Nvidia RTX 5090 Scam Exposes a Hollow Market for Used Flagship GPUs

Nvidia’s RTX 5090 became the center of a German scam after two buyers reportedly received genuine-looking cards stripped of their most valuable components.

Both transactions involved personal handoffs, according to reports published on September 6 and September 8, 2026. That detail makes the Nvidia RTX 5090 scam more troubling than a basic online bait-and-switch. The victims received physical cards, boxes, and supporting paperwork instead of an obviously unrelated product.

The scheme exploited a dangerous gap between visual authenticity and functional authenticity. A used graphics card can have the correct cooler, circuit board, lighting, and packaging while lacking its graphics processor and memory. An in-person exchange cannot close that gap without a working system and a live test.

Similar stripped-card cases previously appeared around Nvidia’s RTX 4090 and, later, the RTX 5090. The German cases show that the method has spread beyond isolated repair-shop discoveries. They also suggest that scammers can recover value twice, first from the removed chips and again from the convincing remains.

Two Hollow RTX 5090 Cards Point Back to One Seller

The reported transactions were not simple counterfeit sales because the buyers received real RTX 5090 hardware with the essential silicon removed.

The first known buyer found a used Palit GeForce RTX 5090 on Kleinanzeigen during the summer of 2026. He reportedly traveled from Stuttgart to Bremen for a personal collection, hoping that an in-person exchange would reduce the risk.

The seller presented paperwork issued roughly one year earlier for two RTX 5090 cards. According to the account, the seller claimed that one card was no longer needed. The buyer could inspect the product, but he could not test it inside a functioning computer before completing the exchange.

The graphics card produced no display output after the buyer returned home. He eventually sent it to Roman Hartung, the German hardware specialist known online as der8auer, for diagnosis.

Several warning signs appeared before Hartung powered the card. A factory seal covering one cooler screw had been broken, indicating that someone had opened the assembly. The card’s serial number also differed from the number printed on its box.

Those signs did not prove that the card was hollow. Owners sometimes open graphics cards to replace thermal material or install water blocks. Packaging can also become mixed during resale, especially when one owner has several identical products.

Testing produced a stronger clue. The card’s lighting activated, but the board drew no measurable power through its 12V-2x6 connector. It also generated no video signal.

Hartung then removed the backplate and cooler. The underlying circuit board appeared to belong to a real Palit RTX 5090, but the GPU package and every GDDR7 memory module were missing. The published hardware examination described an otherwise convincing board emptied of its defining components.

A second buyer later contacted the publication’s community with a similar account. That person said another RTX 5090 had been obtained through an in-person handoff and arrived without its GPU or memory chips.

The second transaction had a different serial-number pattern. Its invoice and box reportedly matched, while the number on the card itself had been scratched away. The buyer posted an image of the stripped circuit board.

Reporting linked both transactions to what appeared to be the same seller. However, that connection rests on the buyers’ accounts and available transaction details. Authorities had not publicly identified a suspect when the reports appeared.

The first buyer filed a police complaint, according to a separate German investigation. The full origin of the cards and the identity of whoever removed the components remained unknown.

Together, the two reported sales generated several thousand euros for hardware that could never function as advertised. More importantly, they established a repeatable pattern rather than a single buyer’s unexplained loss.

Why the Nvidia RTX 5090 Scam Works in Person

The scam succeeds because a personal handoff proves possession, not the condition or internal identity of the hardware.

Conventional marketplace advice often treats collection as the safer alternative to shipping. That guidance remains useful because collection prevents empty-package tricks and gives buyers a chance to inspect an item. It also keeps payment and delivery in the same place.

Kleinanzeigen itself says collection with payment during the handoff is among the safer transaction methods. Its marketplace safety guide also tells users to examine profiles, images, payment requests, and suspicious pressure from sellers.

A hollow graphics card defeats several of those checks. The seller has a tangible product. The buyer can inspect a branded cooler, feel its weight, compare the box, and see an invoice.

