AI’s Three-Body Problem: no single force can dictate the outcome

In the AI economy of 2026, the system is likened to three celestial bodies orbiting one another, each exerting a significant influence on the others. These bodies consist of closed-source frontier labs, such as OpenAI and Anthropic, open-weight models primarily from China, and application companies built on top of both. Each of these entities is powerful enough to reshape the others' trajectories, but none can dictate the ultimate outcome.

Recent events have heightened the instability within this system, including the surge in demand and revenue growth experienced by the frontier labs, leading to a greater focus on demonstrating return on investment and finding more cost-effective alternatives. This shift in spending habits has prompted concerns among enterprises, as some are overspending on AI tokens without proportional productivity gains.

Simultaneously, competition among the frontier labs has intensified, with Meta and xAI releasing increasingly capable models. Meanwhile, Chinese companies have made strides in open-weight models, with Zhipu's GLM 5.2 and Moonshot's Kimi K3 matching or surpassing the performance of frontier models, all while being priced affordably.

U.S.-based open-weight models, such as Thinking Machines' Inkling and Nvidia's Nemotron 3, are also gaining traction, offering domestic alternatives to Chinese releases while providing customers with greater control and lower costs. As these developments unfold, some broad trajectories can be identified for the second half of 2026.

Firstly, concerns over frontier pricing are expected to dissipate as competition drives down prices and returns on AI investment begin to demonstrate merit. Secondly, the emergence of a multi-model world driven by competition will likely lead to the development of distinct capabilities for each model. Finally, U.S.-based open-weight models will become genuine alternatives to their Chinese counterparts, garnering real-world adoption and establishing clearer business models.

Ultimately, the distinction between open and closed models may diminish as frontier labs integrate personalized models for specific customer needs, and application companies deepen their engagement with the model stack. While the competition and shifting dynamics may appear temporary, the true question remains: who will capture the value generated by AI's growth – the frontier labs, open models, or application companies that own the customer base?