During an interview with CNBC on Wednesday, Aravind Srinivas, CEO of Perplexity, emphasized that the companies that can derive the greatest economic value from their AI energy consumption will secure the highest market valuations.
Srinivas highlighted the importance of achieving what he termed the "most token value per watt per user," suggesting that this will be a decisive factor for future success in the industry. According to him, the organization that optimally balances several key factors—including accuracy, latency, cost, privacy, and intelligence—will emerge as the leader in the long run. He shared these insights in a conversation with CNBC's Elaine Yu.
In the context of AI, a "token" represents the fundamental unit of data processed by AI models. When a chatbot is tasked with a query, it segments the information into tokens, each of which requires energy for processing. Srinivas believes that companies that achieve the most efficient conversion of energy into economic success will stand out in a competitive landscape.
He further remarked on companies that may appear profitable due to their high-priced models, labeling such financial performance as merely a temporary spike in revenue.
Perplexity is increasingly focusing on "agentic AI," designed to tackle sophisticated tasks that extend beyond basic queries. In February, the company revealed Perplexity Computer, an agent equipped to manage intricate tasks over extended durations.
While Perplexity does develop some of its own AI models, it primarily integrates technologies from other AI firms, such as Anthropic. A central objective for the company is enhancing efficiency to facilitate optimal outcomes while simultaneously reducing energy consumption. In pursuit of this aim, Perplexity recently unveiled Personal Computer, a tool categorized as an "orchestrator."
