Bloomberg Data Science PhD Fellowship: $45k Stipend + Full Tuition (Next Round Not Yet Announced)

Status as of 30 July 2026: applications are closed. The Bloomberg Data Science Ph.D. Fellowship for the 2026–2027 academic year closed on 14 December 2025 (11:59 PM AoE), and Bloomberg said recipients would be informed by 9 February 2026. Bloomberg has not announced a 2027–2028 cycle on the official programme page. There is therefore no new deadline or application process to follow yet; do not treat a date from an aggregator as an official Bloomberg deadline.

The durable place to watch is Bloomberg’s own programme page. The verified canonical address is:

https://www.bloomberg.com/company/values/tech-at-bloomberg/data-science/academic-engagement-programs/data-science-ph-d-fellowship/

The rest of this page records what the 2026–2027 fellowship paid and what that closed application required. Treat it as an archive of the last confirmed cycle, not as instructions for an open application.

What the fellowship is

If you are doing a PhD in AI, machine learning, or data-heavy computer science, this is a substantial industry fellowship. In the 2026–2027 round it covered 100% of tuition, provided a USD 45,000 stipend, offered Bloomberg mentors, and required a paid 14-week research internship in New York, London, or Toronto. The fellowship could be renewed annually for up to three years at Bloomberg’s discretion.

The programme is aimed at doctoral research connected to Bloomberg’s data science and AI work. The official topic list includes large language models, information retrieval, AI for finance, and trustworthy AI, among other areas.

Terms from the 2026–2027 round

Note the eligibility window carefully. “Graduation by the end of 2029” was written for the 2026–2027 cohort. If Bloomberg runs a 2027–2028 round, that window will almost certainly shift forward by a year — so do not rule yourself in or out on the 2029 number until the new call is published.

What the money actually buys you

Full tuition coverage. Your university tuition was covered in full for each year you held the fellowship. That removes the pressure to teach multiple sections or pick up side work just to stay afloat, and it lets you plan a multi-year research arc instead of surviving semester to semester.

A $45,000 stipend. On top of tuition. Bloomberg says the stipend is intended for living expenses, professional conferences, and research expenses such as computer hardware. The official description does not prescribe a detailed personal budget, so applicants should confirm any university or tax treatment before relying on a specific use.

A paid 14-week research internship. Each year of the fellowship required a summer internship at Bloomberg in New York, London, or Toronto. Bloomberg describes the internship as paid and research-focused. Treat the location and work-authorisation requirements as part of the funding decision, not as an optional extra.

Mentorship and career support. Bloomberg lists mentors and career counseling alongside the financial award. The page does not promise a particular mentor, publication outcome, job offer, or research result, so applicants should view this as structured professional support rather than a guaranteed placement.

Renewal for up to three years. Support was not limited to one year, but renewal was annual and explicitly at Bloomberg’s discretion. The three-year maximum is not a promise that every recipient will receive three years of funding.

Research areas Bloomberg funds

Bloomberg is specific about what it wants. To be competitive you either work in one of these areas or make a clear case that your work connects to one. From the last round’s call:

• Agentic AI and LLM reasoning — systems where models reason, plan, and act over time rather than predicting the next token.
• Agent evaluation and LLM judges — methods for models evaluating other models’ outputs, quality, safety, or correctness.
• Code generation, semantic parsing, Text2API — turning natural language into programs, queries, or API calls that actually execute.
• Human-AI interaction and data annotation — how people work with models, and better ways to label and curate data.
• Information retrieval and conversational systems — search, recommendation, and chat over large, complex information spaces.
• Knowledge graphs and structured reasoning — explicit relationships supporting reasoning and query answering.
• LLM post-training, domain adaptation, and alignment — fine-tuning, instruction-tuning, finance-specific specialisation, and safety work.
• Multi-modal models for document understanding — text, tables, and figures handled together, which is directly relevant to filings and news.
• Summarization and content generation — condensing long reports, earnings documents, and news, or generating controlled content.
• Time-series modeling and AI for finance — prediction over markets, volatility, and risk, and decision support.
• Trustworthy AI and interpretability — inspecting and explaining model behaviour in regulated, high-stakes settings.

