Perplexity CEO: Value Per Watt Will Define AI Winners

Aravind Srinivas, CEO of Perplexity, says AI companies that deliver the most economic value for the power their systems consume will secure the highest valuations. 

He frames success as “taken value per watt per user”, balanced across accuracy, latency, cost, privacy and intelligence.

Aravind told CNBC: “Whoever is able to maximise this particular objective really will, by balancing accuracy, latency, cost, privacy and intelligence all together, they are going to win long term.” 

Measuring value per watt

When an AI chatbot handles a task, it processes information as tokens, the basic units of data a model consumes. Each token requires energy, so every interaction has a power cost.

Aravind argues that the winners will be those which best translate that energy into useful economic output. In practice, that means raising utility per token while lowering the power needed to produce it.

He describes this as optimising for value per watt per user across the entire experience. The focus is not just on raw model quality, but on the end-to-end outcome a user receives.

Companies that strike the right balance between capability and efficiency are likely to be strongest over time. In this view, energy economics becomes a central competitive moat.

Building an agentic operating system

Perplexity is expanding its agentic AI strategy to serve complex, multi-step workflows. In February, it introduced Perplexity Computer, an agent that can run deep research, write reports and integrate with workplace tools across hours or even months.

While Perplexity develops its own agents, its core products also integrate models from larger AI firms, including Anthropic. A key priority is to improve efficiency so outcomes are strong while energy use remains low.

To that end, Perplexity has introduced Personal Computer, an “orchestrator” that automatically routes tasks through the most efficient path. It connects to local files and applications to complete work with minimal friction.

Perplexity says Personal Computer is rolling out on Microsoft Windows, enabling connections to apps such as Word and Outlook, alongside files stored on a user’s device. The tool is already available on Apple’s Mac.

Orchestration across devices and chips

AI firms are moving to embed models within software and across hardware, including laptops and phones. Running AI closer to the user can cut power needs and latency while improving privacy, as data may not need to leave the device.

Personal Computer directs processing via the most efficient route, whether locally or in the cloud. The goal is to optimise for speed, cost and security in a single user workflow.

“The data centre is coming to your laptop,” Aravind says, stressing the need for an AI operating system that unifies models, apps and hardware into one experience.

“We believe that by solving that, we’ll be building a pretty valuable company that has endurable, long-term advantage.”

A platform-agnostic path to compete

Competition is intensifying as OpenAI, Anthropic and Google accelerate their agentic AI initiatives. Larger technology companies such as Microsoft and Apple are building integrated AI stacks across their ecosystems.

Aravind expects many companies to develop their own AI systems. Even so, he believes Perplexity’s platform-agnostic approach will help it compete across a wide range of setups.

He says: “We believe we are 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.”

Perplexity says its integration of Anthropic models has helped it triple annualised revenue since early 2026, citing model advances delivered by Anthropic.