The NeurIPS Booth Is Where Quants Close. Source Its 9,108 Papers First

If you're hiring ML researchers for a startup or a non-finance AI team, the quant firms you're competing with started recruiting in August, and the NeurIPS 2026 accepted-paper list is already public. You have roughly five weeks before the Sydney site opens on Dec 6, and that is the only stretch where you get first conversations instead of fighting over final decisions.

Bloomberg reports that quant firms including CFM, QRT and Jump are funding PhDs and building AI labs. Recruiter estimates put first-year ML PhD packages at $450K to $650K, and nearly half of NeurIPS 2026's top-tier sponsors are financial firms.

The NeurIPS booth is where quant firms close, not where they start

Quant firms opened their summer 2027 internship and 2027 full-time graduate applications in August 2026 and review them on a rolling basis. So by December, most students they want have already interviewed. The booth is where they close candidates, not where they find them.

Jane Street, Citadel, Hudson River Trading, Two Sigma, D.E. Shaw, Optiver and SIG all opened applications in August. Rolling review means the earliest applicants get the earliest slots. A strong fourth-year PhD who applied in September could be holding an offer by the time NeurIPS starts.

The sponsor list shows how fast this has grown. Diamond is NeurIPS's top sponsorship tier.

• NeurIPS 2024: 5 of 17 Diamond sponsors were trading or finance firms (Citadel, G-Research, Jane Street, Jump, D.E. Shaw). That's about 29%. Optiver and ADIA sponsored at Platinum.
• NeurIPS 2026: nearly half of top-tier sponsors are financial firms, per Bloomberg.

If you start NeurIPS 2026 recruiting at the conference itself, you're turning up for the last round of a contest that began in August.

What ML PhD new grad compensation looks like in 2026

Bloomberg, citing recruitment-industry estimates, puts first-year quant packages for ML PhDs at $450,000 to $650,000, with some reaching $800,000. These are recruiter estimates, not figures the firms disclosed, so read them as the top of the market rather than the norm.

$650K

The more interesting part is how the money gets spent. Quant firms aren't only raising cash offers. They're building things that look like academic labs:

• QRT Labs (Qube Research & Technologies) sponsors more than 70 PhD and postdoctoral researchers.
• CFM's Paris lab, where CPO Laurent Laloux leads on the product side, is set up so researchers "are not under the pressure to build alpha models."
• HRT AI Labs (HAIL) and XTX's AI lab are both named by Selby Jennings' Karysca Gill.
• Millennium Management is building an AI lab under Head of AI Gideon Mann, with an ML team of about 50.
• Jump Trading funds academic fellowships, and G-Research has a dedicated head of academic outreach, Charles Martinez.

Here's why that matters to you. Trading firms already win on salary. Building research labs is an admission that salary alone doesn't land top ML PhDs. Candidates also want to publish, be seen, and work without a P&L watching them. Startups can offer that by default.

How many ML PhDs can you actually source?

At least 2,701 across the US, UK and France, before you open a single paper. Refolk's index of professional profiles has 1,465 US profiles with a PhD student, candidate or doctoral researcher title and a Machine Learning skill, and another 1,236 across the UK and France.

Three things jump out:

• New York is the most common US region in the sample, and most US quant firms are based there. Your US candidates live where quant recruiters already are.
• Europe is nearly US scale. The UK and France pool is 84% the size of the US pool. London and Paris are home turf for G-Research, XTX, QRT and CFM.
• The finance side is already big. 2,244 people with ML skills already hold quant or ML researcher titles at finance firms, nearly one for every PhD student in the combined pool. That ratio is rough, but it shows how much ML research capacity the trading industry has already absorbed.

One caveat: these counts only include profiles that list Machine Learning as a skill, so they're lower bounds. The regional splits come from a 25-profile sample.

Why arXiv-only sourcing misses 42% of NeurIPS 2026

alphaXiv, which tracks arXiv preprints by conference, links only 5,265 of NeurIPS 2026's 9,108 accepted papers. So a search that stops at arXiv misses about 3,843 papers, roughly 42% of the list.

42%

You should know the paper count itself is disputed. alphaXiv reports 9,108 accepted papers. Another tracker, csconfstats, reports 7,900 accepted out of 30,709 submissions, a 25.7% acceptance rate. Even using the lower count, about a third of papers have no linked preprint. The conclusion doesn't change: the search tools most recruiters use stop at arXiv, and arXiv isn't the whole conference.

The full source is the OpenReview accepted list, the public peer-review record NeurIPS uses, plus university news pages that list their accepted papers.

A five-week playbook to hire PhD students graduating 2027

Sourcing AI researchers from conference papers means turning a list of author names into verified, contactable, graduating people. Here's how to do that before Dec 6.

