‘Are we under pressure to IPO?’: Perplexity's Aravind Srinivas breaks silence on funding, product rumours

After a wave of speculation about Perplexity’s product updates and future plans, CEO Aravind Srinivas took to X to set the record straight. Addressing questions around the company’s financial health, model choices, and the motivations behind recent interface changes, Srinivas clarified one thing up front: the AI startup is not running out of money, nor is it rushing toward an IPO.

“Are we running out of funding and facing market pressure to IPO? No. We have all the funding we've raised, and our revenue is only growing,” Srinivas posted. “The objective behind Auto mode is to make the product better, not to save costs.”

He added, “If anything, I have learned it's better to communicate more transparently to avoid any incorrect conclusions. Re IPO: We have no plans of IPOing before 2028.”

The statement came as users questioned the rationale behind the newly introduced Auto mode and missing model selectors. Srinivas explained that Auto mode is aimed at reducing clutter and simplifying the experience: “The user shouldn't have to learn so much to use a product.” Auto mode lets the AI determine the complexity of a query—from quick answers to deep research—without overwhelming the user.

On inconsistencies in model availability, Srinivas noted that not all models suit every mode. “o3-mini and DeepSeek R1 don't make sense in the context of Pro Search,” he wrote. “As for Deep Research, it's a combination of multiple models that all work together... there's absolutely nothing to control there.”

He added that model selectors are evolving: “Pro is customizable. Pro will persist across follow-ups. Reasoning does not, but we intend to merge Pro and Reasoning into one single mode.”

As for Deep Research reverting to Auto for follow-ups, the team found that slow response times were discouraging engagement: “15–20% of Deep Research queries are not seen at all bc they take too long.”

Finally, on GPT-4.5, Srinivas said it's simply too slow: “The decoding speed for GPT-4.5 is only 11 tokens/sec... Until we can achieve speeds similar to what users expect, we will have to hold off.”