Stanford Math PhD’s AI Startup Targets $300 Million Valuation

Mathematicians have been enthralled with artificial intelligence that can solve difficult math problems. And some developers behind these models at OpenAI and elsewhere believe AI that learns to solve hard math can use similar reasoning methods to solve other types of problems.

But since today’s AI still hasn’t proven it can beat human experts at advanced math, some founders see an opening to develop models that specialize in the field. One of them is Carina Hong, a math PhD student at Stanford University, who is now in talks with investors to raise $50 million for her new startup, Axiom, which aims to develop math-solving AI.

Hong hasn’t developed a product yet but has been in talks to raise capital at a valuation of $300 million to $500 million, according to three people who have spoken to Hong. The deal isn’t closed and the terms could change.

B Capital, which has backed well-known AI firms including Glean, Perplexity, Writer, and Poolside, is in talks to lead the round, which has not yet closed, one of the people said. 

Hong declined to provide a comment for this story. B Capital spokespeople did not respond to multiple requests for comment.

Hong told potential investors Axiom will develop AI trained on formal mathematical proofs, which show the truth of a mathematical statement using proven theorems, also known as axioms.

Axiom aims to sell its eventual product to customers such as hedge funds and quantitative traders who need rapid answers to complex math problems that relate to broader questions about asset values or the stock market, Hong has told potential backers.

Hong, who is from China, graduated with a degree in math and physics from the Massachusetts Institute of Technology in 2022 and subsequently attended Oxford University before enrolling at Stanford in 2024, according to her online biographies, published by the universities. She has previously published research papers on combinatorics, an area of math that involves counting combinations of objects. It isn’t clear which AI researchers she has recruited to help develop the models she envisions.

The deal talks show there’s still an appetite among investors to agree to high valuations for AI-related business ideas, despite other investors’ warnings that it’s too late for newcomers focused on developing their own models.

AI from top developers including Anthropic, OpenAI and Google still struggle to answer basic math questions, such as giving incorrect answers to questions that ask whether 9.11 is smaller or bigger than 9.8, but have shown lots of promise in answering complex queries.

OpenAI’s o4-mini recently outperformed teams of human mathematicians in a math evaluation for AI called FrontierMath. Last year, Google’s AI completed the International Math Olympiad at the level of a silver medalist and earlier this year solved geometry problems at the level of a gold medalist. 

Coincidentally or not, Google’s math model couldn’t solve combinatorics problems—Hong’s specialty.

Hong has previously downplayed some of those models’ performance on math. She said in a recent post on X that OpenAI’s models may perform well on math exams because the firm had access to the exam problems when training the models, and she said in an April interview that those models were good at answering questions but not proving that those answers are correct.

“What we want to build at Axiom is to use programming language to train the machine to be able to speak the language of formal proof,” Hong said in the interview.

Her startup bears a resemblance to other math-focused AI startups founded in the past year. Harmonic, an AI startup founded in 2023 by Robinhood CEO Vlad Tenev, raised $75 million at a $325 million post-money valuation from Sequoia and other backers last fall with a similar pitch, claiming it would build a model capable of solving advanced math problems. 

Here’s what else is going on…

Court Watch

Google has agreed to spend $500 million over 10 years and create new compliance committees to settle shareholder litigation accusing it of antitrust violations, according to court documents filed Friday.

Deals and Debuts

See The Information’s Generative AI Database for an exclusive list of private companies and their investors.

Microsoft announced that it would invest $400 million in Swiss AI and cloud companies.

xAI is allowing employees to sell $300 million worth of shares to investors, in a deal that values the company at $113 billion. That’s the same price for the company Musk announced in March, when xAI bought his social media service X.

Netic, which uses AI to help plumbers and electricians automate client calls, raised $20 million in funding from investors including Greylock and Founders Fund.

pWin.ai, which develops AI-powered writing software for government contractors to bid on contracts, raised $10 million in seed funding, led by MicroStrategy co-founder Sanju Bansal  and Blue Delta Capital.

Deepdots, formerly Magic Feedback, which develops AI software for analyzing customer interactions, raised $6.3 million in seed funding led by Dawn Capital, with participation from Maki.vc and Geeta Schmidt.

Literal Labs, which develops explainable AI models for regulated industries, raised $6.2 million in pre-seed funding led by Northern Gritstone and Mercuri, with participation from Sure Valley Ventures and Cambridge Future Tech SPV.

Outter, which lets developers at small companies easily integrate chatbots in their websites, is targeting a seed round up to $1.37 million, Tech Funding News reported.

A U.K. government study found that civil servants saved 26 minutes per day, or 2 weeks per year, by using Microsoft’s Copilot AI.

The Food and Drug Administration is rolling out an AI tool called Elsa the agency will use to help employees with reviewing drug trials and writing code, Axios reported.

New From Our Reporters

Exclusive

How Alibaba Helped China Take the Lead From the U.S. in Open-Source AI

By Juro Osawa and Qianer Liu

Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: [email protected].

If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.