Perplexity AI CEO Aravind Srinivas has predicted how companies can win the AI race. He believes that the company that generates the most economic value from the energy consumed by artificial intelligence (AI) systems will ultimately emerge as the long-term winner in the AI race. Speaking to CNBC, Srinivas said the most important metric for AI companies is "token value per watt per user", a measure that combines efficiency, performance and economic output. According to him, the ability to maximise this metric will be more important than simply building larger or more expensive AI models.“Whoever is able to maximise this particular objective really will, by balancing accuracy, latency, cost, privacy and intelligence altogether; they’re going to win- that’s what’s going to win long term,” Srinivas told CNBC.A token is the basic unit of data processed by an AI model. Every AI prompt is broken down into tokens, each of which requires computing power and energy to process. Srinivas argued that the companies capable of delivering the highest value from that energy usage will be in the strongest position over time. He also suggested that high-priced AI models generating strong revenue today may not necessarily translate into long-term leadership.“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 added.How Perplexity is focusing on improving efficiency of AI agentsSrinivas’ comments come as Perplexity expands its efforts in agentic AI, which refers to systems capable of carrying out more complex tasks with limited user intervention. Earlier this year, the company introduced Perplexity Computer, an AI agent designed to handle tasks over extended periods.To improve efficiency, the company recently launched Personal Computer, a product it describes as an "orchestrator." The tool determines which AI model should be used for a particular task, how different AI agents should work together and where processing should take place. Srinivas said the future of AI will involve moving more processing from large data centres to local devices such as laptops and smartphones.“The data center is coming to your laptop,” Srinivas said, adding that AI systems require a unified platform capable of coordinating models, hardware and operating systems. Perplexity has expanded the Personal Computer platform to Microsoft's Windows operating system, allowing it to interact with applications such as Word and Outlook, as well as files stored on a user's device. The product was previously launched on Apple's Mac platform.Perplexity faces competition from companies including OpenAI, Anthropic and Google, all of which are investing heavily in AI agents and foundation models. Srinivas said Perplexity's approach is to remain platform-neutral by supporting multiple AI models, hardware platforms and operating systems.“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,” he said.According to Srinivas, improvements made by model providers such as Anthropic directly benefit Perplexity because their technologies are integrated into the company's products. He said these advancements have helped Perplexity triple its annualised revenue since the start of the year.