ML-heavy vs maths-heavy quant firms: what they say, and what it means for your interviews

One quant firm says on its careers site that "deep learning is the future of quantitative trading." Another publishes an engineering post saying that "if a linear model is as good as a random forest model, we'd prefer the linear model." Read quickly, that looks like two camps: the machine-learning shops and the statistics shops. Read carefully, it is mostly two ways of describing the same discipline, with real differences in emphasis that do change who gets hired and how they are interviewed.

This guide sticks to what firms say publicly. Every claim is labeled as official (a firm's own site, job posting or press release), reported (press), or self-reported (candidate forums), with the year. Nobody outside these firms knows how their models actually work, and we do not pretend to.

Last reviewed 2026-10-04. Firm messaging, compute numbers and interview formats change often, so read this as a dated snapshot of public evidence. It does not rank firms.

A spectrum of positions

Every serious quant firm uses statistics, and almost every one now mentions machine learning somewhere. The useful question is what a firm chooses to lead with when it describes itself to candidates, because that framing tends to show up again in its job requirements and interview loops.

Below: where those labels come from, and why the middle of the table is crowded.

The firms that lead with machine learning

XTX Markets is the clearest example. Its homepage says it "uses state-of-the-art machine learning technology to produce price forecasts for over 53,000 financial instruments", lists "over 25,000 GPUs in our research cluster" and "over 1 Exabyte of usable storage", and gives a headcount of 300 employees. (Official, homepage, accessed 2026-10-04.) That cluster has grown in public: a February 2024 press release described "100,000 cores and 20,000 A/V100 GPUs". (Official, 2024.) In January 2025 XTX announced plans to invest over €1bn in a data centre project in Finland, with the first building targeted for completion in 2026; its homepage currently says it "is also constructing a large-scale data centre in Finland". (Official, 2025 and 2026.)

The same February 2024 release launched XTY Labs in New York, an ML division with an AI residency offering contracts of 6 to 12 months. In the release, Alex Gerko said the programme was "a fantastic opportunity for some to eventually transition into the core quant team at XTX." (Official, 2024.) Notably, XTX also says its teams have "backgrounds in pure math, programming, physics, computer science and machine learning", so ML-first does not mean ML-only. More detail on the firm is in our XTX Markets guide.

Jane Street puts it bluntly on its machine-learning page: "We believe deep learning is the future of quantitative trading." It describes "Tens of Thousands of high-end GPUs", asks candidates to "think of Jane Street as a research lab with a trading desk attached to it", and recruits three distinct ML profiles: ML researchers, ML research engineers, and ML performance engineers. (Official, accessed 2026-10-04.) eFinancialCareers reported in September 2026 that the firm, long known for OCaml, introduced Python, "which has been very useful for the firm's machine learning efforts", citing CTO Ron Minsky on the firm's Signals and Threads podcast. (Reported, 2026.)

G-Research describes its research as "blending mathematical precision with state-of-the-art machine learning" and says researchers apply "cutting-edge machine-learning techniques, whether drawn from the latest research or developed in-house." (Official, accessed 2026-10-04.)

Hudson River Trading is a useful signal from the accounts side. eFinancialCareers reported in September 2026 that HRT's UK services entity recorded £391m of additional spending on property, plant and equipment, which eFinancialCareers put at about £2.4m per head across HRT's UK staff, and noted that now-expired HRT job postings mentioned "very high GPU-to-research ratios". (Reported, 2026, from UK company accounts.)

The firms that lead with statistics, economics and method

D.E. Shaw describes its quantitative analysts as people who "apply mathematical techniques and write software to develop, analyze, and implement statistical models for our computerized financial trading strategies", and says the firm "quickly became a pioneer in computational finance." (Official, accessed 2026-10-04.) More in our D.E. Shaw guide.

Man AHL, founded in 1987, says markets "exhibit persistent anomalies, such as price trends, mean reversion, carry or other repeatable patterns, which can be identified through careful statistical analysis." Its own timeline lists "Machine Learning" as a 2014 milestone, and it highlights its links with the Oxford-Man Institute. (Official, accessed 2026-10-04.) A statistics-first shop that has used ML for over a decade is a good reminder that the labels overlap.

Winton says it has pioneered trend following "since 1997" and that "all our strategies are rooted in our belief that original research can provide an investment edge", describing a culture of "methodological rigour". AQR describes its evolution as taking place "at the nexus of economics, behavioral finance, data and technology." (Both official, accessed 2026-10-04.) Neither page leads with ML.

Two Sigma sits in the middle. Its homepage leads with "rigorous scientific inquiry", but also says "AI is embedded across our platform" and that researchers build models "using mathematics, advanced techniques, generative AI, and deep domain expertise." Its campus quant researcher posting asks you to "use a rigorous scientific method" and to "apply quantitative techniques like machine learning to a vast array of datasets." (Official, accessed 2026-10-04.) See our Two Sigma guide.

PDT Partners ("Predictions based on data", with ideas "peer reviewed, and empirically validated") and WorldQuant, whose Mumbai quant researcher posting requires "Knowledge of Linear Algebra, Statistics, Machine Learning", fit the same middle ground. (Official, accessed 2026-10-04.)

