Nvidia plans to raise AI server prices by more than 15% for major customers
The chipmaker is set to push up the cost of its most advanced AI server systems next year, hit by soaring memory chip bills it can no longer absorb alone.
Key points
- Nvidia plans to raise prices on AI server systems by more than 15% for many large customers, Bloomberg first reported on 19 July 2026.
- The increases will cover servers built around its newest chip families, including Grace Blackwell and Vera Rubin.
- How much each customer pays will depend on the chip generation and how much memory is packed into the system.
- Shipments carrying the new prices are expected to go out in 2027.
Nvidia, the company whose chips power nearly every major AI system from chatbots to self-driving cars, is about to get more expensive. The cloud operators and data-centre buyers that purchase servers by the thousands have been told to expect increases of more than 15% on many orders, according to Bloomberg.
The affected products are servers built around Grace Blackwell, which began shipping in late 2025, and Vera Rubin, the generation following it. Both are AI accelerators: specialised processors designed to run the number-crunching that large AI models need. When you stack hundreds of them into a single server, that's how companies build the computing power behind tools like ChatGPT.
Why is Nvidia raising prices now?
Memory costs are the driver. Every AI chip needs large amounts of a fast, specialised type called HBM (high-bandwidth memory) to feed data to the processor quickly enough to be useful. Its price has risen sharply, and Nvidia says it can't absorb the full cost anymore. A server with more memory, or built around a newer chip generation, will see a bigger jump.
What does this mean for ordinary people?
Most people won't get a bill from Nvidia directly. The companies that do, including Amazon and Microsoft, pass infrastructure costs on through cloud pricing. If you pay for AI tools at work or subscribe to any service running on rented computing power, higher Nvidia server bills will eventually filter down. Not overnight, and not as a line item you'll see. More likely as a slow drift upward in subscription prices, or a reason your employer cites for delaying AI projects.
For buyers, the timing is awkward. Data-centre spending is already at record levels. A 15%-plus price jump on hardware that can cost tens of millions of dollars per rack forces a real decision: pay more, or look harder at alternatives from AMD or custom chips built in-house by the cloud providers themselves. Our 19 August story on Nvidia's $600 billion computing deal with OpenAI showed just how deep those infrastructure commitments have already run.
What happens next?
The new prices apply to systems shipped next year, so buyers still have time to lock in existing contracts or renegotiate. Nvidia hasn't issued an official statement confirming the figures. Expect the company's next earnings call to be the first place it addresses the increases on the record.
Here's what this beat tells me: Nvidia still holds enough pricing power to push costs outward rather than absorb them. Its customers, for now, don't have many places else to go. That's the number worth watching, not the 15% itself.
