The artificial intelligence industry is experiencing a significant paradigm shift, moving away from a competition based solely on developing the largest AI models towards optimizing for cost-effectiveness, smart system orchestration, and task-specific efficiency. Companies like Perplexity are building “harnesses” that route tasks to the most appropriate AI model, whether open-source or proprietary, based on specific needs and budget constraints. This trend, supported by the growing capability of open-weight models, is expected to put economic pressure on major AI providers and could lead to a more hybrid AI infrastructure blending cloud and local computing.
For the past two years, the competition in artificial intelligence has largely focused on who could build the biggest models and achieve the best benchmark scores. However, this simple scorecard is rapidly becoming outdated as the industry evolves.
As businesses transition from merely testing AI to integrating it into live products and daily workflows, the emphasis is shifting. It's no longer about finding the single "best" model, but rather identifying the most suitable one for a particular task, considering factors like cost-effectiveness, data requirements, and deployment environment. This evolution is sparking a new phase of AI competition, moving beyond sheer model size to focus on intelligent routing, optimized costs, greater control, and efficient compute resources.
"The model alone is no longer the product," explained Aravind Srinivas, CEO of Perplexity. "It is the harness, the orchestration system that puts the model inside a very capable harness and pairs the model with a lot of tools."
This means modern AI products are becoming sophisticated systems capable of autonomously determining which AI model to employ, when to activate it, and what external tools or proprietary company data sources are necessary. For instance, a routine customer service inquiry might not warrant an expensive, high-capacity model, whereas a complex coding challenge likely would. A simpler internal task could run efficiently on an open-source model, with more demanding steps escalated to a powerful, proprietary solution.
"The answer is always use whatever is the best for the task," Srinivas reiterated.
This emergence of diverse model options arrives at a time when corporate America is increasingly scrutinizing its AI expenditures. This trend poses a significant challenge for industry leaders like OpenAI and Anthropic, who have thrived by offering cutting-edge, often premium, AI technology.
Perplexity recently demonstrated this shift by previewing a new system for its computer-use product, leveraging GLM 5.2, an open-source model developed by China's Z.ai. This innovative system is designed to allow a more economical model to handle the majority of tasks, only engaging a more robust model when absolutely essential.
This strategic approach mirrors a wider market transformation. Open-weight models, which can be downloaded, customized, and operated directly by companies, are rapidly enhancing their capabilities. Crucially, they also offer a more cost-effective alternative to the premium proprietary models from leading AI research labs.
Peter Fenton, General Partner at Benchmark, believes this shift could be profound.
"A maybe contrarian view that is becoming consensus is our belief that 90-plus percent of the tokens created will come out of open-weight models over the next 18 to 24 months, possibly even by the end of the year," Fenton shared with CNBC. Tokens are the fundamental units of data that AI models process and generate.
"The inference margins generated by the frontier model companies, I think, are going to come under pressure when you can run those without the markup that they're providing, when you have good enough models from open weights," Fenton added.
Fenton emphasized that the move towards open models isn't solely driven by cost savings. In many instances, smaller models meticulously fine-tuned for a specific application can outperform larger, general-purpose models in both speed and accuracy.
'Where it runs and how it runs'
This rationale underpinned Benchmark's investment in Ollama, a company dedicated to simplifying the process for developers and enterprises to download, operate, and manage open models.
"One thing is where the model's from and where it was created and trained," stated Jeff Morgan, CEO of Ollama. "But the more important thing to these businesses we speak to is where it runs and how it runs."
Morgan reported that Ollama has been adopted by over 85% of Fortune 500 companies, including those in heavily regulated sectors like aviation, insurance, and health care. He noted that many organizations typically begin with smaller, locally run models, close to their data, before scaling up to larger open models as their comfort and confidence grow.
The ascendancy of open models also introduces a significant strategic challenge for the U.S. Many of the most competitive open-weight models are originating from Chinese laboratories, including Z.ai and DeepSeek. This development elevates open-source AI from a business concern to a critical policy issue and a matter of national competitiveness.
Srinivas advocates for U.S. support of open models, arguing they democratize AI by making it more affordable and accessible.
"If you want the benefits of AI to be widely distributed to small businesses in America and American allied countries, then you really need AI to be a lot more affordable," Srinivas asserted. "And open source is the only way to do that."
This paradigm shift could also influence the massive data center expansion currently underway across the tech industry. The prevailing AI boom assumes a continuous flow of demand towards large cloud data centers equipped with high-end chips. However, Srinivas suggests that some AI tasks might eventually execute locally on consumer or business devices.
While this wouldn't negate the need for data centers, it could foster a more hybrid AI ecosystem, where routine operations are handled locally, and only the most complex computational challenges are routed to powerful cloud-based models.
For investors, a key question remains: can the largest AI labs sustain their premium pricing power as open models continue to improve and companies become increasingly discerning about their AI solutions?
Watch: OpenAI's Sam Altman says Chinese open source models are getting very good
