Perplexity CEO Aravind Srinivas said AI companies that generate the most economic value from the least amount of computing power will be best positioned to succeed as competition intensifies across the sector. Speaking to CNBC’s Elaine Yu, Srinivas argued that maximizing “token value per watt per user” will become a key metric for evaluating AI businesses.

Balancing Accuracy, Latency, Cost And Privacy

Srinivas explained that every AI token, the basic unit of data processed by a model, consumes energy. Companies that can optimize the relationship between energy use and economic output, he said, will have a significant advantage. “Whoever is able to maximize this particular objective by balancing accuracy, latency, cost, privacy and intelligence really will win in the long term,” Srinivas said. His comments reflect a growing focus within the AI industry on efficiency as companies face rising infrastructure and computing costs.

Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram

Advancing Agentic AI With A Platform-Agnostic Approach

Perplexity continues to expand its work in agentic AI, systems designed to complete complex tasks rather than respond to individual prompts. In February, the company introduced Perplexity Computer, an AI agent capable of handling multi-step tasks over extended periods. While Perplexity develops its own models, its products also incorporate technology from companies including Anthropic.

The company recently launched Personal Computer, an orchestration layer that routes queries to the most appropriate processing resource. Srinivas described the shift as a move toward bringing more AI capabilities directly onto personal devices rather than relying exclusively on centralized data centres. According to Srinivas, this approach can reduce energy consumption while improving privacy and security.

Integration Across Leading Platforms And Growing Competition

Personal Computer is currently available on Apple’s Mac devices and is expected to expand to Microsoft’s Windows platform. The system is designed to work across applications, including Word and Outlook. The strategy also differentiates Perplexity from competitors such as OpenAI, Anthropic and Google, which are building increasingly integrated AI ecosystems around their own models and platforms.

Despite rapid growth among rivals, Srinivas said Perplexity’s focus remains on creating a system that works across different models, chips and operating systems. “We believe we’re building the most versatile operating system by making it work across different models, chips, and operating systems,” he said.

Future Outlook: Sustainable And Enduring Advantage

Perplexity’s approach allows the company to incorporate advances from multiple AI providers rather than relying on a single model ecosystem. The strategy has coincided with strong business growth. According to Srinivas, the company’s annualized revenue has tripled since the beginning of the year. As competition intensifies across the AI sector, efficiency, infrastructure costs and cross-platform integration are becoming increasingly important factors for companies seeking to scale their products and services.