Nvidia AI servers face 15 per cent price hike

NEW DELHI: AI may be getting smarter, but the servers powering it are getting costlier.

Major customers of Nvidia are facing AI server price increases of more than 15 per cent in many cases as rising memory chip costs push up the cost of computing infrastructure, according to a Bloomberg News report.

The increases are expected to affect systems scheduled for shipment early next year, including servers based on Nvidia’s Vera Rubin and Grace Blackwell platforms. The exact increase will depend on the chip generation and memory configuration used in each system.

Server manufacturers supplying large data centre operators have already informed some customers about the expected price changes, people familiar with the matter told Bloomberg.

Companies including Microsoft, Google and Oracle are among the major operators expected to be affected.

Nvidia did not immediately respond to requests for comment.

The price increases come as demand for memory chips used alongside AI accelerators continues to outstrip supply.

Samsung Electronics, SK Hynix and Micron Technology account for most of the world’s DRAM production, but their output has struggled to keep pace with demand, according to the report.

Memory is a critical component of AI computing systems. Nvidia’s accelerators depend on high-performance memory to process the huge volumes of data required to train and run increasingly sophisticated AI models.

As memory becomes more expensive, the pressure is being passed through to complete AI server systems.

Nvidia has considerable pricing power in the AI accelerator market, where its chips can sell for tens of thousands of dollars each. The latest increases, however, show that even a company with significant control over AI chip pricing cannot completely shield its customers from higher component costs.

The price increases could complicate infrastructure planning for some of Nvidia’s largest customers as technology companies continue to spend heavily on AI data centres.

Amazon, Microsoft, Google and Meta Platforms are all developing their own AI chips, partly to reduce dependence on Nvidia.

But building alternative processors does not eliminate the broader memory constraint. Access to memory supplied by Samsung, SK Hynix and Micron remains critical to the wider AI infrastructure ecosystem.

For technology companies planning large-scale AI deployments, higher server prices could therefore add another layer of pressure to already substantial infrastructure budgets.

The memory squeeze arrives as companies push ahead with ambitious AI data centre projects around the world.

The sector is already facing a range of challenges, including project delays, labour shortages, tighter capital markets and opposition from local communities to new data centre developments.

Higher hardware costs could add to those pressures, particularly for companies planning large deployments that require thousands of AI servers.

For Nvidia, the developments come at a closely watched moment. The company is scheduled to report its fiscal second-quarter results on August 26 and has increasingly become a key indicator of the health of the broader AI infrastructure boom.

The latest price increases highlight a less glamorous side of the AI race: as demand for computing power soars, the cost of the memory needed to power that intelligence is becoming a bottleneck of its own.