Nvidia AI server prices could rise more than 15%, report says

# Nvidia customers reportedly face AI server price hikes of more than 15%

The cost of building massive AI infrastructure could be about to climb even higher.

Some of Nvidia’s largest customers have reportedly been told that servers powered by the company’s AI chips could become **more than 15% more expensive in many cases**, as surging memory costs put fresh pressure on the hardware behind the artificial intelligence boom.

The reported price increases are expected to affect systems based on Nvidia’s latest **Vera Rubin and Grace Blackwell platforms**, according to Bloomberg News, as cited by Reuters.

The higher prices are expected to apply to systems shipping from **early 2027**, although the exact increase could vary depending on the generation of hardware and the amount and type of memory used.

Nvidia has not publicly confirmed the reported increases.

## AI servers are getting more expensive

The reported price increase highlights a growing problem for companies racing to build increasingly powerful AI data centers.

Modern AI systems require enormous amounts of specialized memory alongside GPUs.

As demand for artificial intelligence infrastructure has accelerated, advanced memory — particularly the high-bandwidth memory used alongside AI accelerators — has become an increasingly important part of the cost of an AI server.

According to the report, rising memory prices are now significant enough that Nvidia and its server partners are preparing to pass part of those additional costs on to customers.

Server manufacturers supplying large cloud and data-center operators have reportedly already begun communicating the new pricing expectations to their own customers.

That could ultimately make the next generation of AI infrastructure considerably more expensive to deploy at scale.

## Nvidia's Vera Rubin systems could be affected

One of the most important platforms mentioned in the report is Nvidia’s new **Vera Rubin architecture**.

Nvidia describes Vera Rubin as its next-generation rack-scale computing platform for large AI workloads.

A Vera Rubin NVL72 system combines **72 Rubin GPUs and 36 Vera CPUs** along with Nvidia networking and data-processing technology into a single rack-scale AI system.

Nvidia announced earlier this year that Vera Rubin had entered full production as cloud providers, AI companies and server manufacturers prepared to deploy the platform globally.

The Rubin GPU itself contains **288 GB of HBM4 memory** and is designed for extremely demanding AI inference and training workloads.

That enormous dependence on advanced memory helps explain why changes in memory pricing can have such a large impact on the final price of AI systems.

## The AI infrastructure race is already enormously expensive

The reported Nvidia increases arrive as technology companies are committing unprecedented amounts of capital to AI infrastructure.

Training and running the largest AI models increasingly requires data centers containing thousands — and eventually potentially hundreds of thousands — of accelerators connected through high-speed networking.

Nvidia has positioned Vera Rubin specifically for this new generation of extremely large AI factories.

Microsoft, CoreWeave and other infrastructure providers are among companies planning deployments based on Nvidia’s newest architecture.

A price increase of more than 15% therefore matters far beyond the cost of an individual GPU.

For hyperscalers ordering entire racks and constructing multi-gigawatt data-center campuses, relatively small changes in hardware prices can translate into enormous differences in overall capital spending.

## Nvidia remains at the center of the AI boom

Despite increasing competition from AMD and custom AI accelerators developed by major cloud companies, Nvidia remains one of the central suppliers powering the current AI infrastructure buildout.

Its newest hardware is also moving beyond traditional model training.

Nvidia says Vera Rubin is designed specifically for increasingly demanding **agentic AI and inference workloads**, with the company claiming substantially higher performance and efficiency compared with its previous Grace Blackwell generation.

That makes the reported pricing development especially important for companies planning their next generation of AI capacity.

The question is no longer simply how many GPUs companies can obtain.

It is increasingly becoming how much it will cost to power, cool, connect and equip the enormous computing facilities required to run the next wave of AI models.

## Nvidia earnings are coming this week

The timing of the report is also notable.

Nvidia is scheduled to release its **second-quarter earnings on August 26**, making the company one of the biggest events for markets this week.

Investors will be watching closely for evidence that spending on AI infrastructure remains strong and for any signals about demand, supply constraints and Nvidia’s next-generation products.

For now, the reported server price increase has not been confirmed directly by Nvidia.

But if customers ultimately face increases above 15%, the rising cost of memory could become yet another major expense in an AI infrastructure race that is already costing the technology industry hundreds of billions of dollars.