The Greensheet: December 2025

Key Themes

• Memory Shortage Horizon Extends to 2027: Leading OEMs and enterprise customers now project the DRAM/NAND supply crunch will persist well into 2028, not 2026 as previously expected. Samsung, SK Hynix, and Kioxia have reportedly already sold out all 2026 NAND capacity and are negotiating 2027 allocations with hyperscalers.
• Nexperia Partially Resumes but Trouble Far from Over: While select Nexperia distributors in Xiamen and Dongguan have resumed quoting and shipping (domestically and to Hong Kong), global OEMs are permanently exiting the brand. Automotive Tier 1s now demand full traceability and are qualifying On Semi, Diodes, and Infineon as drop-in replacements at scale.
• Broadcom Faces Looming Supply Shock from TSMC Prioritization Shift: TSMC is reportedly cutting 50% of its Broadcom wafer allocation to redirect capacity toward memory and AI chips. Broadcom distributors warn of significant price hikes and extended lead times across networking and storage ICs starting Q1 2026.

Product Updates

Integrated Circuits

Foundry Shifts Constrain Legacy IC Supply

• TSMC is phasing out older, smaller 6-inch wafers and consolidating 8-inch wafer production to improve efficiency and focus resources on advanced 12-inch processes (like 3nm and 5nm) and advanced packaging, which are used by Broadcom, Nvidia, and Apple.
• Lead times for Marvell components, particularly the 88E-xxx series, have been significantly extended beyond the standard 26 weeks. Most existing backlog orders are now on hold, awaiting updated delivery. This delay stems from TSMC’s strategic reallocation of wafer capacity toward AI-related memory production, resulting in a ~50% reduction in wafer allocation for Marvell.
• GigaDevice’s flash memory is in short supply, with lead times stretching past 2026.
• The Xilinx XC7 series has recently been extended from 14 to 23 weeks due to raw material supply and costs.

CPUs 

Demand Clusters Around Proven Architectures

• The market is overwhelmingly favoring Intel’s 12th, 13th, and 14th Gen CPUs (Alder Lake and Raptor Lake) over newer architectures. Adoption of Meteor Lake, Lunar Lake, and Arrow Lake remains limited due to significant performance and design trade-offs. As a result, demand for older-generation Intel CPUs remains intense, exacerbating supply shortages and driving spot-market premiums.
• Intel is no longer providing full allocation for EOL Rocket Lake CPUs, accelerating end-of-life logistics for legacy server platforms.  
• AMD server CPU continues facing a supply issue on the Genoa and Turin series. Especially for 9554, 9654, 9334. Supply is limited with booking LT around 8-12 weeks.
• For non-server segments, AMD desktop CPU lead times have stabilized at 3–4 weeks, offering a rare bright spot in an otherwise constrained market.  

GPUs

Workstation GPU Availability Pushes Into 2026

• NVIDIA RTX 2000 and 4000 Ada workstation GPUs have had allocation pushed from November 2025 to January 2026, with lead times now exceeding 8 weeks.  
• NVIDIA RTX 5090 lead time stretches to February 2026 due to massive demand for GPUs for AI development and potential shortages with GDDR Memory.

Memory

Consumer Memory Exits Reshape Supply

• Micron has officially announced it will discontinue its consumer-focused Crucial memory and SSD product lines. Shipments will cease by the end of Q2 FY2026, after which the business line will be formally shut down.

RDIMM

Hyperscaler pull continues to crowd out channels

• DDR4 RDIMMs and DDR5 RDIMMs are both heavily constrained, and market pricing is moving up each week. More than 75% of memory production is now dedicated to hyperscalers and AI/cloud companies, leaving less than 25% available for traditional distribution channels. 

SSD

Pricing Resets as NAND Tightens

• Samsung and Solidigm Enterprise and Consumer SSD shortage is getting worse in Q1 2026. With reports of at least a 20% price increase for enterprise SSDs.
• Most storage specialists confirm a sweeping price adjustment is coming in January, with increases ranging from 20% to 50% depending on the series. With many customers reporting receiving about 1/3 allocation.
• Samsung is discontinuing all 250GB and 500GB SSD models moving forward. The bare minimum will now be a 1TB Capacity.
• Transcend has increased the SSD price by 20-25% and postponed its shipments to customers due to a lack of NAND. 

HDD

High-capacity models remain structurally constrained

• Lead times now exceed 52 weeks across all vendors, including Seagate, Western Digital, and even Toshiba, with the highest demand for models featuring 20TB and 24TB capacities.

Passives

Extended lead times and price increases persist

• Kemet tantalum capacitor lead times have stretched from 32 to 42 weeks.  
• Panasonic and Kemet have both increased prices and are expecting to increase pricing even more in 2026
  

Supply Chain Trends 

• TSMC expands CoWoS capacity with Nvidia: Semiconductor equipment suppliers reveal that TSMC and non-TSMC players, including ASE, Amkor, and UMC, are accelerating the expansion of advanced packaging Chip on Wafer on Substrate (CoWoS) capacity.
• NVIDIA's H200 chips require a circuitous route before landing in China: The Trump administration has authorized exports of NVIDIA's H200 artificial intelligence processors to China.

Manufacturer News and Updates 

• Wingtech, the Chinese parent company of Netherlands-based chipmaker Nexperia, said it has invited the court-appointed custodians of the Dutch firm for talks on control of the company, a possible first step toward easing relations. [Source: CNA]
• SK Hynix is reaping the largest benefits from the H200 export approval. It is reportedly the primary supplier of the fifth-generation high-bandwidth memory (HBM3E) used in the H200, meaning its supply volume will inevitably rise, making it the biggest beneficiary. [Source: Trendforce]

Explore our latest Market Intelligence on Agentic AI

A focused look at how Agentic AI workloads are reshaping demand for memory, GPUs, advanced packaging, and infrastructure—and what sourcing teams should be watching next. Read now.