Samsung Electronics announced its earnings guidance for the second quarter of 2026 on July 7, estimating consolidated sales of approximately KRW 171 trillion and consolidated operating profit of approximately KRW 89.4 trillion. Reuters also reported that AI demand continues to drive higher DRAM and NAND prices, helping memory suppliers benefit from stronger data center spending and a broader recovery in the memory cycle.

The signal is important because AI demand is no longer limited to the most advanced HBM products. As AI training, inference and data-intensive workloads expand, demand is spreading into server DRAM, NAND, high-capacity storage and enterprise-grade memory solutions.

On July 8, Samsung announced mass production of PM1763, its PCIe 6.0-based enterprise SSD optimized for next-generation AI and HPC server environments. The product uses Samsung’s 9th-generation V-NAND and a 4nm controller, and is designed to support demanding AI workloads where data must be accessed quickly, reliably and with improved power efficiency.

Samsung also highlighted that the rapid growth of AI training and inference is increasing the importance of enterprise SSDs in AI infrastructure. This reflects a broader industry trend: as GPU clusters scale, storage throughput, latency and energy efficiency are becoming critical performance factors.

Reuters reported on July 7 that DeepSeek is developing its own AI chip, with the project mainly targeting inference workloads. The report said the initiative is still at an early stage, while the company has been hiring chip design engineers and exploring cooperation across chip design, foundry and memory-related supply chains.

The move reflects a broader trend in the AI sector. As model providers seek better cost control, more stable access to compute and differentiated performance for specific workloads, custom silicon is becoming an increasingly important strategic option.

Reuters reported on July 9 that Changxin Memory Technologies, China’s leading DRAM manufacturer, plans to begin book-building on July 15 for its Shanghai IPO. The company is aiming to raise about RMB 29.5 billion, or approximately USD 4.34 billion, with proceeds expected to support production line upgrades and technology development.

The IPO plan comes at a time when memory remains one of the most strategically important areas in the semiconductor supply chain. Capital investment, equipment access, yield improvement and process migration will all influence how quickly additional DRAM capacity can translate into stable market supply.

Reuters reported on July 8 that Apple will spend more than USD 30 billion on U.S.-made chips from Broadcom under a multi-year agreement. The deal covers radio-frequency components such as FBAR filters used for wireless connectivity, and Broadcom is expected to invest in expanding its Fort Collins, Colorado facility.

This agreement shows that supply chain localization is not limited to processors or advanced logic chips. RF components, filters and connectivity-related devices are also being pulled into longer-term sourcing agreements as major OEMs strengthen supply security.

As AI data centers move toward higher power density, server power architectures are becoming more complex. Demand is rising across PMICs, MOSFETs, IGBTs, SiC/GaN devices, power modules and gate drivers. TrendForce previously reported that TI was expected to raise prices for selected segments including PMICs, MOSFETs and industrial control chips from July 1, while Infineon-related pricing pressure also reflected rising supply chain costs and strong demand.

These components are essential for power conversion, voltage regulation and system reliability. Any pressure in this area may affect not only AI servers, but also industrial equipment, automotive electronics and high-reliability power systems.

The expansion of AI clusters is increasing the amount of data moving between GPUs, servers and racks. This is driving stronger demand for high-speed optical modules and related components, especially 800G and above. TrendForce has highlighted rapid growth in laser diode capacity, reflecting the need to support AI data center deployment.

Optical modules rely on a broader set of upstream components, including lasers, DSPs, TIAs, driver ICs, optical engines and precision assembly capacity. As cloud service providers continue to scale AI infrastructure, these smaller but critical devices may become more important sources of supply risk.

Reuters previously reported that Google had held talks with Chinese suppliers, including Envicool, to buy liquid-cooling systems for data centers. The report reflected rising demand for higher-efficiency thermal management as AI servers generate more heat and traditional air cooling becomes less suitable for high-density computing environments.

Liquid cooling is not a traditional semiconductor category, but it affects the wider AI server supply chain. CDU units, cold plates, pumps, valves, sensors, connectors, control boards and monitoring ICs may all benefit from growing liquid-cooling adoption, while also facing tighter availability as deployments accelerate.