Just last year, Inflection AI received the highest profile for a startup, releasing a best-in-class AI model that it claims has the potential to outperform technologies from OpenAI, Meta, and Google. This is a stark contrast from today, as Inflection's new CEO told TechCrunch that his startup is no longer trying to compete in that space.
Of course, a lot has changed at Inflection between then and now. Microsoft hired then-CEO Mustafa Suleiman to run its own AI business, hired most of its staff and paid the startup $650 million to license the technology. . A few months ago, Inflection announced that it would begin restricting the use of its consumer AI chatbot Pi, with a focus on enterprise customers.
Instead, Inflection announced Tuesday that it has acquired three AI startups in just the past two months to build tools it can offer to enterprise customers around the world using currently available AI models. . The company also hasn't ruled out licensing AI models from former competitors in the future.
The Federal Trade Commission is reportedly investigating Microsoft's partial acquisition of Inflexion, Inc. to see if the deal is structured in a way that reduces competition.
Inflexion's new CEO, Shaun White, who took over after the deal, says his startup is no longer competing in building next-generation AI models, but it can still compete on the enterprise side.
“We're not going to compete with, and we don't feel the need to compete with, the companies that are going to build the next 100,000 GPU systems,” White said in an interview with TechCrunch, referring to a handful of well-funded companies. It seems like it is. Companies that can now build cutting-edge AI models include Microsoft, the new home of Inflection's founders.
“I think when we say we can't compete with them, it's also because we don't want to compete with them who are trying to make that next generation model,” White clarified. “I actually think we're still competing with them, especially when it comes to enterprises. But ultimately, our solution of how to build this and the tools that we provide are It actually meets the needs of the enterprise segment.”
White believes that today's AI models are sufficient to meet the needs of most businesses today. He goes a step further and says he is skeptical about how the test-time compute scaling of what many are calling the next generation of AI models will address business use cases. Inflection's CEO says the AI lab has cleverly reframed high latency as “thought” to give consumers a better impression of its models.
“So there's a little part of me that goes, 'Ha!'” We all have delays in reasoning right now, so we just say, “Oh, these things are getting harder and harder. So instead of saying, 'We're just adding more delays,' let's call it thinking,” White said.
Rather than promoting cutting-edge AI research, Inflexion is now thinking more pragmatically about providing AI tools for companies. As part of its efforts, Inflexion on Tuesday announced the acquisition of two small startups: Jelled.AI, which uses AI to manage employee inboxes, and BoostKPI, which provides AI data analysis tools. Last month, Inflexion announced it had acquired European automation consulting firm Boundaryless to expand its presence overseas.
White said that while Inflection still uses its own models, that doesn't mean it won't use other AI models in the future.
Part of Inflection's value proposition today is that its AI can run on-premises, compared to products from big AI labs that have to run in the cloud. This is especially attractive for businesses that want to keep their data safe.
These acquisitions have helped Inflection build a diverse workforce and products. But the startup will also face stiff competition in the enterprise AI space. Salesforce has been going all-in on AI agents in recent months, and Meta recently announced a new business AI division. As for startups, Anthropic and Cohere continue to develop specialized products for enterprise customers. That said, Inflection feels it is better suited to compete in today's enterprise space than to go up against cutting-edge AI labs aiming to create increasingly capable AI models. .