Inflection AI sets off fireworks with $1.3B funding, highlighting surging interest in LLMs (and Nvidia H100s)

In a pre-July 4th weekend surprise, Inflection AI, the Palo Alto-based startup founded by Mustafa Suleyman, cofounder of DeepMind, and LinkedIn co-founder Reed Hoffman, announced that it has raised $1.3 billion in an eye-popping round that brings its valuation to $4 billion.

Forbes reported that Microsoft and Nvidia led the round, along with Hoffman, Microsoft cofounder Bill Gates and former Google CEO Eric Schmidt. Nvidia was the only new investor — which is notable given that Nvidia and its service provider CoreWeave worked with Inflection to develop Inflection’s current H100 cluster, and Inflection worked with Nvidia to help fine-tune models on a recent MLPerf test that set records on current AI model training benchmarks.

The Forbes report also said Nvidia and CoreWeave are now helping Inflection install a cluster that will consist of a whopping 22,000 H100s — which Inflection believes to be the largest GPU cluster for AI applications in the world (even ahead of Meta’s 16,000 GPU cluster announced in May).

Does surging investor interest signal an AI bubble?

Not surprisingly, the surging investor interest in powerful LLMs to create “personal” chatbots has some already chattering about an AI bubble. Nic Carter, a general partner at Castle Island Ventures, said, “AI is making the crypto bubble look like child’s play.”

Inflection AI raised $225 million when it launched one year ago

Inflection, which is only a year old, made eyes water right from the beginning, when it announced it had launched and already raised $225 million with plans to use AI to “generate language to pretty much human-level performance.” 

And late on a Friday afternoon in March, the Financial Times reported that Suleyman and Hoffman were seeking up to $675 million in funding, even though they had yet to release a product.

That quickly changed: In May, the company launched Pi, which it said was named for “personal intelligence” and was meant to be “empathetic, useful and safe” — that is, acting more personally and colloquially than OpenAI’s GPT-4, Microsoft’s Bing or Google’s Bard, while not veering into the super-creepy. During a panel last week at the Bloomberg Technology Summit, Hoffman said that the Pi chatbot takes a more personal, emotional approach compared with ChatGPT. “IQ is not the only thing that matters here,” he said. “EQ matters as well.”

Last week, Inflection also announced that it would release a new LLM to power Pi, called Inflection-1, which it said outperforms OpenAI’s GPT-3.5.

Source