The Greensheet: February 2026

Key Themes

• Memory Crisis Enters "Post-Settlement" Era: The DRAM/NAND crisis has moved beyond allocation shortages into a structural reset of commercial terms. Major manufacturers (Samsung, SK Hynix, Micron) are enforcing post‑settlement pricing clauses within shorter‑term, NCNR contracts, obligating customers to pay market‑rate price differences after delivery. This effectively removes price certainty, with final pricing only determined when products ship at a future date.
• Automotive Sector Facing Existential Memory & Discrete Crisis: Automotive Tier 1s (Toyota, Denso, Marelli, Bosch) are reporting critical gaps in eMMC, LPDDR4/5, and legacy DRAM, with manufacturers prioritizing AI server margins over automotive-grade reliability. Simultaneously, the Nexperia supply chain fragmentation is forcing emergency qualifications of On Semi and Vishay, though lead times for these alternatives have stretched to 40–52 weeks due to capacity overflow.
• Storage Supply Chain Collapse & Strategic EOLs: HDD capacity for 2026 is reported as 100% sold out by Seagate and WD, with lead times pushing into 2028 for high-capacity drives. Manufacturers are aggressively forcing migrations by EOL-ing low-capacity drives (1TB–4TB) and restricting SSD production in favor of higher-margin products. We are seeing severe bottlenecks for enterprise storage.
• GPU & Networking Bottlenecks Shift to Interconnects & Legacy Nodes: While Blackwell GPUs remain in high demand, the bottleneck has shifted to Mellanox networking cards (ConnectX-7/8) and Mellanox NICs, with lead times exceeding 50+ weeks. Finisar optical transceivers have also emerged as a critical constraint. Export controls on H200s to China continue to create volatile spot markets for compliant interconnects like the MCX75310AAS-NEAT.
• T-Glass Substrate Shortage Triggers Cross-Category Decommitments: A critical shortage of T-Glass substrate and glass fiber cloth is causing shortages across analog, FPGA, and CPU sectors. Analog Devices (ADI) has decommitted orders for the LT8 series, Intel cites it for 10–20% CPU price hikes, and Lattice Semiconductor cannot support NPI demand with lead times exceeding 44 weeks. The bottleneck is rooted in AI demand, with NVIDIA‑driven consumption of T‑Glass substrates crowding out supply and creating downstream shortages that are more restrictive than wafer capacity across multiple semiconductor categories.

Product Updates

Integrated Circuits

• Analog Devices is implementing widespread price increases of 15–30% across power and signal chain portfolios. Lead times have stretched to 24–40 weeks, with specific decommits reported for automotive customers during the CNY period.
• Texas Instruments' lead times for booking orders have extended to 6 months (180 days). TI is enforcing stricter quotation protocols, requiring detailed end-customer data and project plans before releasing pricing, effectively halting free-market purchasing. Price hikes of 10–30% are being enforced on legacy and power ICs.
• Nexperia's supply situation remains critical. Lead times for Malaysia/Philippines production have slipped to 30–44 weeks. Automotive customers are unable to secure allocation, with orders placed in late 2025 now pushed to late 2026. Expect pricing to increase by ~10-15%.
• Infineon & On Semi are seeing lead times double to 30–52 weeks as they absorb overflow demand from Nexperia. Infineon has announced price increases effective April 1st, while On Semi is facing wafer capacity constraints at its Le Shan, China facility.
• Lead times for Xilinx Spartan-7 and Lattice devices have exploded to 32–52 weeks (up from 8 weeks), driven by AI server demand consuming packaging and glass fiber raw materials.
• Sony projects CMOS Sensors IMX series shortages starting April 2026, allocating only 10% to certain distributors. A global shortage of glass substrate (Nittobo) limits production to 70% of demand through 2027.

CPUs 

 Structural Shortages and Foundry Shifts 

• Intel Server (Xeon) shortages persist for Sapphire Rapids and Emerald Rapids. Intel has confirmed price increases of 12–20% effective March 29, 2026. Lead times for embedded 10nm parts (Elkhart Lake, Raptor Lake-E) are extending into 2027–2028, with some allocations completely decommitted.
• AMD Server (EPYC) Genoa and Turin series remain sold out through 2026. Booking lead times are uncertain, with specific high-core count SKUs (9555P, 9654P) facing 8-month+ delays as AMD prioritizes hyperscaler AI projects.
• Intel is reducing rebates and increasing prices on desktop CPUs by ~10% starting late March. Small core (N-series) availability remains tight with frequent decommits.

GPUs

• NVIDIA Blackwell (RTX 6000 / 5090) server edition is seeing price hikes of 25%, with some offers reaching $8,300–$8,800. Consumer RTX 5090 supply is constrained by GDDR7 memory shortages.
• Nvidia RTX 4000/5000 Ada lead times remain extended (48–52 weeks), and prices are increasing by 5-10% with production being curtailed to focus on Blackwell.

