TL;DR

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Footage shared by HKEPC on October 23, 2026, shows large numbers of GeForce RTX 5090 cards at a Chinese workshop, where technicians reportedly remove their GPU dies and memory for installation on custom AI-server boards. The video documents a reported conversion process, but does not establish how many cards are being modified, who operates the facility, or whether the cards arrived through authorized channels.

Footage shared by HKEPC on October 23, 2026, shows shelves stocked with GeForce RTX 5090 cards at a Chinese workshop, where technicians are reportedly removing the GPU dies and 32GB of GDDR7 memory for installation on custom boards intended for AI servers. The footage points to another use for the high-end consumer card amid reports of graphics hardware being repurposed for AI workloads, but it does not independently verify the operation’s scale or the cards’ supply chain.

In the video, the workshop appears to hold hundreds of RTX 5090 cards from multiple manufacturers, according to Wccftech’s account of the footage. The displayed stock reportedly includes premium models such as the ROG Astral RTX 5090, as well as Founders Edition cards. The video is evidence of cards present in a facility, but it does not confirm when they were made or how they were acquired.

The reported modification removes the GPU chip and memory from each card’s original board and transfers those components to a custom PCB with a dual-slot blower-style cooler. That narrower cooling design is intended to fit more readily into server systems with limited space, unlike the large multi-fan coolers used on many retail gaming cards. The footage does not show independent testing of the modified hardware or establish its performance in AI workloads.

Wccftech says RTX 5090 conversion for AI use has been reported before, including in the previous year. The card’s 32GB of GDDR7 memory and GPU capabilities make it a candidate for some AI applications, but the images alone do not show which models or server configurations the workshop is building. Nor do they establish that the pictured cards were officially sold for server use.

Why 32GB Cards Suit Server Builds

Repurposing the RTX 5090’s core components could give system builders a way to assemble compact AI accelerators from hardware designed for desktop PCs. Replacing a bulky retail cooler and board with a blower-equipped custom design may make the components easier to install in dense server enclosures. That is a plausible practical advantage, though the footage provides no measurements of system density, operating temperatures, or output.

The reported activity also matters to consumers because demand for AI hardware can compete with demand from gamers for high-end GPUs. If substantial numbers of retail cards are redirected into server builds, availability and pricing could be affected. The video does not reveal sales volumes or prove a market-wide effect, so any impact on retail supply remains uncertain.

The operation also highlights a gap between a chip’s original product category and its later use. The RTX 5090 is marketed as a consumer graphics card, but its memory capacity and compute resources can attract buyers seeking AI hardware. The footage shows a reported route from retail-style cards to custom server boards; it does not establish whether that route is authorized by the manufacturers or how widely it is used.

Earlier RTX 5090 AI Conversions

Wccftech reports that large inventories of RTX 5090 cards have been converted for AI use in China before. It also describes the RTX 5090 as a card that remains popular for AI applications in the region. The new video, shared by HKEPC on social media, adds images of what appears to be a workshop-based process rather than independently verified sales or deployment records.

Wccftech’s account refers to claims that the RTX 5090 had been discontinued, but also says NVIDIA recently rejected those rumors. The two points should be kept separate: the footage may show cards being redirected into AI systems, but it does not establish the product’s manufacturing status or prove that current production is devoted solely to AI-server demand. No direct statement from NVIDIA about the workshop is included in the reporting.

Supply Routes and Scale Unverified

The video does not establish how many cards are actually being modified, whether the visible inventory represents a single shipment, or how many finished server cards have been produced. It also does not identify the workshop’s operator, customers, or the origin of the GPUs. Wccftech’s report describes the cards as entering China despite restrictions, but the footage itself does not document their route or prove that any particular shipment violated export rules.

Other important details are absent: there is no confirmed information about the custom boards’ specifications, reliability, cooling performance, price, or AI workload results. The video also does not show whether NVIDIA or the card manufacturers knew about or approved the conversions. Claims about the company’s awareness or financial incentives are not substantiated by the footage described in the report.

Further Verification Needed

The next useful evidence would be a response from NVIDIA or the card manufacturers, along with independent confirmation of the workshop’s location, inventory, and conversion process. Details about the custom PCB and its performance would help establish whether the modified cards are practical server accelerators or simply components shown during disassembly.

Until such information is available, the footage supports a limited conclusion: RTX 5090 cards are reportedly being dismantled in a Chinese workshop, and their core components are being prepared for custom AI-server boards. It does not settle how extensive the practice is, how the hardware reached the facility, or whether it has affected consumer availability.

Source: Wccftech

Key Questions

What are technicians reportedly doing to the RTX 5090 cards?

They are reportedly removing the GPU dies and GDDR7 memory from the original cards and fitting those components to custom boards with dual-slot blower-style coolers.

Why convert a gaming GPU for an AI server?

The RTX 5090’s 32GB of memory and compute capabilities may suit some AI workloads. A custom board and smaller blower cooler can also be easier to fit into a space-constrained server, although the video does not provide performance tests.

Does the video prove RTX 5090 production has ended?

No. The footage shows cards at a workshop, but it does not establish the card’s manufacturing status. Wccftech says NVIDIA recently refuted discontinuation rumors.

Is the workshop’s inventory independently confirmed?

No. HKEPC’s video reportedly shows a large quantity of cards, but the exact count, ownership, supply route, and scale of completed conversions have not been independently established.

Source: Wccftech