Entrepreneurship April 22, 2026 0 Comments

Why Top Wall Street Math Geniuses Are Abandoning Finance—and the Surprising Industry Luring Them Away Revealed by Neel Somani

Ever wonder why some of the sharpest math minds are packing up their Wall Street calculators and heading straight for the buzzing labs of AI research? It’s not just about chasing bigger paychecks anymore — though those have gotten pretty eye-popping on both sides. The real kicker? It’s about the kind of puzzles they want to solve and where their skills will actually matter tomorrow. Neel Somani, who’s played on both fields — from the high-stakes trading floors of Citadel to the cutting edge of AI interpretability research — throws down a challenge to old assumptions about what a “big win” means for mathematically gifted pros. As hedge funds and tech giants square off in a battle for talent, the question isn’t just who pays more, but who offers a career that keeps compounding in value long after the initial offer is signed. So, where should you aim your mathematical genius when the future’s calling? Stick around, because the answer might surprise you. LEARN MORE

Table of Contents

Key Takeaways

Top mathematical talent is increasingly moving from Wall Street to AI labs due to evolving career incentives and opportunities.
AI labs now offer highly competitive compensation packages, narrowing the historical pay gap with quant finance.
Skills developed in AI research tend to be more transferable across industries compared to specialized finance expertise.
Quant finance remains intellectually rigorous, but operates in a mature, highly competitive environment with diminishing edges.
AI research offers exposure to unsolved, high-impact problems where experience and insight compound over time.

Something noteworthy has been happening at the recruiting dinners that hedge funds throw. The firms still use the same pitch they have had for years: intellectual rigor, fast feedback loops, competitive compensation, and the quiet satisfaction of knowing whether ideas actually work. What has changed is the competition they are up against.

OpenAI has reportedly offered junior quants compensation packages totaling $3 million. Anthropic has hosted dinners specifically to lure high-frequency trading researchers away from their desks. Jane Street and Two Sigma have responded by restructuring offers, accelerating vesting, and making the case for why finance still has more to offer than whatever Silicon Valley is offering this quarter.

The talent war between Wall Street and AI labs is not new. But in the past eighteen months, it has become something different in character. It used to be a skirmish at the borders, with a handful of researchers each year weighing whether to cross industry lines. Now, it is a systematic effort by some of the most well-capitalized companies in history to pull the same group of people in two directions at once. The people in question, those who can combine deep mathematical training with software engineering fluency and genuine research instinct, are not abundant, and both sides want them.

Neel Somani has been on both sides of that divide. He spent time as a quantitative researcher at Citadel, one of the most selective and highly compensated firms in the world for people with his profile. He then founded Eclipse, a blockchain infrastructure company that raised $65 million. He has since returned to technical research, focusing on problems at the junction of mechanistic interpretability and the formal verification of neural networks, work that sits squarely in the domain that frontier AI labs care about most right now.

When Somani considers where mathematically talented people should direct their careers today, he does not see it as clear. He thinks AI labs are the better bet, and his reasoning runs counter to the traditional framing.

Conventional Wisdom and Its Limits

The standard argument for quant finance goes something like this: the work is intellectually serious, the compensation is immense, the feedback is real, and the firms that do it well have built institutional knowledge and culture that is genuinely hard to replicate. All of this is true.

The standard argument against it, from the AI side, is usually framed in terms of impact or meaning: at an AI lab, you are working on something that matters to the world, whereas at a hedge fund, you are extracting value from markets rather than creating it. This argument is weaker than it sounds, and serious people at quant firms are right to be unimpressed by it.

Somani’s argument is simpler and more straightforward. It is about what the work looks like over time, and where the value of the skills being built compounds most effectively.

In quantitative finance, the system operates within a known, increasingly competitive system. The firms competing for alpha in liquid markets are excellent at what they do. The signals are harder to find than they were a decade ago. The strategies that work get competed down as more capital chases fewer edges. A quant researcher who spends a career at a top firm builds genuine, hard-won expertise, but that expertise is specialized in ways that limit its portability. It translates well within finance, though it does not translate as naturally into the broader economy that is being reshaped by AI.

There is also a structural feature of quant careers that industry observers have noted for years, but that rarely makes it into recruiting pitches. The career arc tends to be front-loaded. Early-career researchers may be highly productive relative to their compensation, generating alpha from genuinely novel methods given their seniority. As seniority increases, compensation rises, but the ability to find new edges does not necessarily keep pace. Even talented people can find that their most productive years come earlier than expected, and that the domain knowledge they have accumulated, while real, no longer provides the leverage it once did.

Mathematicians working in AI labs — photo credit: Tima Miroshnichenko / Pexels

What AI Labs Offer That Finance Doesn’t

The contrast with AI research careers is not primarily about compensation, though the gap that once existed in finance’s favor has narrowed considerably. It is about the nature of the work and what it builds toward.

Frontier AI research is operating in a domain where the science is still being established. The researchers doing the most important work are not optimizing within a known framework. They are figuring out what the framework is. That changes the relationship between experience and productivity. The accumulated judgment of a senior AI researcher, about which problems matter, which approaches are likely to generalize, and where the real obstacles lie, tends to compound rather than decay. This is the opposite of what happens in competitive financial markets, where the edges that made a senior researcher valuable often get competed away by the time their career hits its middle stretch.

