A new industry analysis argues that the open and closed AI model ecosystems are no longer competing on the same terms. Instead, they are evolving on separate growth curves, with the closed frontier labs capturing premium demand at the top end of the market and the open model stack spreading more broadly across enterprises and infrastructure providers.
The argument, laid out by Interconnects AI writer Nathan Lambert, centers on economics as much as technology. Lambert says the key question is whether users will keep paying large premiums for the most advanced proprietary models. He points to coding agents as the first major use case where many customers appear willing to pay more for better performance, especially when the tools clearly raise productivity on complex knowledge work.
At the same time, Lambert says frontier labs are likely to respond by keeping their best models more tightly controlled. That could mean slower API releases for top systems, along with more emphasis on protecting compute capacity, limiting distillation, and serving only higher-margin use cases. In his view, those choices would help preserve the value of closed models while also narrowing what gets exposed to developers through APIs.
The analysis describes Anthropic and OpenAI as the leading examples of this closed frontier group, with Google expected to be a stronger contender over time. Lambert argues that these companies have advantages in integration, including model weights, tools, serving infrastructure and product design. He says that combination allows them to optimize not just raw benchmark scores, but the utility users get per second or per watt.
He also argues that the most valuable customers will continue to pay for the strongest models, even as some workflows remain imperfect. In his view, the returns from using the best tools are high enough that many professionals will not settle for cheaper alternatives if the top systems materially improve output.
Lambert draws a contrasting picture for open models, which he says are likely to take longer to mature but ultimately reach a wider set of users and businesses. He argues that many companies want to adopt open models, but the systems are still not consistently strong enough for unusual or out-of-distribution tasks. Over time, he expects open model builders to stop trying to match frontier proprietary models on the same benchmarks and instead focus on tasks where lower price points and flexibility matter more.
According to the analysis, the open ecosystem is structurally different because it is less integrated. Multiple companies can contribute to hosting, tuning, routing and deployment, which pushes pricing toward commodity levels. That creates room for enterprises to build in-house agents and tools around models that are good enough for specific tasks and relatively cheap to run.
Lambert also points to the growing open model fine-tuning ecosystem, citing companies such as Tinker, Fireworks and Prime Intellect as examples of a stack that could make custom deployment easier and widen adoption.
He expects that as open model usage expands, a larger share of inference will run through cloud platforms and AI infrastructure companies rather than through the leading proprietary labs. That, he says, would reflect a wider diffusion of AI across the economy rather than concentrated value capture at a handful of companies.
The analysis rejects the idea that recursive self-improvement would necessarily give closed labs an unbeatable lead. Lambert says progress should continue quickly across the sector, but not in a way that locks in one ecosystem permanently. In his view, closed models are finding product-market fit at the high end first, while open models will build value more slowly across a broader base.
His conclusion is that the two ecosystems are not just competing, they are compounding differently. Closed models, he argues, are following an integrated path that rewards premium intelligence. Open models are following a wider and more distributed path that could eventually touch far more of the market.