The question animating much of the current AI investment debate is not whether artificial intelligence will transform the economy. It is who captures the value when it does. In a new commentary published by Janus Henderson Investors (October 19, 2026)1, Research Analyst Shaon Baqui offers a clear-eyed framework for thinking through the open-versus-closed model divide and what it implies for the entire AI infrastructure stack.
Two Models, One Accelerating Market
The distinction Baqui draws is foundational. Closed models, he explains, are "proprietary, off-the-shelf models where the code and the weights are largely fixed" -- ChatGPT, Gemini, Claude. Open models, by contrast, are "highly customizable in the sense that you can download them on your own server, you can modify them, and really fine-tune them to your particular workload" -- Llama, Qwen, DeepSeek.
Historically, open models traded performance for price. That tradeoff is narrowing fast. Baqui notes that "in many cases, open models are quite good at running some of your standard workflows like search, summarization, and office productivity." The performance gap, while not closed, is no longer the decisive barrier it once was. What has changed is enterprise intent. In recent months, corporations have moved quickly toward open models to gain control and customization while simultaneously working to "bring down the cost of inference and prevent vendor lock-in into one or two sort of frontier models."
Cheap AI Is Expansionary, Not Deflationary
The counterintuitive insight at the centre of Baqui's analysis is worth dwelling on. Falling costs of AI inference are not a negative for the sector broadly. Quite the opposite. As he puts it, "cheap AI is a huge positive." By reducing the cost per token, enterprises run more agents, automate more workflows, and generate exponentially more tokens overall. The volume effect swamps the price compression. The beneficiaries are broad: "compute, networking, storage, and memory" all stand to gain from a world where AI becomes cheap enough to deploy at scale across standard enterprise operations.
This is not a novel dynamic in technology. Lower costs of computing historically expanded demand rather than contracting revenue. The same logic applies here, and Baqui's framing makes the investment implication explicit: infrastructure spending is tied not to the price of AI but to the volume of AI activity. As open models unlock adoption by enterprises that could not justify the cost of frontier model inference, aggregate AI activity accelerates.
Hyperscalers: Value-Agnostic by Design
One of the more durable investment observations Baqui makes concerns the hyperscalers. Their structural position as hosts of open models makes them, in effect, model-agnostic intermediaries. "By acting as the gatekeepers, these guys host these open models, and they can route workflows to the relevant model," he observes. "They can add value potentially no matter which model wins."
This is an important framing for investors anxious about model competition displacing incumbents. The hyperscalers are not competing in the model layer in the same way AI labs are. They are the distribution infrastructure. Whether an enterprise routes a legal review to a frontier closed model or a productivity task to an open model, the workflow runs on hyperscaler infrastructure. The competitive moat is in the plumbing, not the intelligence.
The AI Labs Are Not Standing Still
The concern that open models will commoditize the model layer and squeeze AI lab economics has some surface plausibility. Baqui addresses it directly and dismisses it with data. While the majority of token volume flows through open models, "the majority of token revenue is actually being generated by closed or frontier-level models." Volume and revenue do not move together here, and the distinction matters enormously for assessing lab economics.
Furthermore, Baqui notes that labs are responding competitively. "Some of the latest low- to mid-range offerings such as GPT, Terra, Luna, Opus 5.5, are quite compelling in terms of price and performance versus their open-model counterparts." The labs are not ceding the mid-market. They are repricing into it.
Five Key Takeaways for Advisors and Investors
- Cheaper AI drives more AI, not less infrastructure spending. Falling inference costs expand enterprise adoption, generating more tokens, more agents, and more workflow automation. The net effect on AI infrastructure demand is positive across compute, networking, storage, and memory.
- Hyperscalers hold a structurally advantaged position. As hosts and routers of both open and closed models, they capture value regardless of which model layer wins. Their role as gatekeepers insulates them from model-level competition.
- AI lab revenue is more durable than token volume suggests. The bulk of token revenue still flows to closed frontier models despite open models commanding the majority of volume. Price compression at the volume tier does not translate directly into lab revenue pressure.
- The market will not be winner-takes-all. Sophisticated, high-stakes workloads such as scientific research, advanced coding, and legal review will continue to command frontier model pricing. Commodity workflows will route to open models. The AI ecosystem is stratifying, not collapsing.
- The evolving open-versus-closed mix is the variable to watch. As Baqui frames it, "the key question isn't whether open models will cannibalize closed models; it's just how that evolving mix of open versus closed really changes the equation on value creation in the overall AI ecosystem." Investors who track that mix have an analytical edge.
October 19, 2026
Footnote:
Baqui, Shaon. "Who Benefits as Open Models Make AI Cheaper to Run?" Janus Henderson Investors, 19 Oct. 2026, https://www.janushenderson.com/en-us/advisor/article/who-benefits-as-open-models-make-ai-cheaper-to-run/.