None of those observations proves that the GPU package remains attached to the printed circuit board. The cooler and backplate hide the critical components during any normal inspection.

Opening the card before payment is rarely practical. Disassembly takes time, risks damage, and may create a dispute over warranty seals. A legitimate seller might reasonably refuse that request.

Weight provides little certainty. Modern flagship cards use large heatsinks, metal frames, fans, and backplates. Those parts can dominate the object’s size and apparent substance, even after smaller silicon packages have been removed.

Lighting can also create false confidence. RGB LEDs and auxiliary circuitry do not require a working graphics processor. The first German card illuminated despite having no GPU or video memory.

Paperwork adds another layer of credibility without proving continuity. A valid invoice can show that someone purchased a real product, but it does not establish that the presented card is unchanged. It also does not connect a box to a board when serial numbers differ or disappear.

The reported seller’s paperwork allegedly covered two cards. That detail could make an ordinary explanation sound plausible while supporting repeated transactions. It does not establish who modified the cards, but it illustrates how genuine documents can reinforce a false story.

The effective check is a functional test under conditions controlled by the buyer. The card should be installed in a known working computer, identified by trusted software, and tested under a sustained graphics workload.

A desktop display alone remains incomplete evidence. A manipulated card could fail intermittently, report an unexpected identity, or use substitute hardware. Buyers should compare the software-reported model and memory capacity with the listing and physical labels.

A benchmark also makes it harder to disguise a lower-tier board under a more valuable cooler. Temperatures, clock speeds, power consumption, and performance should align with the claimed model. Any large mismatch deserves an explanation before payment.

The safest exchange therefore requires more than meeting face to face. It requires a setting where the buyer can verify the actual product, not merely the seller’s possession of its visible shell.

The Valuable Parts Were Removed Without Destroying the Disguise

A stripped RTX 5090 preserves the parts that create trust while removing the components that create most of its computing value.

A graphics processing unit is the large silicon processor that performs rendering, parallel computing, and AI workloads. GDDR7 is the high-speed memory surrounding that processor on an RTX 5090 board.

Nvidia lists the GeForce RTX 5090 with 32GB of GDDR7 memory in its official specifications. Removing the GPU and memory eliminates the card’s core processing capability, regardless of what remains attached.

The printed circuit board still contains connectors, power-delivery components, display ports, and smaller control devices. Its cooler may also remain nearly untouched. From the outside, those parts preserve the silhouette and branding buyers expect.

Removing the GPU and memory requires more skill than taking out a removable desktop processor. These components are soldered directly to the board through dense arrays of connections beneath each package.

A technician must apply controlled heat, protect nearby components, and lift each package without damaging it. Reusing the removed silicon also demands specialized rework equipment and experience.

That technical difficulty matters because it challenges the idea of a casual scammer improvising with discarded parts. The person selling the shell did not necessarily perform the extraction. The card could have passed through several hands before reaching the buyer.

Investigators therefore face two connected questions. They must identify who knowingly sold the hollow card and determine where the valuable components went. Evidence for the first answer may not establish the second.

One possible destination is a different circuit board designed for computing workloads. Workshops have previously removed high-end Nvidia chips and memory from retail cards before installing them on custom boards.

That broader supply chain has been associated with demand for AI accelerators and efforts to move restricted hardware in less recognizable forms. However, no public evidence has established that the components from these German cards followed that path.

The distinction is important. A known industry practice can explain why someone values loose GPU and memory packages. It cannot prove the origin, destination, or motive behind a specific stripped card.

The remaining assembly still has limited value as a source of repair parts. A technician might reuse its cooler, fans, connectors, power stages, or undamaged board components. That residual value is far below the value implied by a complete working RTX 5090.

Past cases demonstrate the same economic logic. In June 2025, a repair shop examined a Zotac RTX 5090 that lacked its GPU core and GDDR7 modules. The card retained other board components and its large triple-fan cooler, according to an earlier stripped RTX 5090 case.

