Perplexity CEO Aravind Srinivas Says Efficiency Will Separate AI Winners: “Token Value Per Watt...

OpenAI rival Perplexity CEO Aravind Srinivas has zeroed in on what he believes will ultimately determine the winners in the artificial intelligence race: the ability to deliver maximum economic value from the energy consumed by AI systems.

In a CNBC interview on Wednesday, Srinivas argued that the companies best able to maximize "token value per watt per user", essentially delivering the highest useful output per unit of energy and per user, will command the highest valuations in the long term.

A token is the basic unit of data that an AI model processes when handling a query or task. Each token requires computational power and, by extension, electricity.

"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.

He acknowledged that current AI spending patterns raise legitimate questions. Many companies are pouring billions into infrastructure with limited visibility into returns, creating significant waste.

"You hear companies saying, I am spending a ton of money on AI. And I know some great stuff is happening, but I know there's a ton of waste. How long do I have to wait for it to really show up in revenue, and how long do I have to wait to really get the costs under control?" he asked.

Srinivas views this as a temporary phase. He believes the industry will quickly improve efficiency, but the companies that solve the energy-to-value equation first will build durable advantages.

The Shift Toward Agentic AI and Orchestration

Perplexity is positioning itself at the center of this efficiency drive through a strong focus on agentic AI - systems capable of handling complex, multi-step tasks autonomously rather than simple prompt-and-response interactions. In February, the company launched Perplexity Computer, an agent designed to execute extended workflows.

A key innovation is the newly introduced Personal Computer tool, which Perplexity calls an "orchestrator." This system intelligently decides which AI model to use for a given task, how different agents should collaborate, and whether processing should occur locally on a device or in the cloud.

On Wednesday, Perplexity announced that Personal Computer is now available on Microsoft's Windows operating system, allowing integration with apps like Word and Outlook, in addition to its existing availability on Apple's Mac platform.

Srinivas emphasized the importance of this layer, saying: "The data center is coming to your laptop… 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."

By acting as a neutral orchestrator that works across different models (including those from Anthropic), chips, operating systems, and hardware providers, Perplexity aims to create a versatile "AI operating system" that optimizes for multiple objectives simultaneously - cost, speed, privacy, and performance.

Platform-Agnostic Strategy in a Crowded Field

Perplexity, last valued at around $20 billion, trails far behind larger rivals such as Anthropic (nearing $1 trillion) and OpenAI (over $850 billion). Anthropic confidentially filed for an IPO this week, highlighting intense investor appetite for AI companies.

The competitive landscape is intensifying. OpenAI, Anthropic, and Google are all ramping up agentic capabilities, while Microsoft and Apple are building their own AI agents and assistants. Microsoft unveiled new coding and reasoning models on Tuesday, and Apple is updating Siri using Google's AI technology.

Srinivas remains confident in Perplexity's differentiated approach.

"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. That hybrid neutral orchestration layer is what we are doing, and that allows us to balance all the different objectives simultaneously," he said.

He noted that Perplexity has tripled its annualized revenue since the beginning of the year, largely "thanks to model advances that have been made by Anthropic," whose models are integrated into Perplexity's platform. This highlights a key strength of the orchestration strategy: the company benefits from rapid improvements by frontier model providers without bearing the full cost of developing them in-house.

Why Efficiency Will Define the Next Phase of AI

Srinivas's emphasis on energy efficiency and orchestration addresses a growing tension in the AI industry. Hyperscalers are spending hundreds of billions on data centers and chips, yet real-world utilization rates remain low in many cases, and returns on investment are still being proven at scale. The focus on edge computing, running AI locally on devices like laptops and phones, could dramatically reduce power consumption, improve speed and privacy, and lower latency.

This shift from cloud-only to hybrid cloud-edge architectures represents a potential inflection point. Companies that master orchestration are considered to be better positioned to deliver cost-effective, responsive AI experiences while minimizing environmental impact and energy costs - factors increasingly important to both regulators and corporate buyers.

For Perplexity, this strategy aims to create a sustainable moat in a field where raw model performance alone may not guarantee long-term leadership. The company hopes to carve out a valuable niche as the "AI operating system" layer by focusing on intelligent routing and system-level optimization rather than solely chasing the largest models.