NVIDIA’s planned increase in server prices is forcing Silicon Valley to rethink how it uses expensive AI chips. NVIDIA is raising prices for servers containing its accelerator processors by more than 15%, according to Bloomberg News.
The shift could push major technology companies towards more efficient software instead of simply buying more computing power. The company has strong pricing power in the AI chip market, with gross profit margins of about 75% and an estimated 70% to 90% share of the global AI chip market.
Higher costs are unlikely to stop major customers such as Meta Platforms, Microsoft and OpenAI from buying Nvidia hardware. They could, however, make those companies look more closely at how efficiently they use the chips.
Chinese AI companies have faced tighter access to advanced NVIDIA processors for years. Those restrictions have encouraged companies to find ways to achieve more with limited computing resources.
Hangzhou-based DeepSeek is one example. The company developed a mathematical technique called multi-head latent attention to reduce the memory required by its AI models. The approach reportedly cut memory needs by about 96%, allowing DeepSeek to develop capable AI systems with lower hardware requirements.
Silicon Valley companies have also worked on improving software efficiency. OpenAI and Meta have developed tools to keep GPUs working more effectively. These efforts have not fully addressed the wider software challenges inside large AI data centres.
Some industry estimates suggest AI companies use less than 15% of the theoretical capacity of Nvidia GPUs for customer-facing workloads. Other estimates put effective utilisation as low as 5%.
For years, technology companies have dealt with inefficient chip use by purchasing more processors. The rapid pace of AI development made that approach easier to justify, even as data-centre costs climbed.
The pressure is now increasing. One AI semiconductor entrepreneur described the situation by saying, "We're right now in a world where we need more performance than ever, and the physics has run out,"
Nvidia is also investing in software aimed at improving chip utilization. The company recently spent USD 6 billion on licenses from San Francisco-based Poolside AI, whose software helps manage data movement between thousands of GPUs in data centres.
Poolside co-CEO Eiso Kant said, “You spend all of your time improving the factory itself.”
He also argued that huge AI development budgets do not always reflect the actual cost required to build capable models. Poolside says it trained its coding model Laguna in about eight weeks using efficiency techniques.
Also Read: Apple to Acquire AI Chip Startups to Close the Gap with Rivals?