Verified July–August 2026 reports on the evolving landscape of global hardware shortages and supply chain volatility.
IDC reports PC shipments fell 5% YoY in Q2 2026 to 68.2 million units, ending nine consecutive quarters of growth. The memory shortage is pricing budget PCs out of the market, and IDC warns it will not ease until early 2028.
SK Hynix CEO Kwak Noh-jung told Reuters that 2027 will be the worst supply year in the industry's history and that customer demand will remain higher than the company's supply capacity beyond 2030.
US lawmakers are targeting Chinese memory vendors YMTC and CXMT, which could remove a potential source of relief for the global DRAM/NAND shortage.
Nvidia proposed a financing program to help emerging AI cloud providers acquire GPUs. Early customers include Sharon AI (up to 40,000 GB300 GPUs in Australia) and Firmus (up to 170,000 Nvidia GPUs in Indonesia).
Nvidia demonstrated its Vera Rubin platform, claiming the NVL72 compute tray can be assembled in one minute versus 90 minutes for GB200, and early CoreWeave tests show 10x more tokens per watt than GB200.
ASE raised CoWoS and Fan-Out Chip on Substrate (FoCoS) packaging quotes by more than 20%, citing higher raw-material costs and long-term investment needs.
ASE increased its 2026 capital expenditure from US$8.5 billion to US$10.5 billion and raised its 2027 leading-edge advanced packaging revenue target to at least US$7.5 billion.
Samsung Electronics reported Q2 2026 revenue of W171.5 trillion ($119 billion) and operating profit of W89.5 trillion ($62.4 billion), up 19x YoY. Samsung's EVP of Memory said the supply shortage will persist through 2028.
TrendForce revised its 2026 AI server shipment forecast up from 28% to nearly 31% YoY. The top nine CSPs are expected to spend more than US$886.7 billion on CapEx in 2026 (+~90% YoY) and about US$1.3 trillion in 2027 (+~50% YoY).
TrendForce says DRAM supply will remain tight through 2027 and HBM4e validation schedules are uncertain. NVIDIA is evaluating lower HBM configurations for Rubin Ultra beyond the original 12-Hi HBM4e design. HBM bit shipments are projected to grow 50-60% YoY in 2027, still belo...
TSMC is expanding outsourcing of the CoW (Chip-on-Wafer) step to ASE, Amkor and SPIL to ease AI chip packaging bottlenecks. NVIDIA has reportedly reserved 800-850 k wafers of TSMC CoWoS capacity for 2026.
TrendForce expects the DRAM sufficiency ratio to be approximately -1% to -2% in 2026 and the gap to widen in 2027 as demand outpaces supply. NAND Flash is expected to enter a looser supply environment in 2H27.
Both Seagate and Western Digital have fully allocated nearline production for 2026, with orders now stretching into 2028. Industry research puts lead times at 52–104 weeks and average nearline price gains at 46%, driven by AI storage demand absorbed through multi-year contracts. [ledger: STORAGE-NARLINE-SOLDOUT-LEADTIMES-2027-2028 · NRL-HDD-CLOUD-ALLOCATION-PRICING]
Seagate's Mozaic 4+ (44 TB) has shipped in volume to two leading hyperscalers since March 2026, with 50% of HAMR exabytes planned on the platform by year-end. The published roadmap: 50 TB customer validation in 2027 on Mozaic 4, then 100 TB on the 5 TB/platter generation in 2028. [ledger: STORAGE-CAPACITY-PATHWAY-MOZAIC-50TB-100TB-2028 · STORAGE-MOZAIC4PLUS-ASP-HYPERSCALE-2026]
Fiscal Q4 2026 revenue reached $3.63B (+48.5% YoY) with gross margin above 52%, up from the low 40s a year earlier. The CFO confirmed the 5 TB/platter milestone; HAMR platforms now carry roughly 40% of nearline exabytes shipped. [ledger: STORAGE-HAMR-MARGIN-DIVORCE-SEAGATE-FQ4-2026 · NRL-STX-HAMR-MOZAIC-LEADERSHIP]
Western Digital started shipping its 40 TB ePMR/UltraSMR nearline platform and expects it to exceed 50% of nearline exabytes by Q3 2027. The dual-track roadmap runs an ePMR line toward 60 TB by 2028 and a HAMR line toward 100 TB by 2029. [ledger: STORAGE-WDC-ULTRASMR-EPMR-HAMR-ROADMAP-2027 · NRL-WDC-EPMR-ULTRASM-RAMP-TWO-LINE]
Toshiba's M12 Series uses an 11-disk FC-MAMR shingled design rather than Seagate's HAMR laser route, with CMR models topping out at 28 TB. With HAMR at ~40% of Seagate's nearline exabytes, the density gap between the camps is now roughly one full product generation. [ledger: NRL-TOSHIBA-FC-MAMR-LAG-BEHIND (official Toshiba release is 403 to automated fetchers)]
The enterprise SSD/HDD cost ratio has widened to ~16× from 6× twelve months ago as NAND makers convert capacity toward DRAM/HBM margins. Phison's CEO confirms every NAND manufacturer is sold out for 2026; enterprise SSDs now run $200–$300/TB against $100–$150/TB high-capacity nearline. [ledger: STORAGE-SSD-HDD-COST-RATIO-BIFURCATION-2026 · MEMORY-BW03-NAND-REALLOCATION-SSD-IMPACT-SEP26]
For the first time all three DRAM majors supply the same next-gen HBM tier: SK Hynix holds ~70% of Vera Rubin HBM4 allocation, Samsung targets 25–30%, and Micron qualified alongside them in June 2026. SK Hynix's base-die respin remains the schedule risk for Rubin. [ledger: COMPUTE-HBM4-ALLOCATION-SKHYNIX-70-MICRON-2026 · MEMORY-BW03-YIELD-SAMSUNG-VS-SKHYNIX-SEP26]
Figure AI committed $3.5B initially (scaling beyond $6B) for up to 100,000 Rubin GPUs deployed via Nscale, targeting deployment from H2 2027 in Barstow, Texas. One order of that size equals an estimated 70–90% of annual Vera Rubin supply — demand-side concentration on top of supply-side scarcity. [ledger: COMPUTE-FIGURE-NSCALE-100K-GPU-ORDER-2026]