TL;DR
Thorsten Meyer AI’s late-June report says High Bandwidth Memory has become a central driver of the 2026 memory crunch because AI accelerators need it and suppliers are shifting wafer capacity toward it. The report says HBM is sold out through 2026, but pricing, allocation and the pace of HBM4 supply remain partly uncertain.
High Bandwidth Memory has become a central driver of 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 DRAM and tightening supply for memory and some graphics cards.
The report identifies HBM as the product memory makers increasingly prefer because it sells into AI accelerators at far higher value than standard DDR5. HBM stacks multiple DRAM dies vertically, connects them with through-silicon vias, and places the stack close to the GPU so AI chips can receive data fast enough to avoid bottlenecks.
Thorsten Meyer AI says one HBM bit can consume roughly three to four times the wafer area of one DDR5 bit, while a defect in a stacked tower can ruin the finished component. The report cites estimated per-stack pricing of about $200 for HBM3, around $300 for HBM3E, and about $500 for HBM4, with supplier pricing and availability still moving.
The report says SK Hynix, Samsung, and Micron are the three main suppliers, with SK Hynix leading the market and Micron described as sold out for 2026. It also says all three had qualified for HBM4 by June 2026, shifting the contest from basic qualification to volume, yield and customer allocation.
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 Demand Reshapes DRAM Supply
The significance for readers is that the shortage is not limited to data centers. If fabs allocate more wafers to HBM for AI chips, fewer wafers are available for DDR5 memory and other DRAM products used in PCs, servers and consumer hardware.
The report says the impact has also reached graphics cards. With suppliers prioritizing HBM, GDDR7 used in consumer GPUs has reportedly become tighter, and Thorsten Meyer AI cites reports that Nvidia RTX 50-series production was cut by a third or more in the first half of 2026. That figure is presented as reported, not independently confirmed in the source material.
High Bandwidth Memory HBM4
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Niche Part To Bottleneck
HBM was once a specialty memory product, but the report says it has moved from niche status to a component that influences broad memory pricing within about three years. The shift tracks the rise of AI accelerators such as Nvidia’s H100, H200, B200 and planned Rubin platform, as well as AMD’s MI300 series.
The generational pace is part of the pressure. The report lists HBM3 at about 819 GB/s per stack, HBM3E at about 1.18 TB/s, and HBM4 at an estimated 2.8 TB/s. Higher bandwidth helps AI workloads, but it also raises manufacturing demands and keeps supply tight when orders rise faster than capacity.
“The thing the factories make instead of your RAM is a tower of stacked memory bolted to every AI chip.”
— Thorsten Meyer AI report
GPU with HBM memory
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Supply Claims Still Need Watching
Several figures remain dependent on fast-moving market data. The report labels per-stack pricing as estimated or point-in-time, and market-share ranges for SK Hynix, Samsung and Micron vary by source.
It is also not yet clear how quickly HBM4 output can grow, how much relief it will bring to buyers, or whether AI demand will stay strong enough to absorb the added supply. The reported Nvidia production cuts and the extent of GDDR7 shortages also need continued confirmation from suppliers and public filings.
High bandwidth memory modules
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
HBM4 Output Becomes The Test
The next milestone is HBM4 volume production and allocation across AI-chip customers. If SK Hynix, Samsung and Micron can raise yields and ship more stacks, supply pressure could ease over time.
If AI demand cools, the report says HBM is likely to be the first pressure point because capacity, pricing and customer commitments have moved so heavily toward it. For now, the report describes the market as sold out through 2026.
AI accelerator memory modules
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the actual news development?
A late-June 2026 report from Thorsten Meyer AI says HBM demand has become a central factor behind the current memory squeeze, with effects reaching ordinary DRAM and some GPU supply.
What is confirmed versus claimed?
It is confirmed that HBM is used in major AI accelerators and that SK Hynix, Samsung and Micron compete in the market. Pricing, share ranges, 2026 sellout status and reported GPU production cuts are attributed claims from the cited report and its industry sources.
Why does HBM use more fab capacity?
HBM stacks DRAM dies vertically and requires larger dies, TSV connections and more complex packaging. The report says one HBM bit can use roughly three to four times the wafer area of one DDR5 bit.
How could this affect consumers?
If more wafers go to AI-focused HBM, supply of DDR5 and GDDR7 can tighten. That can affect prices and availability for PCs, servers and graphics cards.
What happens next in the memory market?
The key test is whether HBM4 supply grows fast enough to meet AI demand. If it does, pressure may ease; if demand stays ahead of output, shortages could continue through 2026.
Source: Thorsten Meyer AI