1. Pull the full OpenReview accepted list, not the arXiv subset. Save it as a sheet with title, authors and affiliations.
2. Filter to student authors. First authors at universities are usually PhD students. Check OpenReview profiles for "PhD student" and look at the advisor field.
3. Check graduation timing. Lab pages and personal sites often say "on the job market" or give an expected graduation date. A fourth- or fifth-year student is your 2027 hire. A second-year student goes on your internship list.
4. Match each name to a profile. Link the OpenReview name to a lab page, a GitHub account and a LinkedIn profile. Common names make this the slowest step.
5. Rank by geography and fit. Decide whether you're competing in New York, London and Paris (quant territory) or recruiting more widely.
6. Write to them about the paper. Mention a specific result. Generic "saw your NeurIPS paper" messages get treated like spam.

Take one real example. "Recursive Language Models" by Alex Zhang, Tim Kraska and Omar Khattab (MIT) is a NeurIPS 2026 paper on alphaXiv. From there you'd confirm which author is the student, find their lab page and GitHub, and read the repo before you write anything. That's real work for one paper, and the list has thousands of them.

Step four is where most teams stall, and it's the exact gap Refolk closes. You describe the person in plain English and get back matched profiles across GitHub, LinkedIn and the open web, so you don't have to cross-reference OpenReview names by hand.

Even with good tools, five weeks is tight for a seed-stage company with no recruiter. The founder is still shipping product, the role is a research hire nobody on the team has hired for before, and an agency fee on a research salary is a big hit to a small budget. Hard-to-fill roles fall apart in exactly this situation: everyone agrees the hire matters and nobody has the hours to source it.

Which NeurIPS 2026 site puts you head to head with quants?

NeurIPS 2026 is split across three sites. Paris and the London talent pool behind it are where startups compete most directly with quant firms. Sydney, the main site, probably overlaps the least.

Organizers accepted 102 workshops out of 454 valid submissions, and Sydney got nearly half of them. Workshop authors belong on your sourcing list too, not just main-track authors.

In Europe, you're competing head on. The UK and France pool is 84% the size of the US one, with Inria and Télécom Paris as the top employers. CFM, QRT, G-Research and XTX all recruit in that same pool, so contact Paris authors in early November, not during the conference week.

How to recruit machine learning PhDs without matching $650K

Don't try to beat the salary. Offer what quant firms are spending money to imitate. Bloomberg says AI labs win on "prestige, visibility and, in some cases, the prospect of a windfall from owning a stake." A trading firm can copy the first two only partly and the third hardly at all.

Quant firms are building labs to look like academia. Startups already look like that.

What to put in the first message:

• Freedom to publish. Say clearly that the candidate can keep publishing at venues like NeurIPS. CFM advertising a lab without alpha pressure shows how much this matters.
• Equity with real upside. For some candidates, a stake beats a fixed bonus. Spell out the vesting and the strike price.
• Public credit. Their name on papers, blog posts and open-source repos, instead of work that stays inside a trading firm.
• Ownership of the problem. One research direction they own, not a slot in a team of 50 like Millennium's.
• The research itself. Name the open problem their NeurIPS paper connects to. That's the line that shows you actually read it.

Why a quant offer isn't the end of the pipeline

An author who signs with a quant firm in 2027 is still worth contacting again 12 to 24 months later. Researchers move between trading firms and AI labs in both directions.

Bloomberg's reporting names several moves:

• HRT, which has run an AI lab for years, has lost several employees to OpenAI.
• Two Sigma quants have moved to Anthropic.
• A few Google DeepMind scientists have joined Jump and Citadel Securities.

Use the NeurIPS 2026 author list as a long-running list, not a one-off campaign. Tag the people who sign with quant firms, and set reminders for 12 and 24 months after they sign. By then they know exactly what the secrecy of a trading firm costs them, and your pitch on publishing and equity lands differently.

FAQ

When should startups start NeurIPS 2026 recruiting?

Now. Quant firms opened 2027 graduate applications in August 2026 and review on a rolling basis, so the strongest candidates are already in their pipelines. The five weeks between early November and the Dec 6 Sydney opening are your chance to start a first conversation. After that, you're competing against offers that are already on the table.

How many papers did NeurIPS 2026 accept?

Sources disagree. alphaXiv reports 9,108 accepted papers, of which 5,265 have linked preprints. csconfstats reports 7,900 accepted out of 30,709 submissions, a 25.7% acceptance rate. Either way, a large share of papers has no arXiv preprint, so source from the OpenReview accepted list.

What is ML PhD new grad compensation at quant firms in 2026?

Recruitment-industry estimates reported by Bloomberg put first-year packages at $450,000 to $650,000, with some reaching $800,000. These aren't figures the firms disclosed, and they describe the most aggressive buyers, so treat them as the ceiling you're competing against rather than a number you need to match.

Where do I find graduating ML PhDs besides arXiv?

Start with the OpenReview accepted list. Add university news pages (Warwick published all 20 of its accepted papers with authors), lab pages that say who is on the job market, and GitHub repos linked from papers. In Refolk's index there are 1,465 US and 1,236 UK and French ML PhD profiles to cross-check those names against.