TL;DR
HBM has moved from a specialty memory part to the component shaping the 2026 memory crunch, according to a late-June Thorsten Meyer AI report. The source says AI chip demand is pulling wafer capacity toward HBM, pressuring DDR5 RAM and GDDR7 graphics memory while SK Hynix, Samsung and Micron prepare HBM4.
High Bandwidth Memory has become the capacity-setting part in the 2026 memory crunch, according to a late-June Thorsten Meyer AI report, as AI accelerator demand redirects DRAM fabs toward stacked memory and away from ordinary RAM and graphics memory. The shift matters because HBM now influences prices and availability across PCs, servers and some consumer GPUs.
The report identifies HBM as the part manufacturers are making instead of more conventional memory. Unlike flat DDR5 modules, HBM stacks eight, twelve or sixteen DRAM dies, connects them through through-silicon vias, and places the stack beside an AI GPU on an interposer to feed the chip data at much higher bandwidth.
That design is useful because AI accelerators are bandwidth-bound, the source says. HBM can deliver roughly 5-10 times the bandwidth of normal graphics memory, but it is much harder to build. The report says one HBM bit consumes about 3-4 times the wafer area of a DDR5 bit, and a single defect can damage the value of a whole stack.
The source describes a market in which HBM3 stacks cost about $200, HBM3E about $300, and HBM4 is estimated near $500 per stack. It also says Samsung and SK Hynix raised HBM3E prices by about 20% for 2026, while demand still exceeded supply. Those price figures are presented as estimates and point-in-time market readings.
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 Pressure Hits Buyers
The main consequence is that AI hardware demand is no longer isolated inside data centers. If HBM uses several times more wafer area than DDR5, then fab decisions made for Nvidia-class AI accelerators can affect the supply of PC memory, server DRAM and graphics memory used outside AI.
The report says the strain has also reached GPUs. With suppliers prioritizing HBM production, the GDDR7 memory used in consumer cards has reportedly tightened, and Nvidia reportedly cut RTX 50-series production by a third or more in the first half of 2026. That specific production-cut claim remains attributed to reporting cited by the source.
High Bandwidth Memory HBM4 modules
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How HBM Took Fab Space
HBM was once a specialty memory product, but the report says it became central as AI chips moved from Nvidia H100 to H200, B200 and planned Rubin-era parts. A modern AI GPU can use around eight HBM stacks, making memory supply a gating factor for accelerator shipments.
The generational race has raised both performance and manufacturing pressure. The source lists HBM3 near 819 GB/s per stack, HBM3E above 1.18 TB/s, and HBM4 around 2.8 TB/s based on JEDEC and vendor specifications. It says SK Hynix leads with roughly 50-62% share, Samsung holds about 28-40%, and Micron sits near 5-10%, with all three qualified for HBM4 as of June 2026.
DDR5 RAM for gaming and PC builds
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Supply Gains Still Unproven
Several details remain unsettled. The exact wafer allocation between HBM, DDR5 and GDDR7 is not public across suppliers, and the report’s per-stack prices are market estimates, not uniform contract disclosures. The claimed RTX 50-series cut is also described as reported rather than confirmed by Nvidia in the source material.
It is also not yet clear whether HBM4 qualification will translate into enough reliable volume to loosen supply. The report says the market could reach about $100 billion by 2028, up from around $35 billion, but that outlook depends on AI demand, supplier yields and customer orders.
GDDR7 graphics memory for GPUs
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
HBM4 Race Sets Next Test
The next test is whether SK Hynix, Samsung and Micron can ramp HBM4 at scale while keeping yields high enough to add real capacity. Buyers will be watching 2026 lead times, contract pricing and whether GDDR7 and DDR5 supply improve as more HBM capacity comes online.
The report’s series is set to move next to DDR5 now and DDR6 soon, which should clarify how the HBM squeeze is being felt in mainstream memory. For now, the confirmed picture is narrower: AI demand has made HBM the key constraint, while the pace of relief remains uncertain.
AI accelerator memory modules
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is HBM?
High Bandwidth Memory is stacked DRAM placed close to an AI processor or GPU. Its design gives chips far more memory bandwidth than standard graphics memory, which is why it is used in advanced AI accelerators.
Why does HBM affect normal RAM prices?
The report says HBM uses about 3-4 times more wafer area per bit than DDR5. When memory companies assign more wafers to higher-value HBM, fewer wafers are available for ordinary DRAM.
Who are the main HBM suppliers?
The source identifies SK Hynix, Samsung and Micron as the three main suppliers. It says all three had qualified for HBM4 by June 2026, though their market shares and customer exposure differ.
Is the GPU shortage confirmed?
The broader GDDR7 pressure is attributed to supplier prioritization of HBM in the source material. The claim that Nvidia cut RTX 50-series production by a third or more is described as reported, not independently confirmed here.
What should readers watch next?
The key signals are HBM4 production volume, supplier yields, AI accelerator orders and DDR5 or GDDR7 pricing. If HBM supply improves, pressure could ease; if AI demand keeps rising, the squeeze may continue through 2026.
Source: Thorsten Meyer AI