Adjacent work counts. Reinforcement learning for portfolio optimisation, safety-aware sequence modeling, or evaluation frameworks for agent systems all sit close enough — you just have to draw the connection explicitly rather than hoping a reviewer draws it for you.

Who this is for

The 2026–2027 call was written for serious PhD researchers, not people generally interested in data science. Under that round’s rules, a strong candidate was someone who:

• would be a full-time doctoral student during the funded academic year;
• expected to finish the PhD by the end of 2029;
• worked somewhere in AI, ML, data science, NLP, IR, or AI for finance;
• could write for a technical audience and propose concrete research directions;
• could spend 14 weeks each summer in New York, London, or Toronto.

Full-time status had to be maintained throughout. Dropping to part-time or leaving the programme ended the remaining support.

The fellowship was open internationally, but work authorisation was the real constraint. Bloomberg does not sponsor a U.S. visa for this programme. To intern in New York you needed self-sponsored authorisation covering the full period, such as F-1 OPT or another EAD. London and Toronto placements carried their own local work requirements; your university’s international office is the right first stop for assessing feasibility.

Disqualifiers in the last round included holding another industry fellowship concurrently, working outside your university during the fellowship without Bloomberg’s permission, and being unable to commit to the internship or the annual kickoff event at Bloomberg’s New York headquarters.

What the last round asked you to submit

The application was short, which makes each piece heavy:

• A CV. Research-centric, not a job resume. Publications and preprints with clear author order and venue, talks and posters, open-source projects with links, awards.
• A two-page research proposal, excluding references, in Bloomberg’s specified template. This carried the application.
• A reference letter from your advisor, required. An optional second referee — a collaborator or previous internship supervisor — was allowed. You entered referee contact details during submission, the system invited them, and letters had to arrive by the application deadline.

Bloomberg was explicit that applications ignoring the formatting guidelines would not be reviewed. That is a cheap way to lose, and entirely avoidable.

How to use the wait

There is no deadline to work backward from right now, which is genuinely an advantage if you use it. The applicants who do well on a six-week call are the ones who did the thinking in the preceding six months.

Now through the autumn:

• Confirm the structural facts you control — your expected graduation date, whether you will be enrolled full-time next academic year, and whether you could legally work a summer in New York, London, or Toronto without sponsorship. The visa question kills more applications than weak writing does, and it takes months to resolve.
• Talk to your advisor about whether this fellowship fits your trajectory, and get a provisional yes on the letter. A letter written by someone who has read your proposal twice reads very differently from one written in a weekend.
• Read recent Bloomberg research publications and talks in your area. The proposals that land are the ones that connect to work Bloomberg is already doing.
• Draft a one-page informal version of your idea and get it in front of your advisor and one peer. Scope it to something ambitious but achievable in one to three years.

When a call appears:

• Download the required template on day one and write directly in it rather than reformatting later.
• Expand the one-pager into the full proposal, then circulate it for comments with real time to act on them.
• Tighten your CV around research output.
• Give your referees the near-final proposal plus your CV, and the exact deadline.
• Submit at least 48 hours early and verify in the system that letters have been received. Nobody wants to be debugging a PDF upload at 23:57.

What makes a proposal competitive

Bloomberg does not publish a rubric, but the shape of the programme tells you what reviewers weigh.

Research quality and originality. Is the work substantively new rather than a small tweak, grounded in the relevant literature, and likely to produce results in serious venues? Bold but technically credible beats vague ambition.

Relevance to Bloomberg’s problems. They sit at the intersection of finance and markets, news and media, and large-scale data platforms and search. Do not write “this advances machine learning theory” and leave it there. Say how it could improve search across millions of documents, summarisation of financial or news content, forecasting in markets, or interpretability in a domain where opaque models carry regulatory risk.

