its Yongin semiconductor cluster to expand manufacturing capacity. By rapidly increasing both front-end DRAM production and back-end packaging and testing capacity, SK Hynix is expected to strengthen its ability to meet growing demand for AI memory products such as HBM4.

Some industry observers believe the accelerated schedule may also reflect government efforts to speed up domestic semiconductor investment following plans for the Yongin semiconductor cluster and a proposed large-scale semiconductor manufacturing project in Gwangju.

“There is currently strong pressure at the national level to accelerate semiconductor fab construction,” a construction industry source said. “Engineering contractors are submitting proposals to customers outlining ways to shorten construction schedules for each phase of the project.”

Companies supplying utility infrastructure for semiconductor fabs include Wonik Holdings, STI, HanYang ENG, TEMC C&S, and CnG HiTech.

SK Hynix is discussing a plan to jointly operate Intel’s semiconductor fabrication facility in Ohio rather than acquire it, according to a media report.

U.S. news outlet Semafor reported on July 22 that Intel is seeking a partner to operate its semiconductor fab in Ohio. Citing people familiar with the matter, the report said SK Hynix is discussing a joint operating arrangement instead of an acquisition. It added that the two companies have not begun formal negotiations and that discussions remain at an early stage.

The possibility of a joint operation has emerged as Intel’s Ohio fab project continues to face delays. Intel broke ground on the Ohio facility in 2022 with a planned investment of 28 billion dollars. The site spans about 1.2 million pyeong and is large enough to accommodate a total of eight semiconductor fabs. The site is comparable in scale to the Yongin semiconductor cluster being developed by SK Hynix in Yongin, Gyeonggi Province, which covers about 1.26 million pyeong.

Intel had originally planned to open the Ohio fab in 2025 but postponed completion to 2030 last year. The company was facing financial difficulties at the time and also carried out large-scale restructuring of its foundry business. An Intel spokesperson said, “Our commitment to the Ohio fab remains unchanged,” adding that the company continues to invest in accelerating construction of the site.

A shortage of memory semiconductor supply is also strengthening the rationale for potential cooperation between SK Hynix and Intel. SK Hynix is expanding memory production capacity by building additional semiconductor fabs in Yongin, Cheongju, and Indiana, but believes further capacity will still be needed. SK Group Chairman Chey Tae-won also said at the Korea Chamber of Commerce and Industry (KCCI) Jeju Forum on July 15 that the company needs to build additional semiconductor plants, including in the United States, and is currently searching for suitable locations.

Building a new fab requires significant time from site selection through mass production. Utilizing a site where construction has already progressed could allow production capacity to be secured more quickly. The Ohio fab is reported to have completed part of its foundational construction.

SK Hynix did not deny the possibility of a joint operation. Asked whether the company had discussed jointly operating the Ohio fab with Intel, an SK Hynix spokesperson said, “We cannot confirm that.”

Qualcomm Technologies is expanding shipments of its Snapdragon platforms across Samsung Electronics' latest Galaxy device lineup.

​Qualcomm said on July 22 (local time) at Samsung Electronics' Galaxy Unpacked 2026 event in London that it will supply Snapdragon 8 Elite Gen 5, Snapdragon Wear Elite, and Snapdragon AR1 Gen 1 platforms for six newly unveiled Galaxy products.

​Cristiano Amon, President and Chief Executive Officer of Qualcomm Incorporated, who appeared as a speaker at the Unpacked event, said users would "see Snapdragon across more Samsung product categories" and added that the two companies would work together to create AI experiences with a deeper understanding of user context. Qualcomm and Samsung have partnered for three decades.

​Qualcomm said its Snapdragon platforms power AI experiences across Samsung's mobile portfolio through on-device AI processing and high levels of cross-device connectivity. The platforms are designed to enable autonomous AI experiences that adapt in real time to users' daily activities and workflows.

