Over the past two years, the artificial intelligence race has been easy to score: bigger models, better benchmarks, and which company could take the lead, at least until the next launch.
This scorecard is starting to look incomplete.
As companies move from testing AI to deploying it in real products and workflows, it is no longer about choosing the best model, but rather accessing the model that is best suited for a given task, at the right cost, with the necessary data, and in a chosen environment.
This shift opens the door to a new kind of AI competition that focuses less on model size and more on routing, cost, control and computing power.
“The model alone is no longer the product,” Perplexity CEO Aravind Srinivas told CNBC. “It’s the harness, the orchestration system, that puts the model into a very powerful harness and connects the model to a lot of tools.”
This means that AI products will become systems that can decide which model to use, when, and which external tools or corporate data sources are required. A customer service job may not require the most expensive model. A complex coding problem could be. A routine internal workflow could run on a cheaper open model. A harder step could be escalated into a stronger step.
“The answer is: always use what is best for the task,” Srinivas said.
The emergence of alternative models comes at a time when corporate America is tightening its belt on AI spending, posing a further challenge to OpenAI and Anthropic, which have thrived in recent years by selling cutting-edge technology.
Aravind Srinivas, CEO of Perplexity AI.
CNBC
Perplexity this week introduced a new system for its computing product based on GLM 5.2, an open model from Chinese manufacturer Z.ai. The system is designed so that a cheaper model can do more work, while a more powerful model is only called in when necessary.
This approach reflects a broader shift in the market. Open-weight models that can be downloaded, customized and operated by companies themselves are becoming increasingly powerful. They are also less expensive to operate than proprietary premium models from the largest AI labs.
Benchmark general partner Peter Fenton said the shift could be dramatic.
“A potentially contrarian view that is emerging from consensus is our belief that over 90 percent of tokens created over the next 18 to 24 months, perhaps even by year-end, will come from open-weight models,” Fenton told CNBC.
Tokens are the data units that AI models process and generate.
“The inference margins generated by the frontier model companies will, I think, come under pressure if you can operate them without the markup that they provide if you have good enough models from open weights,” Fenton said.
Fenton said the move to open models isn’t just about saving money. In some cases, smaller models tailored to a specific task can be faster and perform better than larger, general-purpose models.
Table of Contents
“Where it works and how it works”
That’s one of the reasons Benchmark invested in Ollama, a company that makes it easier for developers and enterprises to download, run and manage open models.
“One thing is where the model comes from and where it was created and trained,” said Jeff Morgan, CEO of Ollama. “But the most important thing for the companies we talk to is where and how it works.”
Morgan said Ollama has been adopted by more than 85% of Fortune 500 companies, including companies in regulated industries such as aviation, insurance and healthcare. He said many companies start with smaller models that run close to their own data and then expand to larger open models as they become more comfortable with them.
The rise of open models also poses a strategic challenge for the US. Many of the most competitive open models come from Chinese labs, including Z.ai and DeepSeek. This has turned open source AI into a business problem, a political problem, and a national competition problem.
Srinivas said the US should support open models because they make AI more affordable and accessible.
“If you want the benefits of AI to be available to small businesses in America and U.S.-allied countries, then AI really needs to be much more affordable,” Srinivas said. “And open source is the only way to achieve that.”
The shift could also impact the massive expansion of data centers across the tech industry. The current AI boom assumes demand will continue to flow to large cloud data centers with high-end chips. Srinivas says some AI work may instead be done locally, on devices owned by consumers or businesses.
That wouldn’t eliminate the need for data centers, but it could create a more hybrid AI system, with routine tasks performed locally and the most difficult work sent to a more powerful model in the cloud.
The question for investors is whether the largest AI labs can maintain their pricing power as open models improve and companies become more selective about their use.
REGARD: OpenAI’s Sam Altman says Chinese open source models are getting very good
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
https://www.cnbc.com/2026/07/10/the-ai-race-is-shifting-from-bigger-models-to-cheaper-smarter-systems.html
