How a 24-year-old raised $64M to build an AI mathematician

Executive overview

Most mathematical tools — from the abacus to calculus — take centuries to move from invention to real-world impact. AI compresses that timeline dramatically. Axiom is building the first self-improving AI mathematician, combining deep learning, formal proof languages, and mathematical research expertise. The goal: turn every organisation into one that can afford world-class quantitative reasoning.

Mathematical taste and intuition, not just computation, is the frontier AI must cross.

The market opportunity unlocked by AI mathematics

• Quant researchers cost ~$14M/year; an AI mathematician costs $5/hour
• Previously unviable markets (e.g. $8M trading volumes) become economically worth analysing
• Jevons paradox: as tool cost drops, unexpected use cases multiply, driving demand further
• AI mathematicians can collaborate directly with applied scientists — something human mathematicians rarely do
• Centuries-long gaps between mathematical invention and application could compress to years

The three pillars of Axiom's approach

• Three disciplines combined: AI, programming languages, and mathematics
• Formal proof language Lean used to build a deductive knowledge graph
• Team spans deep learning (Hugh Lather, since 2017), AI-for-math discovery (François Charton, transformer-based symbolic integration 2019), and large reinforcement learning (CTO Shubhash Ankupta, OpenGo at FAIR)
• Company name "Axiom" reflects the axiomatic, deductive reasoning Carina encountered at the Ross Mathematics Program age 15

From Olympiad problems to research mathematics

• Olympiad math gives instant dopamine hits — a dozen problems solved per day
• Research math is delayed gratification: months can pass with no visible progress
• The shift requires moving from problem solver to theory builder
• Key skills: patience, looking for unexpected cross-field connections, developing mathematical taste
• Taste — knowing what is natural, interesting, and elegant — distinguishes good scientists from mediocre ones

Why math is the sandbox for AI reasoning

• Math is a fully digital domain: no real-world data collection required
• Models trained on math transfer strongly to coding and other reasoning tasks
• Formal proofs allow strict, rigorous verification of conclusions — not pattern matching
• AI mathematicians can tackle complex systems in applied science that have never been theoretically understood
• Vision: human mathematician guides lemma-to-lemma; AI proves each step, making the journey faster and less isolating