​The Snapdragon 8 Elite Gen 5 will power the Samsung Galaxy Z Fold8 Ultra, Galaxy Z Fold8, and Galaxy Z Flip8. The Galaxy Z Fold8 Ultra and Galaxy Z Flip8 newly adopt Snapdragon processors. The Fold8 Ultra is Samsung's first Ultra-branded foldable model. The previous-generation Flip model used Samsung's Exynos 2500 application processor (AP), although some regional variants of the Galaxy Z Flip8 will feature the Exynos 2600.

​The Snapdragon 8 Elite Gen 5 integrates Qualcomm's third-generation custom Oryon CPU architecture, an Adreno GPU, and a Hexagon NPU. Qualcomm said the platform delivers higher performance than its predecessor while reducing power consumption. It also enhances AI processing through improved smart camera capabilities and network connectivity.

The Snapdragon Wear Elite powers the Samsung Galaxy Watch9 and Galaxy Watch Ultra2. Both devices used Samsung's Exynos W1000 wearable AP in the previous generation. Snapdragon Wear Elite was first introduced at Mobile World Congress (MWC) in March. It is the first wearable processor to incorporate a dedicated NPU.

​The integrated Qualcomm Hexagon NPU enables on-device execution of AI models with up to one billion parameters without relying on cloud connectivity. The platform also features high-performance computing capabilities, advanced connectivity, and sensor fusion technology, which combines data from multiple sensors to improve accuracy.

​The Snapdragon AR1 Gen 1 will be used in Samsung Electronics' new intelligent eyewear, scheduled for release this fall. Samsung also unveiled two additional intelligent eyewear designs at the Unpacked event, developed in collaboration with Gentle Monster and Warby Parker.

​With the Snapdragon AR1 Gen 1, Qualcomm is extending its collaboration with Samsung Electronics and Google beyond Android XR headsets into the smart glasses segment. Android XR is Google's operating system for extended reality (XR) devices, including headsets and smart glasses.⁶

Nvidia has begun shipping its next-generation Vera Rubin AI platform to customers, marking the commercial rollout of the company's latest rack-scale computing architecture for AI factories.

The company announced on July 21 that Vera Rubin has entered volume production and is already running in customer data centers. Nvidia had previously announced Vera Rubin's entry into mass production on June 1. Customers include Google, Microsoft, Oracle and CoreWeave.

Vera Rubin is a rack-scale AI infrastructure platform that tightly integrates 36 Nvidia Vera CPUs, 72 Rubin GPUs, NVLink 6 interconnects and storage into a single system. The compute tray eliminates internal cables, fans and hoses, reducing assembly time from several hours to about one minute.

The Vera CPU is Nvidia's first server processor designed in-house specifically for AI infrastructure. It features 88 custom Olympus CPU cores and is designed to deliver up to 1.8x the performance of conventional x86 processors while providing three times greater inter-core bandwidth and 40% lower memory latency.

Nvidia said Vera Rubin significantly reduces both power consumption and AI inference costs compared with the previous generation. Benchmark testing on CoreWeave's cloud platform showed the system delivers 10 times more AI throughput per megawatt than Grace Blackwell NVL72, substantially lowering the cost per generated token. The company said tokens per megawatt has become a key metric for AI infrastructure because it measures how much AI work can be performed within a fixed power budget.

The platform also supports liquid cooling with coolant inlet temperatures of up to 45 degrees Celsius, enabling chiller-free dry cooling in suitable environments. Nvidia said the design can save millions of gallons of water annually per megawatt of deployed computing capacity.

SK hynix is accelerating investment in its P&T7 advanced packaging facility in Cheongju by bringing forward the project's execution schedule. The company has increased the amount approved by its board of directors to support an earlier cleanroom rollout, although the total planned investment for the project remains unchanged.

SK hynix announced on July 22 that its board approved an approximately 7.09 trillion won facility investment for construction of the P&T7 plant. The investment represents 5.88% of the company's equity. In its regulatory filing, SK hynix said the investment is intended to "secure production capacity in Cheongju to meet growing global demand for AI memory." The investment period extends through December 31, 2032.

