TL;DR
A Thorsten Meyer AI report says high bandwidth memory has become the main pressure point in the 2026 memory crunch, with HBM sold out through the year and taking wafer capacity from DDR5 and GDDR7. The report says AI accelerator demand, led by Nvidia-linked supply chains, has made HBM a pricing and availability driver for much of the memory market.
High bandwidth memory has become the central strain point in the 2026 memory crunch, according to a late-June report from Thorsten Meyer AI, which says AI chip demand is pulling wafer capacity away from ordinary RAM and graphics-card memory.
The report describes HBM as stacked memory used beside AI GPUs, with eight to 16 DRAM dies connected by through-silicon vias and mounted close to the accelerator. That design gives AI chips far more bandwidth than standard graphics memory, with the report placing the gain at about five to 10 times normal GDDR memory.
The supply problem, according to the report, is manufacturing efficiency. It says one bit of HBM consumes about three to four times the wafer area of one bit of DDR5, while defects can spoil an entire stacked package. That makes each HBM order a larger claim on fab capacity than a comparable amount of ordinary DRAM.
The report says the economics are pushing suppliers toward HBM. It cites estimated stack pricing of about $200 for HBM3, $300 for HBM3E and roughly $500 for HBM4. It also says Samsung and SK Hynix raised HBM3E prices by about 20% for 2026 while demand still exceeded supply.
HBM ate the fab
The thing the factories make instead of your RAM is a tower of stacked memory bolted to every AI chip. In three years it went from niche part to the component that sets the price of nearly all the world’s memory — and now a chunk of its GPUs.
A tower, not a sheet
HBM stacks DRAM dies vertically, links them with thousands of through-silicon vias, and sits beside the GPU to deliver 5–10× the bandwidth of normal graphics memory. AI is bandwidth-bound — without it, the world’s most expensive silicon sits starved for data. But stacking is inefficient: one HBM bit eats 3–4× the wafer area of DDR5, and one defect can ruin a whole tower.
≈ 8 HBM stacks wrap every AI GPUThis isn’t artificial scarcity — AI really is bandwidth-bound, HBM really is the fix, and it really does eat 3–4× its weight in fab capacity. The discomfort is structural: one component, coupled to one customer’s demand, now sets the price of nearly all memory and a slice of GPUs. The market is now $35B → ~$100B by 2028, ~41% of all DRAM revenue (was 8% in 2023), and sold out through 2026. The one hope: with all three suppliers finally racing on HBM4, competition can add supply. The matching risk: if AI demand corrects, HBM is where it breaks first. Next: DDR5 now, DDR6 soon.
AI Memory Is Repricing DRAM
The finding matters because HBM demand is no longer isolated to high-end AI servers. The report says the same factory decisions are affecting DDR5 availability, GDDR7 supply and parts of the consumer graphics-card market.
Thorsten Meyer AI says suppliers prioritizing HBM contributed to a shortage of GDDR7 memory used in consumer GPUs. The report cites claims that Nvidia RTX 50-series production was cut by at least a third in the first half of 2026, though it presents that as reported information rather than a confirmed company disclosure.
The broader consequence is that one fast-growing component can influence prices across memory categories. The report estimates the HBM market could rise from about $35 billion to nearly $100 billion by 2028, reaching about 41% of DRAM revenue after accounting for 8% in 2023.
High Bandwidth Memory (HBM) modules for AI GPUs
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
HBM4 Race Reaches All Suppliers
The report frames the market as a three-company race among SK Hynix, Samsung and Micron. It says SK Hynix leads with roughly 50% to 62% share and sends about 90% of its HBM to Nvidia.
Samsung is described as a 2026 comeback contender with about 28% to 40% share and qualification for Nvidia Rubin HBM4. Micron is described as smaller, at about 5% to 10% share, but sold out for 2026 and positioned for HBM4 inference chips.
Generational changes add pressure. The report lists HBM3 at about 819 GB/s per stack, HBM3E above 1.18 TB/s and HBM4 near 2.8 TB/s, with a new base logic die tied to the coming Rubin generation.
“The thing the factories make instead of your RAM is a tower of stacked memory bolted to every AI chip.”
— Thorsten Meyer AI report
HBM3 and HBM4 memory chips
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Supply Gains Remain Unproven
Several details remain unsettled. The report says per-stack pricing is estimated and point-in-time, and it labels the market fast-moving as of late June 2026.
It is also not yet clear how much HBM4 qualification by all three major suppliers will increase usable supply, or whether yields will improve fast enough to ease pressure on DDR5 and GDDR7. Claims about RTX 50-series production cuts are described as reported, not as a confirmed Nvidia statement in the provided material.
ASUS ROG Strix GeForce RTX 4090 OC Edition Gaming Graphics Card (PCIe 4.0, 24GB GDDR6X, HDMI 2.1a, DisplayPort 1.4a), 3 Year Warranty
NVIDIA Ada Lovelace Streaming Multiprocessors: Up to 2x performance and power efficiency
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
HBM4 Output Sets The Test
The next market signal will be HBM4 volume production and how well SK Hynix, Samsung and Micron can raise yields while meeting demand from Nvidia, AMD and other AI chip buyers.
If supply expands, the report says competition could help ease the squeeze. If AI demand weakens, HBM could be the first part of the memory market to show stress because suppliers have shifted so much capacity toward it.
Stacked HBM memory for AI accelerators
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is HBM?
HBM is high bandwidth memory, a stacked DRAM package placed close to an AI GPU. It uses vertical silicon connections to move far more data than ordinary graphics memory.
Why does HBM affect regular RAM prices?
The report says HBM uses more wafer area per bit than DDR5 and is more profitable. When memory makers allocate wafers to HBM, there is less capacity for ordinary DRAM.
Which companies dominate HBM supply?
The report names SK Hynix, Samsung and Micron as the main suppliers, with SK Hynix described as the current leader.
Is the shortage confirmed through 2026?
According to Thorsten Meyer AI, HBM is sold out through 2026. Exact availability can still shift by supplier, product generation and customer allocation.
What could ease the memory squeeze?
The report points to HBM4 competition among all three suppliers as the main source of possible relief, provided production ramps and yields improve.
Source: Thorsten Meyer AI
