Zhang Hongjiang of Source Code Capital: AI Enters the "Industrial Scale" Era

```

Author | Chai Xuchen

Editor | Zhou Zhiyu

AI’s growth continues to stride forward amidst differences.

On September 11, at the 2025 Inclusion·Bund Conference, Source Code Capital investment partner and foreign member of the U.S. National Academy of Engineering, Zhang Hongjiang, shared his thoughts on large language models, AI agents, and the agent economy. He said that the Scaling Law is still valid, and agents together with economic structural transformation will profoundly reshape society.

The Scaling Law is regarded as the first principle for the improvement of large model performance: the larger the parameters, the better the performance. Though the industry believes the scaling law for pre-trained models is gradually slowing, Zhang Hongjiang specifically pointed out that the emergence of inference models has created a new curve for large-scale development—a “scaling law of inference.” He said, "The computational demands for aspects like context and memory will continue to drive scaling up.”

Additionally, Zhang Hongjiang noted that “large language model deflation (LLM flation)” will further reinforce the Scaling Law. The price per token for large models has dropped rapidly over the past three years; as performance continues to improve, usage costs will keep falling.

In terms of AI industrialization, Zhang Hongjiang pointed out that AI is driving massive expansion of infrastructure. He cited projects like OpenAI’s Stargate and Musk’s Colossus—where the power consumption of a million GPUs amounts to one-eighth of Beijing’s electricity usage. By 2025, major U.S. tech companies’ AI-related capital expenditures are expected to exceed $300 billion. The entire AI data center industry chain has undergone a large-scale construction boom over the past year.

In Zhang Hongjiang’s view, large-scale construction in the IDC industry will stimulate the electricity ecosystem and drive economic development; this is the essence of "AI industrial scale-up."

Zhang Hongjiang concluded that humanity is entering the "agent swarm" era—that is, vast numbers of agents interact, execute tasks, exchange data, exchange information, and even exchange tasks with one another. Human interactions with these agent swarms will constitute the so-called “agent economy.”

He stated that models and GPU computing power will become the core assets of future organizations. "Enterprises need to expand computing power to strengthen models and enrich data." He pointed out that agents will reshape enterprise processes, and that "super individuals + agents" will bring tremendous structural transformation.

Risk Disclosure and DisclaimerThe market has risks, investment requires caution. This article does not constitute individual investment advice and has not considered the specific investment objectives, financial situations, or needs of particular users. Users should consider whether any opinions, views, or conclusions in this article are applicable to their own circumstances. Invest at your own risk. ```