Perplexity CEO: Efficiency, Not Raw Power, Will Determine AI Winner
Perplexity CEO Aravind Srinivas told CNBC that the company capable of delivering the most economic value from its energy consumption will ultimately dominate the AI market. He defined this metric as "token value per watt per user" - essentially, how much useful work an AI system produces relative to the power it consumes.
The measure matters because every token an AI model processes requires energy. A company that maximizes accuracy, latency, cost, privacy, and intelligence while minimizing power consumption will outcompete rivals focused solely on revenue growth, Srinivas said.
"Whoever is able to maximize this particular objective really will... they're going to win, that's what's going to win long term," he told CNBC on Wednesday.
Orchestration as Competitive Advantage
Perplexity announced Personal Computer on Tuesday, a tool the company describes as an "orchestrator." The system decides which AI model to use for specific tasks, coordinates how agents work together, and determines where processing should occur - on a device or in a data center.
This approach differs from competitors that route most queries to centralized servers. Perplexity's method can reduce power consumption, improve speed, and enhance security by keeping data on users' devices rather than transmitting it to remote servers.
"The data center is coming to your laptop," Srinivas said, emphasizing the need for an AI operating system that unifies everything into a single system.
Perplexity released Personal Computer on Apple's Mac and announced availability on Microsoft Windows, where it will connect to apps like Word and Outlook. The platform-agnostic approach allows the system to work across different models, chips, and hardware providers.
Integration Strategy Drives Growth
While Perplexity develops some models internally, its products integrate AI from other firms like Anthropic. This strategy has produced measurable results: the company tripled its annualized revenue since the start of the year, partly because improvements in Anthropic's models automatically benefit Perplexity's offerings.
Srinivas acknowledged that major tech companies - Microsoft, Apple, and Google - are building their own AI agents. He argued that Perplexity's neutral orchestration layer gives it an advantage by balancing competing objectives simultaneously across different vendors' technologies.
Perplexity was last valued at $20 billion, trailing Anthropic (nearly $1 trillion) and OpenAI (just over $850 billion). Anthropic confidentially filed for an initial public offering this week as investor demand for AI stocks remains strong.
Agentic AI as Next Frontier
Perplexity is shifting focus toward AI agents and automation - systems capable of handling complex, multi-step tasks beyond simple queries. In February, the company launched Perplexity Computer, an agent designed to execute complicated assignments over extended periods.
This pivot reflects broader industry movement. OpenAI, Anthropic, and Google have all increased investment in AI agents as the market moves beyond chatbots toward autonomous systems that can take independent action.
For professionals building or managing AI systems, Srinivas's efficiency metric suggests a shift in how to evaluate vendors and platforms. The lowest-cost or most capable model alone won't determine winners - the ability to deploy that capability efficiently will.
Your membership also unlocks:
