The AI Commoditization Wave: Why Value is Shifting from Models to Applications
As open-source alternatives like Zhipu's GLM drive down AI costs, developers are entering an era of model abundance where practical application matters more than raw benchmark scores.
The launch of the open-source General Language Model (GLM) by Beijing-based artificial intelligence firm Zhipu has reignited a crucial debate in the technology sector. By competing directly with major Western models like OpenAI’s ChatGPT and Anthropic’s Claude at a tiny fraction of the price, Zhipu is highlighting a familiar economic pattern. In previous decades, Western innovations in solar panels and electric vehicles were rapidly transformed by Chinese manufacturing into cheap, abundant commodities. Now, Western tech giants like Microsoft and NVIDIA, whose high market valuations rely on the assumed scarcity and high cost of advanced AI models, face a highly disruptive economic shift.
The economics of state-backed abundance
This shift is driven by a strategy previously observed in other industrial sectors. Historically, Western scientific breakthroughs have been met with Chinese pricing strategies financed by state-owned banks and directed by regional party committees. These state-backed entities often operate on a negative net present value strategy, selling products below their actual production costs. As economists Andrei Shleifer and Robert Vishny have demonstrated, when politicians rather than private shareholders direct companies, the primary objectives shift from financial returns to political goals such as employment, national prestige, and strategic dominance.
While private laboratories like OpenAI and Anthropic must eventually turn a profit to survive—especially as they prepare for initial public offerings (IPOs) amid currently negative operating profits—state-supported competitors are indifferent to financial losses. This dynamic drives down prices, creates an abundance of comparable alternative models, and erodes the profit margins of the original innovators. Even if American laboratories maintain their lead on standardized benchmarks like GLUE or MMLU, the market default is often decided by cost rather than minor performance margins. Affordable, "good-enough" models quickly become the foundation upon which global software is built.
The limits of static intelligence
To understand whether AI will inevitably follow the commoditization curve of solar panels, it is necessary to examine how these models function. Technologies like batteries and solar cells follow predictable cost-reduction curves similar to Moore's law, where efficiency rises as unit costs fall. Current AI architectures may face a similar destiny due to their static nature. Unlike the human brain, which rewires itself during everyday interactions, a trained AI model is a closed system. It learns only during its training phase; to improve its cognitive capabilities, it must be rebuilt and retrained from scratch using more data, chips, and electricity.
This limitation mirrors Kurt Gödel’s incompleteness theorem, which posits that a closed system of rules cannot prove itself from within. Because today's models cannot expand their cognitive abilities using their existing resources, they remain subject to a decay curve. Unless artificial intelligence transitions to an evolutionary architecture—where models continuously learn and adapt as they process new information—the competitive edge of owning a static model will continually diminish.
What it means for developers
For software engineers and system architects, this shift in AI economics marks a transition from model scarcity to model abundance. As the underlying models become cheaper and more interchangeable, the long-term value of AI shifts away from the raw models themselves and toward the applications built on top of them. Real-world implementation is already proving this theory. In China, companies are focusing on practical deployment rather than marginal benchmark gains: Meituan has deployed automated drone deliveries, Baidu operates driverless taxis, and Alibaba utilizes delivery robots. In these cases, commercial success comes from efficient usage and integration rather than owning the smartest static model.
To capitalize on this transition, developers should focus on creating unique workflows, proprietary distribution channels, and highly integrated products that competitors cannot easily replicate. As model access becomes increasingly commoditized, developers can try top AI models cheaply through one API at https://apixoai.online, giving them the flexibility to test and deploy various systems without committing to a single expensive provider. Ultimately, the future of AI development lies not in training massive, resource-heavy models for minor benchmark improvements, but in leveraging affordable, abundant intelligence to solve practical problems in fields like medical screening, traffic control, and energy grid management.
Source: Made in China AI: new tech, same old story? — The Conversation AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
