Not for 'all of humanity': Startup Cohere focuses on AI for businesses

AI development company Cohere makes headlines far less frequently than its larger peers. According to its cofounder Nick Frosst, that is deliberate: the company does not seek the attention of the general public and focuses instead on “quiet” sales in the corporate segment. However, at the end of October the company announced that it was preparing for an IPO. For investors, that is a reason to take a closer look.

Cohere CEO and cofounder Aidan Gomez grew up in the woods of Ontario.

"We couldn’t get high speed because of where we lived so it was dial-up with that horrible noise and it was so slow you could watch pictures load pixel-by-pixel. Because it was largely out of reach, the internet was a special, magical thing... That scarcity drove me. I would try to squeeze performance out of computers, make them faster, make my internet work a little bit better. It forced me to understand the tech," he recalls.

In 2011, when he was still a 15-year-old high-school student, Gomez founded AdGo Design, which built and maintained websites for local businesses – from a knitting supply store to a family insurance company – for CAD500 per project.

With this backdrop, it is unsurprising that he enrolled in a computer science program at the University of Toronto, graduating in 2018, and immediately began a PhD in computer science at the University of Oxford.

While still a student, he interned at Google Brain’s AI division, and in 2017 coauthored the paper “Attention Is All You Need,” which ultimately shaped the current trajectory of the global AI race, as well as Gomez’s own professional path. 

The paper introduced what was then a new architecture for neural networks known as the "Transformer." With the help of attention mechanisms, it improved English-to-French and English-to-German translation quality while training faster and requiring fewer resources. In addition, the structure of Transformer networks better leveraged GPUs designed for parallel processing. That resulted in faster performance versus sequential word-by-word processing. As noted in the paper, Gomez spent “countless long days” rewriting code, accelerating the research and improving results.

At the time, no one knew that transformers would become the foundation of the generative-AI boom that began in 2022 with the release of OpenAI’s ChatGPT. The GPT acronym stands for Generative Pre-trained Transformer. In effect, in 2017 the Google Brain team, with Gomez’s participation, introduced the last, missing “T” that would later transform the entire industry.

"While we were doing the work, it felt like a regular research project... We already had Google Translate; we wanted to make it a little bit better. We improved the accuracy by a few percent by creating this architecture, and I thought it was done. That was the contribution... I think it took about a year for the community to take notice... then we just started to see this snowball effect where everyone started adapting it to new use cases. It wasn’t just for translation. It started being used for all of these other NLP, or natural language processing, applications. Then we saw it applied toward language modeling and language representation. That was really the spark where things started to change," Gomez told the Verge. 

Two years after the paper was published, Gomez, together with Google Brain colleagues Ivan Zhang and Nick Frosst, founded Cohere.

All three founders are Canadian, so the decision to open the company’s head office in Toronto was unsurprising. At first, however, the operation was modest. Gomez recalls that the “office” was “this tiny little… basically a closet. I don’t know how many square meters it was but, you know, single digits,” where the three founders worked side by side.

They did not yet have a clear vision of what commercial product they wanted to create. There was only a general sense – now clearly validated – that natural-language technology was poised for rapid development and would acquire commercial applications.

“We just focused on building the infrastructure to train large language models on supercomputers using whatever compute we could get our hands on. Shortly after we started Cohere, GPT-3 came out, and that was a huge breakthrough moment that was very validating and gave us [an indication] that we were heading [down] the right path,” Gomez told Contrary Research.

Cohere received its first $40 million funding round in 2021 from Index Ventures, with participation from Section 32, Radical Ventures, and several prominent AI figures, including "AI godparents" Geoffrey Hinton and Fei-Fei Li.

Less than a year later, a second round for $125 million followed. The company said platform usage increased more than 800% between rounds – although, as Gomez noted elsewhere, relative numbers at that stage can be misleading, since the base was extremely small in absolute terms. 

The broader AI boom that followed accelerated Cohere’s fundraising. To date the company has raised $1.7 billion at a total valuation of $7.0 billion, according to TechCrunch. Recent investors include such names as Nvidia, AMD, and Salesforce.

A decade ago, growing a valuation from zero to $7 billion in six years would have been extraordinary. But the rise of OpenAI and Anthropic has made that look quaint: In October, OpenAI’s valuation reached $500 billion, and in early September Anthropic’s was $183 billion.

Unlike OpenAI – whose CEO Sam Altman largely brushes off questions about profitability – Gomez says Cohere, with projected annual revenue of $150 million in 2025, is on “on a clear path to profitability,” and he expects the company to turn a profit “sooner” than 2029, writes Bloomberg.

Frosst argues that Cohere has not been promoting itself aggressively in the public sphere and has remained “below the radar."

"We’re pretty Canadian. It’s not in our DNA to be out there talking about how amazing we are," he joked in an interview with Fortune magazine.

The real reason for this modest public profile is that Cohere has targeted the corporate segment from the beginning. These customers prioritize privacy and security.

So what does Cohere actually do for them? Gomez says the company develops practical AI models that meaningfully increase employee productivity. Cohere's models can be deployed either in the cloud or on a company’s own hardware, ensuring that valuable data does not leave the organization's control perimeter. He cites an MIT study showing that access to the chatbot ChatGPT decreased the time it took workers to complete the tasks by 40%, while output quality, as measured by independent evaluators, rose by 18%.

As an example of how generative AI is used in business, Gomez described a model Cohere built for an insurer. The technology allows the company to submit faster quotes to beat out the competition when there’s a request for proposal from a mining or pipeline firm.

“I never thought that an insurance company for natural resource projects would be adopting large language models,” he told CNBC. “But they are.”

According to Frosst, Cohere focuses on clear return on investment and profit, rather than developing artificial general intelligence “for the benefit of all humanity,” as OpenAI describes its mission, or pursuing “superintelligence,” as Meta presents its goals.

This pragmatic strategy appears much better suited for a (soon-to-be) public firm than the ambitious but somewhat uncertain approach of OpenAI, which does not expect profitability in the foreseeable future.