AI may be changing how companies work, but the cost of running it is becoming harder to ignore.
NVIDIA’s Bryan Catanzaro, vice president of applied deep learning, said the cost of compute for his team is now far higher than the cost of employees. That is a striking comment from the company at the center of the AI hardware boom.
The point is not that NVIDIA is pulling back from AI. It is the opposite. NVIDIA uses AI heavily across its own work, including models, services, and technologies such as DLSS. But the comment shows that AI is not a cheap replacement for people. At least for advanced work, the infrastructure behind it can cost a huge amount.
AI systems need chips, servers, networking, power, cooling, data centers, software, and constant upgrades. As companies build larger AI models and more advanced agentic AI systems, those costs keep rising.
NVIDIA is not alone. Other companies are also seeing AI spending climb quickly. The wider IT market is expected to grow sharply in 2026, with data center systems seeing the largest jump.
| Category | 2025 spending | 2026 spending | 2026 growth |
|---|---|---|---|
| Data center systems | $505.6 billion | $788.0 billion | 55.8 percent |
| Devices | $791.7 billion | $856.2 billion | 8.2 percent |
| Software | $1.25 trillion | $1.44 trillion | 15.1 percent |
| IT services | $1.72 trillion | $1.87 trillion | 9.0 percent |
| Communications services | $1.30 trillion | $1.36 trillion | 4.8 percent |
| Overall IT | $5.56 trillion | $6.32 trillion | 13.5 percent |
The data center number explains why the AI race is so expensive. Companies are not only buying a few GPUs. They are building entire AI factories, and some of those projects now reach multi gigawatt scale.
This also complicates the idea that AI simply replaces workers to cut costs. In some areas, AI can save time and raise productivity. But at the high end, the compute bill can be enormous. The business case depends on whether the AI output is valuable enough to justify the infrastructure.
NVIDIA CEO Jensen Huang has argued that people who use AI will be more productive and that humans should focus on solving problems rather than only performing tasks. That fits NVIDIA’s broader message: AI is not slowing down, and people should learn how to use it.
Still, Catanzaro’s comment adds a useful reality check. AI may be powerful, but it is not free, and it is not always cheaper than people. As AI becomes more common across businesses, the real winners will be the companies that can use it well enough to make the cost worthwhile.
Discussion (0)
Be the first to comment.