The ability to distill the capabilities of a frontier AI model into a smaller, cheaper one is the main factor preventing an oligopoly. Even if a lab's model is continuously improving, competitors can continually distill its public-facing behaviors, ensuring that no single company can maintain an insurmountable lead.

As more of the internet and code repositories are generated by leading AI models, any new model trained on this public data inadvertently "distills" the knowledge and quirks of those proprietary systems. This blurs the line between original training and outright copying.

Enterprise software choices may be dictated by which platforms AI labs use for reinforcement learning. Meta reportedly switched to Slack because frontier AI models are trained in it, making them more effective agents within that specific environment. This creates a powerful, emergent moat for incumbent software that becomes a default AI training ground.

A niche, services-heavy market has emerged where startups build bespoke, high-fidelity simulation environments for large AI labs. These deals command at least seven-figure price tags and are critical for training next-generation agentic models, despite the customer base being only a few major labs.

Fears of a single AI company achieving runaway dominance are proving unfounded, as the number of frontier models has tripled in a year. Newcomers can use techniques like synthetic data generation to effectively "drink the milkshake" of incumbents, reverse-engineering their intelligence at lower costs.

As reinforcement learning (RL) techniques mature, the core challenge shifts from the algorithm to the problem definition. The competitive moat for AI companies will be their ability to create high-fidelity environments and benchmarks that accurately represent complex, real-world tasks, effectively teaching the AI what matters.

Unable to build frontier models from scratch, some Chinese companies gain a competitive edge by using "scale distillation." This involves training smaller, open models on the outputs of larger, proprietary US models, effectively piggybacking on American R&D to create capable, low-cost alternatives.

Instead of just copying outputs for supervised fine-tuning, Chinese labs use frontier US models as automated evaluators in their reinforcement learning loops. This allows their own models to develop capabilities within their native distributions and potentially surpass the teacher model.

The key to creating frontier AI models is no longer just pre-training data or distilling from other models. The real differentiator is building superior interactive environments for reinforcement learning. Labs that create the best environments for specific tasks (e.g., front-end coding) can generate unique improvement loops, leading to state-of-the-art performance.