The recent drastic price reductions in China's LLM market have ignited a fervent debate on the feasibility of AI model monetization. The challenges of LLM monetization are a global concern, with Western companies engaging in price wars for over a year. However, the divergence lies not in the issues LLM-makers are grappling with or their competitive strategies, but in the contrasting reactions of investors and the general media.
Superficially, major Chinese players like Alibaba, Baidu, and Tencent have slashed prices on their LLMs by up to 99% this week. The forms the announcements took have caused intense focus on whether GenAI can ever be value-accretive with investors of these companies. Given these stocks’ recent histories, their investors and analysts are far more conservative than their US counterparts. This was in full display in recent result calls, too: a Baidu-like result call focus on GenAI would have caused an uproar in the US. However, in the Chinese analyst community, it only generated more concerns about its future ROI. Analysts in China are more preoccupied with immediate financial metrics and have little patience for investments – both in infrastructure and customer acquisitions – required to increase the usage and muscle out the smaller players for long-term success.
Effectively, GenAI capex is celebrated in the West and not so in the East.
Let’s compare this with what is going on in the US. OpenAI, valued at an estimated $80-100 billion, boasts less than a million paid subscribers for ChatGPT Plus, generating roughly $240 million in annual revenue. Other major players like Google and X are bundling their premium LLM usages with existing services without raising prices and with little indication of a meaningful subscriber base increase for those services. Microsoft and Antropic’s reluctance to discuss numbers for subscribers strongly points to the triviality of the numbers and falling growth rates. The situation is likely worse for the rest.
Moreover, open-source and closed-source models increasingly offer free access and incentives like credits to attract users and boost adoption. OpenAI, in the latest salvo, offered GPT-4o free to everyone, almost cannibalizing its paid subscriber segment to boost usage.
As we have been highlighting for months, the sustainability of a. compute on cloud and b. inability of the XaaS model to generate sufficient revenues is questionable. The future of AI-era monetization is shifting towards hardware sales. Microsoft's AI-enabled PCs, Google's integration of DeepMind with Pixel, and the growing focus on AI-powered devices in China all point to this evolving trend.
In popularising AI today, many of these companies are paving the path for their other product sales over time. This strategy could be value-destroying for those who do not see the need to pivot towards hardware, like many of the US giants.