Why the line is blurrier than it looks

The most candid public writing on this comes from HRT's engineering blog. In a 2022 post titled "In Trading, Machine Learning Benchmarks Don't Track What You Care About", Iain Dunning argued that "an incremental 0.1% improvement in the accuracy of a neural network's ability to distinguish dog breeds might not translate to predicting the price of a stock." He added that "it is easy to lie to ourselves while resolving small differences in our low signal-to-noise domain", and that "if a linear model is as good as a random forest model, we'd prefer the linear model." (Official blog, May 2022.) A 2023 HRT post on modelling equity returns with linear factor models says the goal is not "the fanciest model of all" but "something simpler and more reliable." (Official blog, March 2023.)

So the ML-forward firms talk constantly about statistical discipline, and the statistics-forward firms list ML in their requirements. The genuine differences that show up in public material are narrower:

• Compute as a strategic asset. XTX and Jane Street publish GPU counts. eFinancialCareers reported in September 2026 that "Jump Trading, QRT, XTX and Jane Street have data centres for their own use", that Jane Street had entered a 15-year data centre lease in Oklahoma, and that Millennium hired a global head of data centres from Citadel. (Reported, 2026.)
• Who the research hire is. G-Research and Jane Street publicly recruit researchers who look like academic ML scientists, plus the engineers who keep large training runs fast.
• How the interview is split. At least one firm openly runs a separate ML track (below).

What it means for hiring profiles

ML-forward research roles lean academic. G-Research says its researchers often join "after completing PhDs or postdoctoral work, with publications at the world's most prestigious conferences", and adds: "There's no need for experience in finance." (Official, 2026.) XTX's residency, as announced in 2024, was framed as a route into its core quant team. (Official, 2024.)

ML also creates engineering demand. Jane Street's split into researcher, research engineer and performance engineer is explicit. On 2026-10-04, XTX's public Greenhouse board listed eight openings, including datacentre operations roles in Finland and Singapore and a C++ engineer, and no research roles. (Official, snapshot of a live board; research hiring may run through other channels.) If you are a strong systems engineer, the GPU build-out works in your favor.

Statistics-forward roles cast a broad technical net. D.E. Shaw looks for "top students in their respective math, statistics, physics, engineering, computer science, and other technical and quantitative programs". (Official job ad, accessed 2026-10-04.) Two Sigma asks for "experience performing an in-depth research project, examining real-world data." (Official, 2026.) Susquehanna quotes a Dublin researcher saying "the ability to develop a new idea or approach to a problem is far more valued than having a preexisting knowledge of finance." (Official, 2026.)

What it means for interviews

The clearest public example of an ML split is G-Research. Candidates take "either a general quantitative aptitude assessment, or an ML specific one, depending on your background." Standard candidates "typically" sit four one-hour interviews, "one of which will focus on in-depth technical questions in mathematics"; ML candidates "complete two one-hour interviews that focus on your ML knowledge" and "should also expect questions on mathematics, programming and statistics." (Official, accessed 2026-10-04.)

Statistics-forward firms publish topic lists that read like a probability and statistics syllabus. D.E. Shaw says questions "may cover probability, mathematical statistics, algorithms, logical thinking, and/or programming", and that it assesses "how deeply you understand their uses and limitations." Two Sigma lists "data analysis/open-ended problem solving", "coding and algorithms", and "statistics or your research domain for PhDs." (Both official, accessed 2026-10-04.)

Candidate accounts suggest the maths does not disappear at ML-forward firms. On a June 2026 Blind thread about an XTX software engineering onsite, one commenter recalled interviewing there for a quant role about ten years earlier as "still among the hardest set of interviews I had, with actual maths and physics puzzles like in college." (Self-reported, 2026, a single anecdote about an old process.)

A practical reading of all this:

1. Statistics is the shared core. Every interview guide quoted above names statistics or mathematics, whatever the firm leads with, and D.E. Shaw lists probability explicitly. The problem bank has dedicated probability, statistics and machine learning categories if you want to drill them.
2. Know linear models cold. Regression, regularisation and factor models are what HRT's own posts reach for first, and they are where "limitations" questions live.
3. For an ML track, depth beats breadth. Be ready to defend your own research, and to explain how you would validate a model on noisy, non-stationary data without fooling yourself.
4. Both sides require code. Every interview guide quoted above names programming or coding.

How to read a firm's positioning

A homepage is a recruiting message. The more reliable signal is the posting and interview guide for your specific role, which is why the sections above quote them. If a firm runs separate ML and general tracks, pick the one your strongest evidence supports, even if the other sounds more fashionable, and ask your recruiter which topics your loop will cover.

QuantReady is an independent prep platform and is not affiliated with, endorsed by, or sponsored by any firm named here. All trademarks belong to their respective owners. Nothing here describes any firm's actual models or strategies, which are not public. If something is out of date, tell us via /contact.

QuantReady is an independent prep platform and is not affiliated with, endorsed by, or sponsored by any firm we write about. All company names and trademarks belong to their respective owners. Read how we source our content.