AI drives Marvell chip price hikes, NVIDIA GB200 orders surge

1.Marvell chip price increase, effective early next year

According to fast technology news, due to the surge in artificial intelligence demand, the U.S. netcom and optical communications chip maker Marvell recently issued a notice announcing that the entire product line will be effective January 1, 2025 price increases, in the field of optical communications in the wave of price increases to take the lead in the action.

Marvell benefited from the high demand for ASIC, silicon optical and other cloud data center-related AI, last quarter's financial results and financial measurements exceeded market expectations, the price increase also means that the company's performance is expected to continue to grow. Among them, 800G PAM, 400ZR data center interconnect (DCI) and other optical communication products have become an important growth engine, driving data center revenue by more than 90% annually.

2. Microsoft GB200 chip order surge, NVIDIA capacity expansion

According to IT Home News, Tianfeng International Securities analyst Ming-Chi Kuo released industry chain order information for NVIDIA's Blackwell GB200 chip, showing that Microsoft is currently the world's largest GB200 customer, with orders surging 3-4 times in the fourth quarter of this year, and orders exceeding the sum of all other cloud service providers.

Ming-Chi Kuo said in the report that the capacity expansion of the Blackwell chip is expected to start at the beginning of the fourth quarter of this year, when shipments in the fourth quarter will be between 150,000 and 200,000 units, and shipments in the first quarter of 2025 are expected to grow significantly by 200% to 250% to reach 500,000 to 550,000 units.

Microsoft's Blackwell GB200 orders in Q4 have surged from 300 to 500 cabinets (primarily NVL36) previously to approximately 1,400 to 1,500 cabinets, of which approximately 70 percent are NVL72, a maximum increase of 400 percent. Subsequent orders will mainly focus on NVL72.

3. Tibbo: NAND flash products welcome price cuts, mobile category most notable

According to FastTech, prices of user-side SSDs are expected to drop by 5%-10% in the fourth quarter due to the combined effects of increased production and weakening demand. Although memory makers have more and larger capacity NAND chips, the AI PC market is not as hot as expected and market demand is weak, resulting in NAND flash product prices facing a downward adjustment.

Consumer SSDs will see a small price increase in the third quarter, however, as price competition intensifies in the fourth quarter, SSD pricing will decline. Enterprise SSDs may experience a 0-5% price increase. eMMC/UFS, which is typically used in mobile devices, may see the most significant price reductions in the fourth quarter, expected to be up to 13%. This trend is largely due to slowing growth in the smartphone market, where device manufacturers are cautious about depleting inventory and resisting price increases.

4.SK Hynix rumored to focus on HBM and AI memory, reduce CIS investment

SK Hynix is cutting back on its less commercial image sensor and foundry business to intensify its focus on high-margin HBM and AI memory, according to sources in Korean media cited by TechCrunch Daily.

SK Hynix is cutting its CMOS image sensor (CIS) R&D investment, and production capacity is expected to be reduced by more than half from last year to less than 7,000 12-inch wafers per month. In addition, personnel in the SoC design department, which is responsible for designing storage controllers, etc., have been transferred and reassigned to the HBM department.

5.Blackwell delayed shipments, rumors of NVIDIA and TSMC blame each other

According to reports cited by Fast Tech, the partnership between NVIDIA and TSMC is under increasing pressure after recent delays in the delivery of NVIDIA's Blackwell chips, which appeared to be due to production issues. According to the report, engineers found that the Blackwell chips would fail to operate in the high-pressure environments common in data centers, which some NVIDIA engineers believe may stem in part from Blackwell design flaws.

Some NVIDIA engineers believe that the slowdown in production of Blackwell chips stems from TSMC's adoption of a new technology that packages different types of chips together. And some TSMC employees said that NVIDIA let TSMC rush to complete the production process, compared to another major customer Apple, NVIDIA left TSMC to solve the problem less time.

6.NXP launched S32J series of automotive Ethernet switch chip

NXP (NXP) recently launched the S32J series of high-performance Ethernet switches and network controllers, the product and NXP's latest S32 microcontrollers and processors share a common switch core NXP NETC, to be able to operate as an extended virtual switch together.

The S32J offers 80Gbps bandwidth with ports ranging from 10Mb to 10Gb and dual Arm Cortex-R52 cores to meet the varying requirements of new vehicle architectures.The S32J devices are TSN automotive-compliant and offer robust ASIL-D security, Hardware Security Engines (HSEs), and MACsec ports for mixed critical data traffic.

The combination of the S32J family with the NXP CoreRide platform provides a production-grade networking solution with pre-integrated software and tools.A virtual development kit for the S32J family will be available in late 2024. The solution will be available to OEMs and Tier 1 suppliers in 2025.

7. Renesas Releases Two IO-Link Chips to Power Industrial Automation

Renesas recently released two IO-Link protocol semiconductor solutions, including the CCE4511 four-channel IO-Link master chip and the ZSSC3286-IO dual-channel resistive sensor signal conditioner chip.

The CCE4511 is the industry's first four-channel master chip for the IO-Link protocol, with a drive current of 500mA per channel, offering many advantages over dual-channel master devices. Energy savings, reduced PCB size, and fewer external components are required compared to competitive products. the CCE4511 has an ambient operating temperature of up to 125oC, making it suitable for use in harsh industrial environments.

The ZSSC3286 is a dual-path sensor signal conditioning chip for high-precision amplification, digitization, and sensor-specific correction of sensor signals. It supports IO-Link connectivity, and the integrated IO-Link stack runs the IO-Link Smart Sensor Profile for digital measurements and switching of sensors. the ZSSC3286 replaces the MCU, saving board space and cost, and also provides two 24-bit parallel ADCs, a measurement scheduler for optimized performance, and advanced system diagnostics.