CNBC reported on June 3, 2026 that Perplexity CEO Aravind Srinivas told CNBC’s Elaine Yu that the company maximizing economic value per unit of AI power will command the highest valuations long term.
The metric
- Srinivas said the winner will provide the “most token value per watt per user” (CNBC’s article text renders the spoken phrase as “most taken value per watt per user”).
- He said success requires balancing accuracy, latency, cost, privacy, and intelligence together.
- CNBC explained that tokens are the basic units models process and that each token consumes energy—Srinivas argued the best energy-to-economic-output ratio wins, even if expensive models produce short-term revenue spikes.
Agentic orchestration
- Perplexity is emphasizing agentic AI beyond simple Q&A, CNBC said, citing the February launch of Perplexity Computer for long-running tasks.
- On Wednesday, Perplexity announced Personal Computer for Microsoft Windows, connecting Word, Outlook, and local files (Mac support was already available).
- Srinivas described orchestration—picking models, coordinating agents, and routing processing between devices and cloud—as the durable advantage, saying “the data center is coming to your laptop.”
Competition and valuations
- CNBC said Perplexity was last reportedly valued at $20 billion, trailing Anthropic (near $1 trillion) and OpenAI (above $850 billion).
- Anthropic confidentially filed for a U.S. IPO the same week, per CNBC.
- Srinivas said Perplexity tripled annualized revenue since the start of 2026, attributing gains partly to Anthropic model improvements integrated into Perplexity.
- Rivals including OpenAI, Anthropic, Google, Microsoft (new coding/reasoning models on Tuesday), and Apple (Siri updates using Google models) are also pushing agents; Srinivas argued Perplexity’s platform-agnostic orchestration across models, chips, and operating systems can still compete.
Primary source: CNBC — Perplexity CEO tells CNBC one metric will determine who wins the AI race (June 3, 2026).