In recent years, the artificial intelligence race seemed to follow a logic of building the smartest model.

OpenAI launched ChatGPT. Google responded with Gemini. Anthropic gained ground with Claude. Meanwhile, the market tracked benchmarks, reasoning capabilities, coding, and multimodality as if each new release defined a winner.

But in recent weeks, the discussion has gained a new factor beyond performance. Who will succeed in getting as many companies, governments, and developers as possible to build on top of their technology?

It was precisely this vision that appeared in Xi Jinping's speech during the opening of the World AI Conference (WAIC), held in Shanghai in July. Announcing the creation of the World AI Cooperation Organization (WAICO), initially formed by 29 countries, the Chinese president argued that artificial intelligence should not be "a solo performance by a single country," but "a symphony of international cooperation."

At first glance, the speech seems merely diplomatic. In practice, it may reveal an economic strategy that could change the AI market once again.

How does the Chinese strategy work?

The market assumed that AI companies would make money the same way software companies do, by charging a subscription for access to the best models.

This remains the path followed by OpenAI, Anthropic, and Google. China, however, seems to be betting on a different logic.

Models like Qwen, developed by Alibaba, DeepSeek, and more recently, Kimi K3 by Moonshot AI, are being distributed as open-weight models, allowing companies to run, adapt, and customize these systems with much more freedom than fully closed platforms.

READ ALSO: Qwen: the Chinese AI that could rival ChatGPT and Gemini by 2026

Open-weight does not necessarily mean open source. In most cases, companies make the model weights available (the parameters learned during training), but keep elements like the data used, full architecture, and training processes private.

Even so, this represents an important shift. Companies no longer depend exclusively on proprietary APIs and can now run models on their own infrastructure, fine-tune them for specific needs, and reduce operational costs.

The goal is no longer to sell every single API call. It becomes getting as many people as possible to use that ecosystem.

What is Kimi K3?

The Kimi K3 model by Moonshot AI was introduced during the same week as the WAIC and has since ranked among the most advanced open-weight models made available, approaching the performance of leading commercial models in coding and reasoning tasks.

By making a model of this caliber available to developers, Moonshot reduces one of the key competitive advantages of American labs, which is exclusive access to cutting-edge technology.

It is a strategy similar to what has happened at other times in the industry. Linux did not become popular because it was the most beautiful operating system; it became popular because thousands of companies started building on top of it. Android followed a similar path. Today, China is trying to do something similar with artificial intelligence.

Why is AI getting cheaper?

The Chinese strategy gained momentum precisely at a time when the market began to compete on price.

In recent days, OpenAI announced a reduction of up to 80% in the cost of GPT-5.6 Luna, its most economical model aimed at high-volume tasks. According to the company, part of these savings came from using AI itself to optimize the components responsible for model inference.

The response came quickly. The following day, DeepSeek put V4-Flash into public beta, offering even lower prices and compatibility with the exact same API used by OpenAI, facilitating the migration of existing applications.

These moves reinforce a perception that models are getting closer in performance, while price has become one of the main decision-making factors for companies.

Does cheaper AI mean lower spending?

AI companies are undergoing an intense price war. Models have become significantly cheaper, while new Chinese competitors have accelerated this reduction by offering lower-cost alternatives with performance increasingly close to that of market leaders.

At the same time, companies have begun to notice a curious effect. Although the price per token is dropping, total spending continues to increase. The more accessible AI becomes, the more it is used.

LEARN MORE: Subscription vs. token LLMs: which offers the best cost-benefit ratio?

This phenomenon has already appeared in various organizations, which have started treating token consumption the same way they monitor cloud or infrastructure costs.

The result is a market where efficiency matters as much as technical capability.

In this scenario, making cheaper models (or even free ones in certain formats) available can further accelerate their global adoption.

Can AI become infrastructure like the internet?

In the early 2000s, infrastructure companies concentrated a large portion of the digital market's value. Cisco, server manufacturers, and network equipment suppliers held strategic positions.

Over time, the internet, cloud, and connectivity stopped being competitive differentiators and began functioning as infrastructure.

Google, Amazon, Facebook, and Netflix did not become among the most valuable companies because they built network cables. They built applications.

Artificial intelligence may be entering a similar phase. If language models become diverse, accessible, and increasingly cheap, the competitive differentiator ceases to be owning the most powerful model. It becomes knowing how to use it to solve real problems.

Where is the value in artificial intelligence?

Distributing advanced models does not mean giving up revenue. A company can make weights available for free and monetize infrastructure, managed services, enterprise support, cloud, security, or development tools.

In addition, there is another benefit. The more developers use a certain model, the larger the ecosystem created around it tends to be.

Tools, integrations, libraries, and applications begin to emerge naturally. In practice, that model can become a market standard.

This logic helps explain why giants like Microsoft, Nvidia, IBM, and Meta have been advocating for the United States to maintain room for open-weight models, while companies heavily reliant on closed models adopt a more cautious stance.

The strategy presented by Xi Jinping does not guarantee that China will dominate artificial intelligence. Nor does it mean that closed models will cease to exist. What it reveals is an important shift in how this market is starting to compete.

During the early years of generative AI, attention was focused on model capabilities. Now, price, ecosystem openness, ease of integration, and freedom for developers have begun to influence the race.

In the coming months, the main question may not be which lab built the smartest AI. It will be interesting to see which of them can convince more companies to build on top of their technology.