The revised approval reflects the company's decision to accelerate the opening of cleanroom space in response to rising demand. Rather than increasing the overall project budget, SK hynix is adjusting the timing of capital expenditures under its existing investment plan. "The total investment for P&T7 remains unchanged at approximately 19 trillion won," the company said. "As the investment schedule has been brought forward, the amount requiring board approval increased, making a new board resolution necessary."

P&T7 will be SK hynix's seventh packaging and test facility and will manufacture advanced packaging products for AI memory, including high-bandwidth memory (HBM). Construction of the plant began in April. The company plans to begin installing the wafer test (WT) line in October 2027, followed by the phased construction of wafer-level packaging (WLP) lines through February 2028.

AMD will begin supplying its Helios AI systems to Microsoft in the second half of this year, the company announced on July 20.

Microsoft will deploy Helios in its Azure cloud platform to support customer AI models and inference workloads. The rollout expands the partnership between the two companies following Microsoft's adoption of AMD's MI300X graphics processing units (GPUs) in 2023.

Microsoft will also introduce two new virtual machine (VM) series, Azure HDv2 and Azure HXv2. The companies jointly designed the HDv2 VM platform, which is optimized for agentic AI workloads, while the HXv2 series targets electronic design automation (EDA) applications.

Beyond Microsoft, AMD has secured Meta, OpenAI, and Oracle as Helios customers.

Meta announced in February that it plans to deploy AMD GPUs across 6 gigawatts (GW) of AI infrastructure and is expected to utilize approximately 1GW of Helios capacity this year. OpenAI and Oracle also plan to deploy Helios beginning this year.

Helios is AMD's rack-scale AI system that integrates GPUs, central processing units (CPUs), networking equipment and cooling infrastructure into a single platform. AMD first unveiled plans earlier this year to launch the system in the second half of 2026.

Each Helios rack is equipped with 72 AMD Instinct MI455X GPUs and 18 sixth-generation EPYC "Venice" CPUs. Built on an open rack architecture, the platform is compatible with third-party GPUs and CPUs.

The system also incorporates AMD's Pensando data processing unit (DPU). AMD strengthened its networking portfolio through the acquisition of Pensando Systems in 2022.

The MI455X is AMD's first GPU to feature HBM4 (High Bandwidth Memory 4). In March, AMD signed a strategic memorandum of understanding with Samsung Electronics to use Samsung's HBM4 memory in the MI455X.

Samsung's HBM4 delivers data transfer speeds of up to 13 Gbps and memory bandwidth of 3.3 TB/s.

Samsung Electronics is preparing to build a production-scale die-to-wafer (D2W) hybrid bonding line equipped with approximately 50 bonders at its Pyeongtaek semiconductor campus to support next-generation high-bandwidth memory (HBM) and advanced logic chips. Equipment installation is expected to begin toward the end of this year.

According to industry sources on July 21, Samsung has selected Dutch advanced packaging equipment maker BE Semiconductor Industries (BESI) as its preferred supplier and has begun the procurement process for around 50 D2W hybrid bonding systems. The line is expected to be installed in the P5 fab at the Pyeongtaek campus.

Richard Blickman, chief executive officer of BESI, previously said during the company's first-quarter earnings call in April 2026 that "more definitive developments" regarding Samsung's adoption of its hybrid bonders were expected around the middle of this year. At the time, BESI's hybrid bonding equipment was undergoing Samsung's qualification process.

The order, however, has yet to be finalized. Samsung has requested design modifications to the equipment, delaying agreement on delivery schedules. Such modifications involve changes to system architecture, components or modules to meet customer-specific requirements.

Pricing has also complicated negotiations. Industry sources estimate that BESI's hybrid bonders cost roughly 6 billion won per unit—about twice the price of competing systems.

