Report Alleges Grey-Market Conversion of RTX 5090 GPUs for China AI Servers
DataNews Editorial Team
A single-source report alleges that new NVIDIA GeForce RTX 5090 graphics cards are being dismantled in China and rebuilt as dual-slot, blower-cooled units for server-oriented AI deployments. The report says the converted cards are tested, repackaged and shipped to data centers or AI laboratories.
- S1 alleges that RTX 5090 components are transferred to dual-slot, blower-cooled boards for server-oriented use in China.
- The report says tested and repackaged cards are supplied to data centers and AI laboratories, but identifies no buyers.
- S1 claims several thousand cards are repurposed almost daily; the scale has not been independently verified.
- The reported starting price for a customised RTX 5090 is US$8,500 per card.
- Any use of such equipment in AI clusters could introduce uncertainty around specifications, support, warranty and supply continuity.
S1 reports that a leaked video shows Chinese workshops dismantling new NVIDIA GeForce RTX 5090 cards and transferring their GB202 GPUs and GDDR7 memory onto dual-slot boards with blower cooling. According to the report, the resulting products are intended for dense server racks and AI compute clusters.
The source says the converted cards undergo testing before being repackaged and supplied to data centers or AI laboratories. It characterises the activity as a grey-market network operating across China. S1 also states that the GeForce RTX 5090 was not officially launched in China and reaches the market through black-market channels.
S1 claims that several thousand RTX 5090 cards are repurposed almost daily. That figure is not independently verified and is not supported by a disclosed methodology, underlying data or identified participants in the supply chain. The source puts the starting price of a customised RTX 5090 at US$8,500 per card.
If the reported practice is operating at the claimed scale, it could support AI-cluster expansion through unofficially converted consumer GPUs while increasing pressure on retail RTX 5090 supply. For operators and buyers of AI infrastructure, the reported channel may introduce non-standard hardware with unclear performance, power draw, reliability, warranty status and long-term availability.
No specific data centers, AI laboratories, conversion-board manufacturers or end buyers were identified in the source material.
If the reported activity is occurring at the claimed scale, it could create an unofficial supply channel for AI compute while tightening retail availability of RTX 5090 cards. For data-center operators, it also raises questions around equipment standardisation, technical specifications, support, warranty coverage and supply-chain reliability.
- TechNews 科技新報 / ChineseSource →
This article was prepared by the DataNews editorial team based on the sources listed above.