Just last year, Inflection AI was as hot as a startup could be, releasing best-in-class AI models it claimed could outperform technology from OpenAI, Meta, and Google. That’s a stark contrast compared to today, as Inflection’s new CEO tells TechCrunch that his startup is simply no longer trying to compete on that front.
Between then and now, there’s of course been a major change at Inflection. Microsoft hired then-CEO Mustafa Suleyman to run its own AI business, and paid the startup $650 million to hire most of its staff and license its technology. Several months ago, Inflection announced it was starting to limit usage on its consumer AI chatbot, Pi, while pivoting more towards enterprise customers.
Instead, Inflection announced on Tuesday it has now acquired three AI startups itself, in just the last two months, to build up the tools it can offer global enterprise customers using AI models available today. The company also is not ruling out licensing AI models from its former competitors in the future.
The Federal Trade Commission is reportedly investigating Microsoft’s partial-acqui-hire of Inflection to see whether the deal was structured in a way that would reduce competition.
According to Inflection’s new CEO, Sean White, who was put in place after the deal, his startup is no longer competing on building the next generation of AI models, but still can compete on the enterprise front.
“I am not going to, and don’t feel the need to, compete with a company that is trying to build the next 100,000-GPU system,” said White in an interview with TechCrunch, seeming to reference the handful of well-funded companies that can build frontier AI models today — including Microsoft, the new home of Inflection’s founders.
“When I say we can’t compete with them, I think part of it is that I don’t want to compete with them trying to make that next generation model,” White clarified. “I do think we’re actually still competing with them, particularly for the enterprise. But in the end, our solution of how we architect this and the tools that we’re bringing are really that enterprise layer that actually is going to meet their needs.”
White thinks today’s AI models are just fine to address the needs of most enterprises today. He even goes a step further, saying he’s skeptical about how test-time compute scaling, which many are calling the next generation of AI models, can address business use cases. Inflection’s CEO says AI labs have cunningly reframed high latency as “thinking” in order to make consumers feel better about their models.
“I mean, there’s a little piece of me that says, ‘Haha! Now we all have latency in our inference, so we’re going to call it thinking,’ as opposed to just saying, ‘Yeah, we’ve just got more latency because these things are getting bigger and harder,’” said White.
Instead of pressing on the cutting edge of AI research, Inflection is now trying to think more practically to offer AI tools for enterprises. Inflection announced on Tuesday it had acquired two small startups as part of that effort: Jelled.AI, which uses AI to manage employee inboxes, and BoostKPI, which offers AI data analytics tools. Last month, Inflection announced it had acquired Boundaryless, an automation consulting firm in Europe to expand its presence overseas.
White says Inflection is still using its own models today, but says that doesn’t mean they won’t use other AI models in the future.
Part of Inflection’s value proposition now is that its AI can run on premise, compared to offerings from leading AI labs, which must be run in the cloud. This can be especially appealing to enterprises who want to keep their data secured.
These acquisition have helped Inflection build up an array of talent and products. However, the startup will also face intense competition on the enterprise AI front. Salesforce has gone all in on AI agents in recent months, while Meta recently unveiled a new business AI unit. In terms of startups, Anthropic and Cohere are continuing to build products specific for business customers. That said, Inflection feels it’s better suited to compete in the enterprise space today, instead of pushing against frontier AI labs to make increasingly capable AI models.