Feasibility. Can you actually do this, in your situation, in a few years? Are the goals scoped for a PhD? Do you have the compute, data, and supervision? Do you name the risks and outline fallbacks? Promising to solve AGI and all of finance in two years is a fast way to lose credibility.

Track record, or clear potential. Mid-PhD applicants should show publications or submissions to serious venues, visible artifacts like libraries or datasets, and evidence of finishing things. Earlier applicants who do not have first-author papers yet should lean on strong technical coursework, prior research experience, preliminary experiments, and a sharp proposal.

Fit with the internship. Your project should benefit from a summer inside Bloomberg, not be something you would set aside for 14 weeks every year. Reviewers favour proposals where the internship and the dissertation reinforce each other.

Write for a technical audience that is not in your exact sub-subfield. Define specialised terms on first use, include equations only when they clarify, and skip the acronym pile-up. If a colleague in a nearby area can read your proposal and explain it back to you, it is in good shape.

Frequent mistakes

Ignoring the formatting rules. Bloomberg said upfront that non-conforming applications may not be read. Write in the template from the start.

A generic ML pitch. “Here is my idea for a better transformer,” with no connection to Bloomberg’s context, is weak. Add at least one concrete use case in finance, news, search, or large structured data.

Jargon overload. If the proposal reads like a stack of paper titles glued together, reviewers disengage.

Late or thin reference letters. A rushed, generic letter sinks applications quietly. Ask early and give your referee material to work with.

No stated “so what.” Describing a method without explaining why it matters, scientifically or practically, is the classic failure. Say what the impact is at the theoretical, methodological, and application levels.

Questions people ask

Is there any way to apply right now? No. The 2026–2027 round closed on 14 December 2025, and Bloomberg said recipients would be informed by 9 February 2026. There is no open call. Any site advertising a future Bloomberg fellowship deadline is guessing until Bloomberg publishes the next cycle.

When will the next round open? Bloomberg has not said. Watch the official programme page for a new announcement rather than assuming that the previous schedule will repeat.

Do I need to have worked with Bloomberg before? No. What matters is that your topic is relevant to their work: complex data, AI, LLMs, retrieval, finance, news, time series, interpretability.

Can I hold this alongside other funding? Not alongside another industry fellowship. Non-industry funding such as government or university scholarships may be permitted case by case, which you would need to clear with Bloomberg.

Is it only for students in the U.S.? No geographic restriction is stated in the archived eligibility criteria. Bloomberg will not sponsor a U.S. visa, so a New York internship requires self-sponsored authorisation such as F-1 OPT covering the full period. London and Toronto have their own requirements.

Can I pick my internship city? The last call specified one of New York, London, or Toronto. Placement in practice depends on project fit, team needs, and your work status. Preferences are preferences.

What if I cannot do the internship one summer? It was a requirement for each year of the fellowship. Assume renewal and continued funding could be affected, and raise it with Bloomberg directly rather than after the fact.

Do I need publications? They help, particularly in relevant venues. Early-PhD applicants can still compete on strong ongoing work, a sharp proposal, and a substantive letter.

Will I get feedback if rejected? Nothing in the call promises it. Assume not, which is another reason to gather as much pre-submission feedback as you can from your advisor and peers.

Track the next call

Check Bloomberg’s programme page directly rather than relying on aggregators:

https://www.bloomberg.com/company/values/tech-at-bloomberg/data-science/academic-engagement-programs/data-science-ph-d-fellowship/

The 2026–2027 call listed [email protected] as the contact address for questions about the programme, which is the reasonable place to ask whether a new round is planned.

This is a hard fellowship to win, and the preparation is not wasted even if you do not. Writing a two-page proposal that has to justify its own significance to a reviewer outside your subfield tends to improve the dissertation underneath it.