TL;DR
A Thorsten Meyer AI report says High Bandwidth Memory has become the key pressure point in the 2026 memory crunch, absorbing wafer capacity that would otherwise go to DDR5 and other memory products. The report says HBM demand is also affecting GDDR7 supply for consumer GPUs, while supplier competition around HBM4 may determine how much relief arrives next.
High Bandwidth Memory has become the component driving much of the 2026 memory squeeze, according to a late-June report from Thorsten Meyer AI, which says AI accelerator demand is pulling wafer capacity away from ordinary DRAM and contributing to tighter supplies of DDR5 and GDDR7.
The report identifies HBM as the higher-value product memory manufacturers are prioritizing over standard RAM. Unlike flat DDR5 modules, HBM stacks multiple DRAM dies vertically, connects them with through-silicon vias, and places the stack near a GPU or accelerator to deliver much higher bandwidth.
According to the report, one bit of HBM can consume roughly three to four times the wafer area of one bit of DDR5. The report attributes that penalty to larger dies, complex stacking and lower yields, since a defect in one layer can affect the whole stack.
The economics help explain the shift. The report estimates HBM3 at about $200 per stack, HBM3E near $300, and HBM4 around $500. It 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 Demand Reprices DRAM
The report’s central finding is that AI accelerator demand is no longer a niche influence on memory markets. It says HBM has grown from about 8% of DRAM revenue in 2023 toward a projected 41%, with the HBM market estimated to rise from $35 billion to about $100 billion by 2028.
That matters for readers because the pressure can show up far from data centers. The report says wafer allocation toward AI-focused HBM reduces available capacity for ordinary memory, which can affect PC builders, server buyers and consumer electronics supply chains.
The same pressure may extend to graphics cards. The report says suppliers prioritizing HBM have contributed to shortages of GDDR7, the memory used in newer consumer GPUs. Citing industry reporting, it says Nvidia reportedly reduced RTX 50-series production by a third or more in the first half of 2026.
High Bandwidth Memory HBM modules
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How HBM Became Central
HBM is central to modern AI chips because training and inference are often limited by how quickly data can reach the processor. The report says HBM provides roughly five to ten times the bandwidth of normal graphics memory, reducing the risk that expensive accelerator silicon sits idle.
Major AI accelerators are built around that need. The report cites Nvidia H100, H200, B200 and the coming Rubin platform, along with AMD MI300-series chips, as examples of products designed around HBM stacks.
The supplier race is concentrated among SK Hynix, Samsung and Micron. The report says SK Hynix leads with about 50% to 62% share, Samsung holds about 28% to 40%, and Micron has about 5% to 10%. It also says all three had qualified for HBM4 by June 2026, shifting the question from basic qualification to shipment quality and volume.
“The thing the factories make instead of your RAM is a tower of stacked memory bolted to every AI chip.”
— Thorsten Meyer AI report
DDR5 RAM for gaming PC
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Supply Relief Still Unproven
Several figures in the report are estimates or attributed to industry sources, including per-stack pricing, supplier shares and the reported RTX 50-series production reduction. The report also labels the market as fast-moving as of late June 2026.
It is not yet clear how quickly HBM4 production can scale, whether yields will improve enough to ease wafer pressure, or how much added supply will reach customers outside the largest AI accelerator buyers.
The report also identifies a risk on the demand side: if AI spending slows, HBM could be the first part of the memory market to feel the break because so much capacity and pricing power are tied to that demand.
GDDR7 graphics card memory
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
HBM4 Output Becomes Test
The next marker is whether SK Hynix, Samsung and Micron can turn HBM4 qualification into reliable high-volume shipments. More competition could add supply and reduce pressure, but only if output rises faster than AI accelerator demand.
The Thorsten Meyer AI series says its next installment will examine DDR5 now and DDR6 soon, placing the HBM squeeze in the broader memory upgrade cycle that PC, server and GPU buyers will face 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 Thorsten Meyer AI report says HBM demand has become a central driver of the memory crunch, affecting DRAM capacity, pricing and some GPU memory supply.
Is this breaking news or analysis?
This is an analysis based on a reported market shift, supplier data, industry estimates and the reported status of HBM4 qualification as of late June 2026.
What is confirmed versus uncertain?
Confirmed technical facts include HBM’s stacked design and its role in high-bandwidth AI chips. Market figures such as pricing, supplier shares and reported GPU production cuts are attributed to the report and cited industry sources, and may change.
Why does HBM affect regular RAM prices?
The report says HBM uses more wafer capacity per bit than DDR5. When manufacturers allocate more wafers to high-priced HBM, less capacity is available for ordinary DRAM, tightening supply.
Could HBM4 ease the shortage?
Possibly, but relief depends on production volume, yields and demand. The report says all three major suppliers qualified for HBM4 by June 2026, but output at scale remains the key test.
Source: Thorsten Meyer AI
