Harvey AI Corp. CEO Winston Weinberg. The company can maintain an edge by going deep into areas tailored for law firms, such as tools for due diligence, Mr. Weinberg says.Ayesha Kazim/Supplied
Ontario Teachers’ Pension Plan has placed a bet in the crowded, fast-growing artificial-intelligence-powered legal software space, investing US$50-million in Harvey AI Corp., one of the leading platforms selling smart digital tools to lawyers.
The investment by Teachers’ Venture Growth (TVG), a unit of the pension giant that invests in later-stage technology companies globally, is an extension of a financing the San Francisco-based company announced earlier this month. At the time, Harvey said it had raised US$550-million led by private capital firms Diffusion and Lightspeed Ventures, at a valuation of US$15.5-billion. The TVG investment, the first by a Canadian institutional investor in the company, brings the round’s size to US$600-million.
Harvey is one of the largest AI-powered software vendors globally, recently topping US$400-million in annualized revenues just four years after it was co-founded by chief executive officer Winston Weinberg, a 31-year-old securities and antitrust litigator. His co-founder, president Gabe Pereyra, was Mr. Weinberg’s one-time roommate and previously worked as a research scientist with Google’s DeepMind AI research arm.
The 1,400-person company serves 20 per cent of the Fortune 500 and has more than 3,000 customers in 70 countries. Its 100-plus customers in Canada include law firms Davies Ward Phillips & Vineberg LLP, Dentons Canada, Gowling WLG and corporations Dentalcorp Holdings Ltd. and RONA Inc. Harvey opened an engineering and client support office in Toronto last year, its first outside the U.S.
AI tool Vero aims to bring down legal costs for small businesses
TVG executive managing director Olivia Steedman said her group had been tracking Harvey’s performance within the highly competitive legal AI technology market, and that it “stood out to us straight from the beginning. They beat all the targets they set out for themselves. The fact they’re getting this kind of adoption not only from law firms but also from internal counsels stood out for us.”
Harvey’s AI-powered legal software can answer routine questions, conduct research, and carry out work such as drafting memos and analyzing deposition transcripts.
As with many AI startups, Harvey has a complicated relationship with large language model (LLM) developers such as OpenAI and Anthropic. Harvey relies on commercial models from these companies to help power its platform, but OpenAI Group PBC and Anthropic PBC are also competitors circling the legal industry. (TVG is also an investor in Anthropic.) Faced with mounting costs for using big LLM models, many AI companies, including Harvey, have begun using lower-cost open-weight models.
With the LLM giants and a surfeit of other players competing in the legal AI technology space – including Canadian companies Spellbook, Clio, Blue J Legal and Thomson Reuters Corp. as well as Sweden’s Legora AB – it’s too early to know how the market will shake out. (Woodbridge Co. Ltd., the Thomson family holding company and controlling shareholder of Thomson Reuters, also owns The Globe and Mail.)
“The total addressable market is truly massive in legal and some of these companies are solving different customer needs and workflows,” Ms. Steedman said.
Mr. Weinberg said in an interview “the foundation models are competitors with every business on Earth. The models are getting better, and you have to make sure that you’re providing a tremendous amount of value to the users.”
Last week, OpenAI introduced a version of its GPT-6 Astra model tailored for the legal industry. OpenAI is going even further by embedding employees at select companies to build custom tools. With global law firm Sullivan & Cromwell LLP, for example, it built a tool to analyze agreements, spot risks and suggest changes.
Mr. Weinberg said Harvey can maintain an edge by going deep into areas tailored for law firms, such as tools for due diligence. Customers also prefer not to be locked into a single model provider. “You’re going to want to work with multiple models, and that’s why I think the app layer has a lot of value,” he said.
Among Canada’s lawyers, the jury is still out on AI
Harvey is training its own models, too. The company introduced Tenet in August, which is built on top of Kimi K3, an open-weight model from the Beijing-based company Moonshot AI. That reduces reliance on third-party models and lowers costs, since Harvey is building on top of an existing model rather than creating one from scratch.
Top-of-line AI models can be expensive to run in work settings. Tapping OpenAI’s Astra to complete all parts of a legal workflow could be more expensive than contracting a human lawyer, Mr. Weinberg said. “If you can get Astra-level performance, but it’s 10 times cheaper, now all of a sudden that’s a huge advantage,” he said.
Ms. Steedman said she wasn’t worried about the competition for Harvey from OpenAI and Anthropic, citing “the capabilities of the team, the product that they’re building, the speed at which they’re moving” at Harvey. “That gives us conviction.”
The Harvey investment is toward the lower end of the range of what TVG typically commits per deal, which ranges from $50-million to $250-million. TVG had $25.9-billion in net investments as of June 30 in 45 companies, including Space Explorations Technology Corp. (SpaceX), Anthropic, Databricks Inc. and Toronto’s ApplyBoard Inc. and StackAdapt Inc.
With reports from James Bradshaw