The bill is starting to become steep for NVIDIA. According to the latest data from Warranty Week’s annual report, the manufacturer saw its expenses related to guarantees explode in 2025, reaching $894 million. A spectacular increase which immediately attracts attention, both for its magnitude and for its timing and which seems closely linked to the persistent problems of the famous 16-pin power connector introduced with the RTX 40 generation.
This sudden jump marks a clear break with previous years and illustrates the concrete financial consequences that controversial technical choices can cause when they are deployed on a large scale on products as widely used as high-end graphics cards.
An explosion in warranty costs in one year at NVIDIA
Until now, NVIDIA kept its warranty expenses within a relatively controlled range. Between 2022 and 2024, these were around 100 million dollars per year, with even a low point at 81 million in 2024. But in the space of twelve months, the situation has completely changed, with a nearly tenfold increase in claims to reach 894 million dollars in 2025.
Â
In detail, the quarterly progression highlights a particularly marked acceleration at the end of the year. After a first quarter already reaching $147 million, the figures remained strong before exploding in the fourth quarter, which alone accounted for $511 million in claims. An anomaly significant enough to suggest a structural phenomenon rather than a simple one-off hazard.
On AMD’s side, the trend is also upward, but in much more contained proportions. The manufacturer went from 110 million dollars in claims in 2024 to 238 million in 2025, with a return rate increasing, but remaining below 1%. A notable development, but not comparable to the situation observed with its direct competitor.
The famous 16-pin connector in the viewfinder
At the heart of the questions is the 16-pin power connector, introduced with the GeForce RTX 40 graphics cards, and in particular the RTX 4090. Since their launch, these models have been associated with reports of overheating, incorrect insertion or even deterioration of the connector, sometimes to the point of partial melting of the connectors. cables.
If NVIDIA has since tried to regulate the use of this standard via stricter recommendations, the problem seems to have continued with the next generation. The RTX 50, and in particular the most energy-intensive models like the RTX 5090 and RTX 5080, continue to present similar cases.
This type of defect is particularly costly, because it does not only concern an isolated component, but can lead to broader damage to the graphics card, or even to the power supply itself, mechanically increasing the average cost of each warranty service.
An economic context which worsens the bill
Beyond technical issues, the economic context also played a determining role in the increase in costs. The end of 2025 was notably marked by trade tensions in the United States, accompanied by a sharp increase in customs duties on certain electronic components.
At the same time, the price of DRAM memory has skyrocketed spectacularly, driven by massive demand linked to artificial intelligence. However, this memory is a key element in the repair of graphics cards, which means that each intervention under warranty today costs much more than before.
Result: NVIDIA must not only manage a sharply increasing volume of returns, but also absorb a significantly higher unit repair cost, which further amplifies the overall financial impact.
Provisions that reach new heights
Another indicator revealing the situation: the provisions set up by NVIDIA to cover these risks. In 2023, the group had set aside $109 million. This amount was multiplied by almost nine in 2024 to reach 948 million, before reaching a new milestone in 2025 with a total of 2.59 billion dollars, including more than a billion in the fourth quarter alone.
This dynamic reflects a clear anticipation on the part of the manufacturer: the costs linked to guarantees are not perceived as a temporary phenomenon, but rather as an underlying trend likely to have a lasting impact on results.
AMD is following a similar trajectory, but on a much more moderate scale, with provisions increasing from 126 million in 2023 to 358 million in 2025.
A warning signal for the entire sector
Beyond the specific case of NVIDIA, this situation highlights the risks associated with the introduction of new hardware standards, particularly when they affect critical elements such as power supply. A failure, even marginal in proportion, can quickly transform into a major industrial problem when volumes are large.
With supply chains still under strain and component costs high, warranty expenses could continue to weigh on GPU manufacturers in the years to come. For NVIDIA, the priority will now be to restore confidence around its solutions, while competition remains particularly aggressive in all market segments.
Â