"BESI supplies hybrid bonders to companies such as TSMC and Micron, and it is reluctant to redesign its equipment exclusively for Samsung," one industry source said. "Samsung still considers BESI's technology the strongest option, but as negotiations become more complicated, it is also evaluating domestic suppliers that can respond more quickly to its requirements."

Samsung's semiconductor equipment subsidiary Semes has also been asked to provide D2W hybrid bonders for the production line and is hoping to supply equipment for two hybrid bonding production lines at the Pyeongtaek campus.

"Semes initially felt significant pressure after Samsung prioritized BESI, but sentiment has shifted recently," another industry source said.

Semes completed development of its D2W hybrid bonder earlier this year and delivered evaluation units to Samsung for performance verification. Industry sources believe the equipment has successfully passed internal testing.

Samsung is also considering Hanwha Semitech as a potential supplier. The company completed development of its second-generation D2W hybrid bonder, SHB2 Nano, in February and delivered an evaluation system to SK hynix in April 2026.

The evaluation platform was supplied as an integrated cluster that combined Hanwha Semitech's bonder and plasma activation module with an EFEM (Equipment Front End Module) from CyMechs and a wafer cleaning module from Zeus.

Samsung's investment in a D2W hybrid bonding production line at Pyeongtaek is closely tied to advanced packaging for HBM and logic semiconductors.

In addition to its existing advanced packaging operations in Cheonan, Samsung is developing next-generation hybrid bonding capabilities at Pyeongtaek for future HBM products.

Replacing conventional thermo-compression (TC) bonding with hybrid copper bonding (HCB) between HBM DRAM dies can significantly improve performance, reduce power consumption and lower heat generation. The technology directly bonds copper and dielectric materials without using microbumps, shortening interconnect distances between chips.

The technology is expected to play a key role in custom HBM (cHBM). Samsung plans to replace the conventional HBM base die with a customer-specific logic die incorporating proprietary compute intellectual property (IP). The logic die will complement external CPUs and GPUs while optimizing data transfer, with HBM DRAM stacks vertically integrated above it in a 3D system-in-package (SiP) architecture.

In June 2026, Samsung Foundry introduced 3D Cube-H, its next-generation hybrid bonding-based 3D vertical stacking solution. The heterogeneous integration technology targets AI chips and high-performance computing (HPC) systems, offering improvements in performance, power efficiency and bandwidth over existing packaging technologies. Samsung Foundry is actively promoting the technology to customers.

Hybrid bonding equipment is expected to begin arriving at Samsung's Pyeongtaek campus later this year. Internally, however, Samsung is targeting 2030 for large-scale commercial production using hybrid bonding, reflecting the technology's greater manufacturing complexity compared with conventional TC bonding.

One industry source said widespread adoption of hybrid bonding is likely to coincide with the introduction of NVIDIA's next-generation Feynman AI accelerator, expected around 2029.

AMD on July 20 announced the launch of the Ryzen 7 7700X3D, a new processor designed to enhance gaming performance using the company's proprietary 3D V-Cache technology. The processor is priced at $329.

The Ryzen 7 7700X3D is an enhanced version of the Ryzen 7 7700X aimed at the high-end PC market. Its primary distinction from the existing model is the addition of AMD's 3D V-Cache technology. According to the company's internal testing, the new processor delivers up to 9% higher gaming performance than the Ryzen 7 7700X. It is compatible with AMD's AM5 platform.

3D V-Cache is a packaging technology that vertically stacks additional cache memory on top of the processor cores. In this configuration, the stacked cache die sits between the cores and the cooler, which limits heat dissipation — a key reason AMD caps clock speeds on first-generation 3D V-Cache chips such as the 7700X3D. AMD currently employs 3D V-Cache technology across its Ryzen and EPYC processor families.

The Ryzen 7 7700X3D supports DDR5 memory with capacities of up to 128GB. It features a total cache capacity of 104MB, a 120W thermal design power (TDP), and a maximum boost clock of 4.5GHz. The processor is equipped with eight cores and 16 threads and is designed to operate at temperatures of up to 89 degrees Celsius.