There is also the question of what the skills transfer to. Someone who spends several years doing serious research, or working on formal verification of neural networks, or building robust evaluation frameworks for large models, has built knowledge that is relevant across a very wide range of applications. Every sector of the economy is in the process of figuring out how to deploy AI systems, verify their behavior, and know when they fail. The researchers who developed a foundational understanding of these problems during the current period will be in demand in ways that extend well past any single employer.

Somani points to something else as well. The problems being worked on at frontier AI labs are not just technically interesting. They are unsolved in ways that matter. The question of how to comprehend what a large model is actually doing internally, how to certify that a system should behave reliably in a specified domain, and how to intervene when it does not, are questions whose answers will matter at a scale that no trading strategy can match. That is not a moral argument. It is an argument about where the most interesting technical problems currently are.

The Talent Migration

The direction of activity in the broader market supports Somani’s take.

Mehtaab Sawhney, a mathematician at Columbia who has been working extensively with AI tools on research-level mathematics, recently took an academic leave to join OpenAI. Carlo Pagano started a joint position at Google DeepMind earlier this year. These are not people leaving mathematics for industry because they ran out of interesting problems. They are moving because they believe the most interesting problems in their field are now being tackled in AI labs.

On the quantitative finance side, firms have responded to the pressure not by questioning their core pitch but by enhancing its terms. Higher base salaries, accelerated vesting, and more flexibility around what researchers work on have all appeared in recent cycles. The firms are not wrong to fight for talent. The work is genuinely interesting, and the compensation is substantial.

But Somani’s argument is harder to counter with better offer letters. If the underlying dynamic is that quant finance is a mature industry where edges are competed thin, and domain expertise has limited portability outside the sector, then improving the near-term terms of employment does not change the long-term trajectory for someone building a career over decades rather than years.

Where the Argument Is Weakest

Somani’s view is not without its counterarguments, and the strongest ones are worth taking seriously.

AI labs are not uniformly perfect backdrops for technical research. The organizational dynamics of large, well-funded, fast-moving companies can work against the kind of deep, patient work that the hardest problems require. Researchers who join labs during periods of rapid scaling sometimes find that the institutional priorities shift in ways that pull them away from the problems they came to work on. The funding environment for AI, while currently robust, has been volatile in the past and could become volatile again.

There is also a risk of concentration that the AI lab argument tends to understate. The value of the expertise being built at frontier labs is currently highly precise because the field is moving fast, and the number of people who can do the work is small. If the field matures, if the methods stabilize, and the problems become more routine, the premium on frontier AI research skills could compress in ways that parallel what happened to quant finance skills over the previous two decades.

Somani acknowledges these risks. His point is not that AI labs are risk-free, but that the risk-adjusted expected value still favors them for people with the profile to succeed in either environment. The downside of several years at a frontier lab, even if the lab does not become one of the defining institutions of the AI era, is that the skills and research record remain broadly applicable.

The Question Behind the Question

There is a version of the debate that is really a question about what kind of life a technically talented person wants to build and what problems they find most compelling. That question does not have a universal answer.

But Somani’s framing suggests that for a significant portion of people weighing this choice, the decision is being made based on outdated information. The compensation tables that made finance the obvious answer a decade ago no longer look the same. The skills being built at frontier AI labs no longer transfer narrowly into a single industry. The problems being worked on are no longer a step removed from what matters in the broader economy.

The talent war, with hedge funds hosting recruiting dinners and AI labs offering packages that would have been unimaginable five years ago, is a consequence of all of this coming to pass simultaneously.

FAQs

Why are mathematicians leaving Wall Street for AI labs?

Mathematically skilled professionals are drawn to AI labs because of the opportunity to work on foundational, unsolved problems with broader real-world applications. The combination of competitive compensation and intellectually expansive work makes AI research an increasingly attractive alternative. This shift reflects a broader trend where long-term career value outweighs short-term financial incentives alone.

How does compensation compare between quant finance and AI research?

Historically, quant finance offered significantly higher compensation, but AI labs have rapidly closed that gap with aggressive offers. Some AI firms now provide multi-million-dollar packages to attract top talent. This parity has removed one of the biggest barriers preventing talent migration.

What makes AI research skills more transferable than finance skills?

AI research builds knowledge applicable across industries, from healthcare to manufacturing and beyond. In contrast, quant finance expertise is often highly specialized within financial markets. This broader applicability gives AI professionals more flexibility in shaping their long-term careers.

Is quant finance still a good career path?

Yes, quant finance remains a highly respected field offering strong compensation and intellectually challenging work. However, it operates in a mature environment where competitive advantages are harder to sustain over time. Individuals must weigh whether they prefer stability within a defined system or exploration in a rapidly evolving field.

Are there risks to choosing a career in AI research?

AI research comes with uncertainties, including shifting organizational priorities and potential market volatility. The field may also mature over time, potentially reducing the premium on specialized skills. Despite these risks, many believe the long-term opportunities still outweigh the downsides.

Learn more: https://www.neelsomani.com/