The companies that can provide the most economic value from the power their AI uses will ultimately command the highest valuations, Perplexity CEO Aravind Srinivas told CNBC on Wednesday.
Srinivas said that whichever company can provide the “most token value per watt per user” will be the winner in the future.
“Whoever is able to maximize this particular objective really will, by balancing accuracy, latency, cost, privacy and intelligence all together, they’re going to win, that’s what’s going to win long term,” Srinivas told CNBC’s Elaine Yu in an interview on Wednesday.
A token refers to the basic unit of data that an AI model can process. When an AI chatbot is asked to carry out a task, it breaks it down into tokens. Each token then requires energy to be processed. Srinivas’ view is that whichever company can provide the best ratio of energy to economic output will be in the strongest position.
“And so it might feel like some model providers are making a lot of money because their models are very expensive … but that’s short-term revenue growth,” Srinivas said.
Perplexity is stepping up its focus on agentic AI, a term that refers to AI systems capable of handling more complex tasks beyond simple queries. In February, the company announced Perplexity Computer, an agent it says can execute complex tasks over long periods of time.
While Perplexity develops some of its own models, its key products integrate models from other AI firms like Anthropic. A key focus for Perplexity is improving efficiency to achieve the best outcomes while minimizing energy use. To support that goal, Perplexity announced Personal Computer on Tuesday, a tool which it calls an “orchestrator.”
The process of orchestration involves a system that makes decisions on what the best model is to use for a particular task, how agents work together, and where the AI should process queries. Much of the actual AI processing today is done in data centers.
But there is an increasing focus from AI firms on enabling these models to be processed on a device like a phone or laptop. Experts say this could reduce the power required to process AI, make it faster and more secure, since the data is not being sent to a server. Perplexity Personal Computer automatically routes the processing to where it deems best.
“The data center is coming to your laptop,” Srinivas said, adding that it’s crucial to have an AI operating system that brings everything together in a single unified system.
On Wednesday, Perplexity said that its Personal Computer product will be available on Microsoft’s Windows operating system, enabling the AI to connect to apps like Word and Outlook, as well as files on a user’s device. Perplexity has already launched the Personal Computer on Apple‘s Mac product.
Srinivas said that Perplexity is focused on creating a “sustainable, durable advantage” versus competitors and that “this is an orchestration problem.”
“We believe that by solving that, we’ll be building a pretty valuable company that has endurable, long-term advantage,” Srinivas said.
Rising competition
Perplexity faces increasing competition as rivals like OpenAI, Anthropic and Google have ramped up their focus on AI agents.
Perplexity, which was last reportedly valued at $20 billion, trails behind Anthropic and OpenAI, whose valuations have climbed to nearly $1 trillion and just over $850 billion, respectively. Anthropic this week confidentially filed for an initial public offering in the U.S. as investor demand for AI stocks continues.
Even as Perplexity expands its products across products from Microsoft and Apple, those very same companies are developing AI agents of their own. On Tuesday, Microsoft announced new coding and reasoning models. Apple is currently working on an updated version of its digital assistant Siri based on Google’s AI models.
Srinivas said that while these companies will build their own AI, Perplexity’s platform-agnostic approach will help it compete.
“I think they absolutely will try to build their own AI systems, but we believe we’re building the most versatile operating system by making it work across different models, across different chips, across different traditional operating systems, different hardware providers, different laptops,” Srinivas said.
“That hybrid neutral orchestration layer is what we are doing, and that allows us to balance all the different objectives simultaneously.”
Whenever Anthropic’s models get better, Perplexity improves too because Anthropic’s models are integrated into Perplexity, the CEO said. This has led Perplexity to triple its annualized revenue since the beginning of the year, “thanks to model advances that have been made by Anthropic,” he told CNBC.
