Price Pressure on Chips
NVIDIA is raising prices for advanced AI chips. Usually a manufacturer passes on cost increases, but now the company is raising prices. This decision could force Silicon Valley to reconsider its plans. Meta Platforms Inc., Microsoft Corp., and OpenAI have spent enormous sums on NVIDIA equipment to train and run models.
- NVIDIA is raising prices for advanced chips.
- Silicon Valley may reconsider its plans.
- Meta, Microsoft, and OpenAI have spent enormous sums on equipment.
If chip costs continue to rise, buying more equipment may not be the best answer. Companies may need to find ways to get more out of the chips they already have. The priority is shifting from the quantity of purchases to efficiency of use.
A Shift in the Logic of the Race
The AI race has largely been about obtaining more computing power. More chips meant the ability to train larger and more advanced models. This approach is becoming harder and more expensive.
| Criterion | Previous priority | New priority |
|---|---|---|
| Primary resource | Number of chips | Efficiency of each chip |
| Goal | Train larger models | Do more work with existing equipment |
| Constraint | Access to computing power | Rising cost and complexity |
The winner is the one who squeezes more work out of the available equipment.
The Experience of Chinese Companies
Chinese AI companies have faced restrictions on access to advanced NVIDIA chips. Instead of simply slowing down, they looked for other ways to improve their AI systems.
- Focused on improving model efficiency.
- Used available equipment carefully.
- Looked for other ways to improve AI systems.
DeepSeek built strong AI models despite having less access to the most advanced chips. When chip access became a problem, the developers implemented a mathematical reduction called multi-head latent attention. It cut the memory requirement of the chips by roughly 96%.
DeepSeek's progress suggests that having fewer top-end chips does not always mean falling behind. For Silicon Valley, the lesson is simple: smarter AI can matter just as much as more powerful hardware.
Software as a Lever for Efficiency
Meta Platforms Inc. and OpenAI are working on new software tools and programming systems. They can help their AI models work more efficiently.
- Better software helps models do more work.
- New tools reduce waste and improve performance.
- Software is becoming a separate criterion when choosing a strategy.
Investment in software and development tools is becoming no less important than buying hardware.
A Practical Criterion
NVIDIA chips will remain important to the AI industry. However, companies will focus less on how many chips they can buy. The question of how much work each chip can handle will matter more.
- Look beyond access to computing power.
- Consider model efficiency.
- Evaluate software solutions.
- Compare the return from existing equipment.
Smarter AI can deliver the same result as more powerful hardware.
