“What you really need is a fleet of F-150 pickup trucks,” Kuhn said. “We build F-150s.”
Founded by former Google AI researchers and backed by Nvidia, Cohere is betting on enterprise generative AI rather than consumer chatbots, which have been the talk of the tech industry since OpenAI released ChatGPT in late 2022.
In June, Cohere raised $270 million at a $2.2 billion valuation, with Salesforce and Oracle participating in the funding round. Company executives have attended AI forums at the White House. Cowher is said to be in talks to raise the amount Up to $1 billion In additional capital.
“We don't comment on rumours,” Kuhn told CNBC. “But someone once told me that startups are always growing.”
The field of generative AI has seen a major boom over the past year, with a record $29.1 billion invested across nearly 700 deals in 2023, a more than 260% increase in deal value from the previous year, according to PitchBook. It's become the catchphrase on corporate earnings calls quarter after quarter, and some form of technology is automating tasks in nearly every industry, from financial services and biomedical research to logistics, online travel and public utilities.
Although Cohere is often mentioned alongside AI heavyweights like OpenAI, Anthropic, Google, and Microsoft, the startup's focus on enterprise-only chatbots is what sets it apart.
Competitors are offering AI products to both consumers and businesses. For example, OpenAI launched ChatGPT Enterprise in August, and Anthropic opened consumer access to its previously business-only Claude chatbot in July.
By remaining focused on just the enterprise, Cohere is able to operate efficiently and keep costs under control even amid chip shortages, rising costs of graphics processing units (GPUs) and ever-changing licensing fees for intelligence models, said Kuhn, who is also the company's chief operating officer. Artificial.
“Rarely in my career have I seen so many companies that can be both consumer and successfully established at the same time, let alone a startup,” Kuhn said. “We don't have to raise billions of dollars to run a free consumer service,” he added.
Current clients include Notion, Oracle and Bamboo HR, according to Cohere's website. Many clients fall into the banking, financial services and insurance categories, Kuhn said. In November, Cowher told CNBC that she saw an uptick in customer interest following OpenAI's sudden and temporary ousting of CEO Sam Altman.
Kuhn acknowledges that the changing dynamics of the hardware industry have presented ongoing challenges. The company has had a reserve of Google chips for more than two years, secured in Cohere's early days to help it pre-train its models, Cohn said.
Now, Cohere is moving toward using more Nvidia H100 GPUs, which power most of today's large language models.
Kohn's relationships with strategic investors are another area where it differs from its AI competitors, Kuhn said. Many companies from the likes of Nvidia and Microsoft have introduced some conditions attached to the use of their software or chips.
Kuhn insists that Cohere has never accepted a conditional investment, and that every check cashed — including from Nvidia — has no conditions.
“In our last round, we had several checks of the same size, and we had no conditions associated with any of them,” Cohn said. “We made this decision clearly so we can say we don't owe it to anyone.”
Cohere's decision to focus on enterprise chatbots only may help the company stay out of the gray area of misinformation concerns, especially with election season looming.
In January, the Federal Trade Commission announced an investigation into artificial intelligence at Amazon, Alphabet, Microsoft, OpenAI, and Antropik. FTC Chairwoman Lina Khan called the matter “a market investigation into the investments and partnerships being formed between AI developers and major cloud providers.” Cowher was not named.
Kuhn says the company's growth so far has been largely around areas like search and retrieval, which require their own separate AI models. He calls it “using the tool,” and it involves training models on where, when and how to look for the information an enterprise customer needs, even if the model wasn't originally trained on that data.
Research is a key part of generative AI that receives less attention than other fields, Kuhn said.
“This will certainly, for institutions, be a real eye-opener,” he said.
In discussing the expansion timeline, Kuhn described 2023 as a “proof-of-concept year.”
“We believe that 2024 will turn out to be the year of widespread adoption,” he said.
Don't miss these stories from CNBC PRO:
He watches: Generative AI will democratize access to enterprise data.
“Typical beer trailblazer. Hipster-friendly web buff. Certified alcohol fanatic. Internetaholic. Infuriatingly humble zombie lover.”