Technology news and discussion on TalkFuse.
There is no end in sight for the GPU supply problem
Demand from AI data centres is consuming memory and drawing manufacturers towards more profitable data-centre products, worsening shortages for consumer GPUs and driving prices sharply higher. The article says there is little sign of relief, with some suppliers warning that tight conditions could continue into 2027 and 2028, and possibly beyond.
The shortage centres on DRAM capacity being redirected to high-bandwidth memory (HBM), which AI systems need and which competes for the same production resources. Three firms produce about 90 per cent of global DRAM, while data centres are projected to use around 70 per cent of supply this year. A JP Morgan report forecasts DRAM prices will be 400 per cent higher by the end of 2026 than in 2024; the article also cites the RTX 5090 selling for $4,920, compared with its $1,999 launch price.
In short
- AI data centres are intensifying the memory shortage.
- DRAM prices could rise 400 per cent by the end of 2026.
- Industry leaders expect supply to stay tight through 2028.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
AI data centres are creating strong demand for high-bandwidth memory, and manufacturers may reasonably prioritise products that earn more and support major computing investments. Redirecting capacity can help meet those needs and sustain investment in advanced memory production, even if consumer GPUs become harder to find and more expensive in the short term.
The case against
The concentration of DRAM production among a few firms makes consumers vulnerable when manufacturers shift capacity towards more profitable data-centre products. Higher prices and prolonged shortages can put consumer GPUs beyond reach and suggest that supply decisions are imposing substantial costs on buyers who have few alternatives.
Explained simply
Graphics processing units, or GPUs, power the graphics in computers and are also used for demanding calculations. They rely on memory chips called DRAM; AI data centres use a specialised form, high-bandwidth memory, to process large amounts of data quickly.
A small group of manufacturers makes most of the world’s DRAM, and the same production resources can serve both AI and consumer products. As data centres take a larger share of supply, fewer chips may be available for graphics cards used in home computers and gaming PCs.
This matters to people buying or upgrading computers because tighter supply can mean higher prices and fewer choices. The balance between AI demand and memory production will help shape what is available to consumers in the coming years.