Since the advent of large AI models, the field of artificial intelligence has been characterized by significant financial investment. This trend has been particularly beneficial for hardware providers like NVIDIA, whose revenues soared as they capitalized on the demand for AI development tools, briefly making it the most valuable company in the world. Recently, several Wall Street firms, including Goldman Sachs and Sequoia, have released reports questioning the sustainability of the AI "gold rush," arguing that substantial investments—amounting to billions of dollars—need justifiable profit returns.
According to U.S. media reports, OpenAI anticipates a massive loss of nearly $5 billion in 2024, with operating costs expected to reach $8.5 billion, leading to cash flow depletion within a year. If even high-profile companies are facing such losses, the industry may need to reassess its approach.
Another emerging trend is that the fundamental capabilities of large models are hitting bottlenecks. Progress has slowed in developing the next generation of products with substantial improvements in factual understanding and logical reasoning. Current advancements are primarily focused on multi-modal functionality, such as video generation, and on improving the efficiency of training and operation through architectural simplifications. While the potential of AI remains significant, the enthusiasm for "burning cash" without clear direction has noticeably waned, with an increasing emphasis on economic viability.
The global development trends in large AI models indicate a shift toward application development, practical implementation, and cost reduction. A handful of large enterprises will continue to invest strategically to enhance the foundational capabilities of their models, striving to remain at the forefront. However, only a few companies are expected to invest heavily in exploring the unknown realms of general artificial intelligence.
With the U.S. government imposing restrictions, some Chinese companies are starting to consider a more adaptive approach to R&D. This strategy is quite viable, given China's substantial research and market scale. Even if there are technological breakthroughs in the U.S., China can quickly catch up and eventually leverage its advantages in application and cost to surpass competitors. Encouragingly, Chinese AI R&D is now shifting to focus more on economic factors, aiming for practical outcomes and industry integration rather than succumbing to anxiety over racing against the U.S.
Looking ahead, the industry needs to align with national conditions and policy directions to avoid falling into harmful competition. Currently, some sectors experience intense cost competition, even with well-established technological capabilities and services pushing companies to cut costs in a bid for market superiority. From a national perspective, market competition should be maintained within reasonable bounds to promote growth, foster effective resource allocation, and prevent the destructive competition that leads to waste.
Utilizing cloud service platforms is a popular approach for Chinese companies developing large models, enabling smaller firms to bypass the need for massive GPU clusters. Beyond providing computational power, cloud providers can also integrate various open-source large models, with pricing based on token usage—where a token represents the smallest unit of text processed. Rates for inputting tokens can vary significantly, often from under ten cents to more than a dime, depending on the model's size.
These accessible cloud services have effectively lowered the barriers to entry for large model development, allowing many companies to customize services to drive their AI innovations. Recent news reports that Alibaba Cloud has raised prices for its domain auction platform—from 69 yuan to 99 or 199 yuan— to cope with rising service costs highlight an important industry trend. Previously, due to market share competition and focus on tangible values, China had undervalued software and service assessments.
The integration of specialized applications like "domain squatting" and "large model training" has significantly facilitated market participation through cloud platforms, allowing stakeholders to concentrate on leveraging their strengths for market applications. Practical evidence shows that cloud platform services are essential for advancing AI development in China. To achieve a flourishing landscape for AI growth and its related applications, both the active participation of numerous small and medium enterprises in AI development and the provision of high-quality services by cloud platform companies are crucial.
Ranking