Later reports described sellers receiving returned flagship cards with the GPU and memory removed. That version reverses the roles. Instead of a buyer receiving a hollow card, a dishonest buyer allegedly returns the emptied shell after keeping the valuable silicon.

Both versions exploit the same design feature. A complete-looking cooling assembly can conceal major internal changes until someone removes it. The visual identity remains intact after the computing identity disappears.

Scarcity Turns an Obvious Warning Into a Tempting Deal

The central conflict is not counterfeit hardware versus genuine hardware, but trusted appearance versus verified function.

Flagship graphics cards create unusually strong incentives for both sides of that conflict. Buyers search for scarce inventory and meaningful discounts, while criminals target products with concentrated value and strong resale demand.

The RTX 5090 sits at the top of Nvidia’s consumer GeForce range. It serves gaming enthusiasts, creators, local AI users, and professionals who can use its large memory capacity and parallel processing performance.

That broad demand gives a suspicious listing several possible audiences. A buyer may believe the seller simply needs a quick transaction. Another may assume that an untested card offers a repair opportunity.

The first German buyer was reportedly skeptical of the offer. He chose personal collection partly because of that concern. The precaution addressed the risk of receiving nothing, but not the risk of receiving a carefully prepared shell.

This is the reversal at the heart of the case. Actions that normally signal caution can increase confidence enough to complete a dangerous transaction. Travel, face-to-face contact, cash payment, paperwork, and physical inspection all feel more concrete than shipping.

The fraud does not need every detail to survive close scrutiny. It only needs a coherent story that explains why testing is unavailable and why the seller has the product. Once those explanations feel plausible, the physical card supplies the remaining credibility.

A supposedly defective card creates another trap. Experienced hobbyists regularly purchase nonworking hardware because many faults are repairable. A seller can frame the absence of output as an ordinary technical issue rather than evidence of missing silicon.

That framing changes the buyer’s calculation. The buyer stops asking whether the card exists and starts estimating whether it can be repaired. Yet a board without its processor and memory is not a typical repair project.

The growing record of RTX 5090 fraud also weakens the value of marketplace reputation alone. Scams have included unrelated objects, switched products, stripped boards, and allegedly fraudulent returns. Each method attacks a different point in the transaction.

A reputable marketplace cannot inspect every internal component in every privately sold card. Manufacturers also cannot authenticate a board remotely when labels have been removed, altered, or separated from the original packaging.

Sellers face pressure too. An honest owner may provide invoice images to reassure a prospective buyer, only to have those documents copied into another listing. A forum discussion about stolen invoice images shows how proof offered in good faith can become material for another scheme.

The result is a trust problem across the used market. Buyers demand increasingly intrusive proof, while honest sellers become reluctant to share serial numbers and paperwork. Scammers benefit from that confusion because no single artifact remains decisive.

This does not make every discounted or opened card suspicious. Enthusiasts routinely replace thermal materials and cooling systems. Boxes become mixed, and older invoices may contain limited product detail.

The lesson is narrower and more useful. A plausible explanation should never substitute for a functional test when the transaction involves a high-value component with hidden internals.

What the Reports Still Do Not Prove

Two matching accounts establish a serious warning, but they do not reveal the complete chain behind the stripped cards.

The public evidence does not identify the reported seller by name. It also does not show whether police have interviewed a suspect, recovered transaction records, or located the missing components.

The claim that both buyers encountered the same seller appears credible within the published accounts. Still, it has not been independently established through a police statement or court record.

The reports also do not prove who removed the GPU and memory. A person can knowingly sell stolen or manipulated hardware without possessing the equipment needed to modify it. The visible seller might be the technician, an intermediary, or the final link.

Likewise, the AI accelerator theory remains an inference based on earlier hardware conversions. High-end GPU packages have uses beyond gaming, and smaller components are easier to transport than complete retail cards.

That context explains a possible incentive. It does not connect the German boards to any specific workshop, buyer, destination, or country.

