Chinese AI models are getting better and cheaper and Australian businesses are taking notice

The models created by Chinese AI have been rapidly improving and are much more affordable to run. The transition is starting to put pressure on the reigning dominance of U.S. tech giants in the AI industry, and it is making businesses around the world rethink their AI strategies.

The Tech World is on the move!

At first, DeepSeek was a surprise and sparked doubts as to whether China can achieve state-of-the-art AI at a lower price. But today, DeepSeek is not the only one. Several models have been released by developers such as Z.ai (formerly Zhipu AI), Moonshot AI that are gaining traction around the world. In particular, Moonshot AI's Kimi K3 has been the highlight of the attention and even some praise from industry experts and a scrum of criticism from Washington.

This growing competition has led to volatility in global markets. Tech-heavy indexes have been correcting as investors wonder if the trillions already invested in AI infrastructure in the United States will be rewarded by enough returns when given to cheaper and more powerful Chinese alternatives.

Australian businesses are embracing ‘Model Routing’.

This intense competition is at its core reshaping the economic paradigm of technology for Australian startups and enterprises adopting AI solutions. Many companies are moving towards not just relying on costly, closed source models from the major developers, in the U.S.

However, the new trend is to adopt the so-called "open-weight" models, many of which have also been created in China and can be downloaded and hosted by businesses. Unlike proprietary developers, which charge a fee every time the model is used, self-hosting removes these token fees, meaning businesses are only responsible for the computational costs needed to operate the models.

With this change has come a new approach called “model routing.” They have the most powerful and expensive business-centric models reserved for more complex, high-level reasoning, and the lower-cost, self-hosted ones for simpler, more repetitive workloads. This combination could reportedly lead to 80% savings in token consumption costs, making AI automation much more scalable.

Security and Trust Concerns

There are obvious economic advantages, but there are some obstacles to adoption. However, Chinese models can be problematic, experts say, especially when dealing with issues that are politically charged. Additionally, there is still regulatory uncertainty and a lack of trust for industries where sensitive information is involved, including healthcare and finance.

Users of AI models that are locally installed should be aware that they all come with risks, cybersecurity experts warn. To counteract these risks, companies going for self-hosted solutions should allocate significant resources to ensuring that they have strong security measures in place and that they are leveraging the trusted local cloud.