The emergence of Chinese AI models such as Kimi K3 and Qwen3.8 Max has reignited a debate that goes far beyond mere technology. To understand whether they represent a real threat or a strategic opportunity, we must examine the economic and technical dynamics behind them. The key lies not so much in raw model capability, but in how their availability changes the marginal costs of AI and, consequently, the foundations of digital businesses.
For years, the dominant software paradigm relied on near-zero marginal costs: once an application was developed, distributing it required little additional resources. Generative AI, and particularly inference on large language models, has brought back a concept that seemed forgotten: cost of goods sold (COGS). Every query to a model consumes computation and therefore money. Kimi K3, for example, charges $3 per million input tokens and $15 per million output tokens. Anthropic’s Fable charges $5 and $30 respectively. The difference seems enormous, but reality is more complex: the number of tokens needed to reach a correct answer varies hugely between models. A less efficient model may consume more tokens, effectively making it more expensive.
In this scenario, intelligence begins to behave like a commodity. When two models solve the same task with identical results, the differentiating factor is not quality but production cost. Variables such as model footprint, inference efficiency (e.g., through Mixture-of-Experts architectures), memory cache usage, and the ability to serve multiple concurrent requests come into play. The player with the lowest cost structure will have a decisive advantage in a market where the selling price is set by the least efficient supplier.
Today, however, demand for frontier models far exceeds supply, allowing labs like Anthropic or OpenAI to maintain high margins. But this situation is temporary. As computing capacity expands, the price of intelligence will tend to equalise downward. Chinese models, being open weights, accelerate that downward pressure. For Western companies, this poses an immediate challenge: it is no longer enough to have the best model; you must be able to serve it at the lowest possible cost.
From a business perspective, this evolution has profound implications. Companies that build custom software must rethink how they integrate AI into their solutions. The availability of competitive, free (in R&D cost) models lowers entry barriers, but demands solid expertise in deployment and optimisation. This is where technical know-how makes the difference. It is not just about choosing a model, but about designing an architecture that minimises inference costs and maximises efficiency.
In parallel, the commoditisation of intelligence opens opportunities in fields like cybersecurity. As a recent incident at Hugging Face showed, defenders turned to an open-source Chinese model (GLM 5.2) because proprietary US models had barriers preventing their use in incident response environments. If defenders cannot access the best models due to political or licensing restrictions, they are forced to seek alternatives, often from countries with divergent interests. For businesses that need to protect their systems, having a technology partner capable of integrating AI models into their cybersecurity protocols becomes critical. The question is not whether Chinese models are safe, but whether we can afford not to have access to the best available intelligence.
In business intelligence, generative AI is transforming how data is analysed. BI tools like Power BI can benefit from language models that automate report generation or trend interpretation. Chinese models, being cheaper, make this integration viable for a larger number of organisations. However, the true competitive advantage is not in the model itself, but in the ability to combine it with a solid cloud infrastructure, whether on AWS or Azure, that guarantees scalability and security. Here, cloud AWS/Azure services provide the ideal environment to deploy these models with control over costs and latency.
The emergence of AI agents executing complex workflows adds another layer of complexity. A poorly optimised agent can consume thousands of tokens per task, skyrocketing costs. Chinese models, with their focus on token efficiency, may be especially suitable for tasks requiring chain-of-thought reasoning. For companies looking to automate processes, the choice of underlying model conditions the economic viability of the project. Here, technical expertise to evaluate and compare alternatives is indispensable.
Beyond economic analysis, there is a geopolitical dimension. China has adopted a strategy of “commoditise your complements”: it promotes open models so that AI becomes cheap and universally accessible, benefiting its robust industrial and robotic sector. At the same time, it weakens US labs by forcing a price war. Distillation (using frontier models to train smaller models) is a common practice that, although controversial, accelerates innovation. The problem is not distillation itself, but the dependence on providers operating under different rules. For Western companies, the solution is to push for a legal framework that allows unrestricted distillation and fosters an open-weight ecosystem of their own.
In this context, the question “threat or opportunity?” admits a nuanced answer. For companies that merely consume expensive closed models, the Chinese disruption will be a threat because it erodes their margins. For those that know how to leverage open models to build differentiated solutions, optimising costs and scaling with cloud infrastructure, it will be a unique opportunity. The key is not to confuse the model with the product: the real value lies in integration, customisation, and service. Companies that master these areas, supported by technology partners with expertise in AI, cloud and cybersecurity, will be best positioned to turn the perfect storm into a tailwind.
