How DeepSeek can change everything

Here are four key predictions about how DeepSeek could change the industry.

1. The cost of artificial intelligence will continue to plummet rapidly

All innovations aim to do more with less money, or, from an economic perspective, reduce the cost per unit. The AI revolution is a prime example. Before DeepSeek emerged, the cost of advanced AI models was already declining by approximately 80% year-over-year for two years, and DeepSeek has only accelerated this trend.

The 30-fold difference in token costs between DeepSeek and OpenAI is due to the former company's algorithmic improvements and the latter's more aggressive pricing strategy. As AI research advances and competition intensifies, this downward price trend will only grow.

2. The AI economy will expand and change the structure of value distribution

As artificial intelligence becomes cheaper and more accessible, its usage will grow, following Jevons paradox—the idea that demand for a resource increases when technology makes it more efficient.

When foundational models become a mass-market product, value will shift toward applications: more resources will go toward their usage rather than training. This will increase demand for specialized XPU chips, which are better suited for specific tasks, while GPUs will remain more suitable for model training.

Nvidia has already noted this trend: in May 2024, it reported that demand for model usage reached 40% and is growing faster than demand for training.

As the scope of AI applications expands, research will become more accessible not only to major players, accelerating innovation. For example, thousands of open-source models have already been built based on DeepSeek R-1.

3. US chip export restrictions require careful reevaluation

DeepSeek's breakthrough was achieved with fewer and less powerful chips than those used by American competitors, proving that restrictions can stimulate innovation.

Although US export measures temporarily constrain the development of DeepSeek and its Chinese competitors, they are unlikely to stop their growth. Moreover, such steps could permanently isolate American technology from the Chinese market.

Restrictions also contradict the US goal of reducing its trade deficit with China. As one Chinese minister noted: "We want to buy from the US, but they restrict exports... How do they plan to reduce their deficit?"

US politicians, regardless of party, continue to tighten control over China. However, the pros and cons of these measures are increasingly balancing out, requiring careful reevaluation.

4. US and Chinese tech giants could foster collaboration

Although DeepSeek's technological breakthrough and its open-source code initially caused concern among American investors, many US IT industry leaders have supported these changes.

Microsoft, AWS, and Hugging Face have already integrated DeepSeek R-1 models, expecting that accessible language models will increase demand for cloud services and, consequently, their revenues.

In the long term, companies will benefit from increased productivity and reduced costs due to the widespread adoption of AI, regardless of its origin. This is especially important for the US, where salaries are higher and the shortage of STEM specialists remains a problem. A similar situation occurred in the 1980s with Japanese automakers, whose "lean production" system boosted the efficiency of the American automotive industry.

AI opens up opportunities for collaboration between the US and China, particularly in the race to create artificial general intelligence. However, trade and political disagreements create risks of a technological split.

It remains unclear whether this will lead to partnership or intensified competition, but DeepSeek demonstrates how rapidly the AI world is changing—and will continue to change.

The article was translated from Reuters.com