Networking

• Mellanox ConnectX-7 lead times have extended to 20–50 weeks (up from 2–4 weeks), with ConnectX-8 facing similar constraints.
• Broadcom warns that current 52-week lead times are "just the start," with constraints tightening further in 2027 due to insufficient 3/5/7nm capacity.
• Coherent (Finisar) has pushed lead times for 400G/800G transceivers out by 6 months, with some materials scheduled only into 2028. Production in Malaysia and China is actively limiting new orders.

Memory

• A new "post-settlement" model is emerging where final pricing is adjusted based on market rates after delivery, shifting all risk to the buyer.
• DDR5 RDIMM 64GB contract prices have jumped 75–125% Q/Q. Spot prices range from $2,300 to $3,000+. Samsung is proposing 110% MoM increases for DIMMs to OEMSs. Hyperscalers are signing LTAs extending to 2030, effectively locking out traditional OEMs and automotive sectors until at least 2027.
• Shortages for DDR3/DDR4 will see an increase in prices, projecting the 2027 allocation to be worse than 2026.
• Severe shortages are impacting 8GB and 16GB automotive‑grade eMMC. Samsung and Micron have effectively stopped quoting, pushing lead times out to 2027, while SanDisk indicates customers should plan for just 15–20% of requested allocations in 2026. Legacy eMMC pricing has tripled as production capacity is reallocated away from legacy memory and standard DRAM toward higher‑margin HBM3E/HBM4 used in AI accelerators.
• High demand for Scandisk low-density eMMC, especially for SDINBDG4-8G-ZA2 and SDINBDA6-64G-ZA1. Lead times are now stretched to 26 weeks.

SSD

• Enterprise SSD prices have increased 50–90% Q/Q, with Solidigm and Samsung implementing weekly price resets.
• Manufacturers are shifting fab capacity from QLC to TLC due to stronger margins, creating shortages in high-density QLC drives used in hyperscale storage.

HDD

• Seagate and WD capacity is 100% sold out for 2026, with some agreements extending into 2028–2030.
• Seagate is planning to EOL 1TB–5TB 2.5" Barracuda drives, while WD is refusing quotes for low-capacity drives, forcing migrations to 12TB+.
• WD high-capacity HDD (18TB–24TB) lead times are expected to range from 3 months to 54 weeks.

Passives

• MLCC shortages are anticipated post-CNY, with lead times stretching to 16+ weeks. Tantalum capacitor lead times exceed 40 weeks, and vendors are warning of potential discontinuities for the low-margin line.
• T-Glass Substrate material constraint is now the primary bottleneck for ADI, Intel, Lattice, and Sony. Glass cloth fabric lead times have extended from 4 weeks to 20 weeks, causing decommits and forcing foundry diversions.
• Lattice lead times are currently very tight. Any orders requested with a lead time of less than 44 weeks will be subject to an expedited fee.
• Automotive customers are investigating potential wafer issues at Vishay, compounding risks as the company struggles to fill the void left by Nexperia diodes.

Supply Chain Trends 

• Meta locks in millions of Nvidia GPUs and CPUs in multiyear AI infrastructure pact: Nvidia and Meta have signed an expanded multiyear agreement under which Meta will deploy millions of Nvidia AI chips across its global data centers, securing long-term GPU, CPU, and networking supply for large-scale artificial intelligence infrastructure. The agreement covers Nvidia's Blackwell GPUs, next-generation Rubin GPUs, standalone Grace CPUs, future Vera CPUs, networking systems, and security technologies.
• Nvidia AI chip sales to China stalled by US security review: Nvidia’s sales of H200 AI chips to China are still awaiting final approval from Washington, nearly two months after Donald Trump greenlit exports, as the US government conducts a national security review before granting licences to Chinese customers.
• Western Digital is already sold out of hard drives for all of 2026: Western Digital CEO Irving Tan has confirmed that the company has “pretty much sold out for calendar 2026.” Western Digital went on to confirm they have firm purchase orders in place, locking in most of its supply throughout the year, with agreements already signed through to 2028 for one customer.

Manufacturer News and Updates 

• GlobalFoundries and Renesas Electronics Corporation announced an expanded strategic collaboration through a multi-billion-dollar manufacturing agreement aimed at strengthening semiconductor supply chains and expanding production capacity across multiple regions. [Source: GF]
• Onsemi's revenue declined in the fourth quarter as its two largest business units continued to post year-over-year sales declines, highlighting persistent weakness across key parts of its portfolio. [Source: Onsemi]
• Texas Instruments has announced an agreement to acquire Silicon Labs, a move expected to close in H1 2027. The transition aims to reshore manufacturing but will likely cause supply disruptions and integration challenges for customers of both firms over the next 24 months. [Source: TI]

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.