The phrase “hollow RTX 5090” also needs precision. The reported cards were not empty plastic shells. They retained real coolers and printed circuit boards, plus enough supporting circuitry for one card’s lighting to activate.

That distinction makes the fraud more convincing. It also means investigators might extract evidence from board markings, repair traces, component residue, packaging, invoices, messages, or payment details.

Serial numbers could help reconstruct the cards’ paths, but only if manufacturers, distributors, and retailers retain matching records. A mismatched box or scratched identifier complicates that work without making it impossible.

The first card’s broken seal showed prior access, yet it could not reveal when or where the modification occurred. The second card reportedly had matching numbers on its invoice and box, while the board’s identifier was obscured.

No published evidence shows that Nvidia, Palit, or Kleinanzeigen caused the reported losses. The transactions involved private resale, and the alleged manipulation occurred outside normal manufacturing controls.

There is also no verified count for how many similar cards circulate in Germany. Two connected cases can indicate repetition, but they cannot establish the scale of a national fraud wave.

Reports from other countries show that stripped GPU schemes are not unique. However, combining every online complaint into one organized operation would overstate the evidence.

Some stories may involve separate criminals copying a known method. Others may reflect return fraud, mislabeled spare parts, or disputes where the full transaction record is unavailable.

Readers should therefore separate three conclusions. Stripped flagship cards demonstrably exist, at least two German buyers reported closely related losses, and the wider organization behind those cards remains unproven.

That cautious framing does not reduce the practical risk. Buyers do not need to know who removed a GPU before requiring a live test. They only need to understand that an authentic exterior no longer guarantees authentic internals.

Three Signals Will Show Whether the Scam Is Spreading

Police findings, marketplace patterns, and manufacturer tracing will determine whether these sales were isolated or part of a larger pipeline.

The first signal is any official update from German law enforcement. Investigators could confirm whether the two buyers dealt with one person, whether additional complaints exist, and whether transaction evidence points upstream.

A confirmed series of linked sales would strengthen the view that stripped cards have become a repeatable resale business. An isolated complaint with no further matches would weaken claims about broad local circulation.

The second signal is a rise in similar reports from repair specialists and marketplace users. The most useful accounts will include board photographs, matching transaction details, serial-number evidence, and documented attempts to test the card.

More verified examples across different sellers would suggest that the technique has spread beyond one source. Recycled screenshots and secondhand social posts would add attention without adding reliable evidence.

The third signal is whether manufacturers or retailers develop better authentication and chain-of-custody tools. Buyers currently depend on labels, invoices, and visual inspection, yet each can become detached from the working hardware.

A serial-number lookup could confirm the original model and seller, but it still could not prove that the silicon remains installed. Tamper-evident construction might expose access, although legitimate repairs and cooler changes would complicate warranty policies.

Marketplaces can also improve transaction guidance for high-value components. A recommendation to collect in person is insufficient when the product can pass a surface inspection. Guidance should emphasize live testing, protected payment, evidence preservation, and refusal when verification is blocked.

For buyers, the immediate response is straightforward. Arrange the meeting where a known working computer is available, bring the required power connections, and run identification software before paying.

Match the card, box, and invoice serial numbers. Examine screws and seals for tool marks. Treat missing or scratched identifiers as unresolved risks, not cosmetic defects.

Ask the seller to demonstrate a sustained workload rather than a brief desktop screen. Confirm the reported model, memory capacity, power behavior, display outputs, and benchmark performance.

Walk away if the seller unexpectedly prevents testing. A compelling explanation does not restore buyer protection after cash changes hands.

Preserve the listing, messages, invoice images, profile details, and meeting information if fraud is suspected. Report the transaction promptly to the marketplace and police rather than attempting private confrontation.

The Nvidia RTX 5090 scam shows that possession, packaging, and even personal collection can create confidence without proving functionality. The deciding question is now simple: will used-GPU buyers demand evidence from the running hardware before trusting the shell?