When Hugging Face shipped the open-source library that became the de facto standard for AI model collaboration, the founders couldn’t have predicted it would also become an HR problem. The company that started as a chatbot for teenagers in 2016 is now a $4.5 billion AI platform, and somewhere along the way, paying its global workforce stopped being something a startup CFO could handle from a spreadsheet.
By 2024, Hugging Face was running global contractor payments through five different systems, splitting payouts across its U.S. and French entities, and processing every payment individually each month. According to the case study, this fragmentation was slow, error-prone, and impossible to scale alongside the company’s rapid AI-driven growth.
The fix was a consolidation play similar to what GitLab did at the EOR layer , except Hugging Face wasn’t consolidating Employer of Record providers, it was consolidating contractor payment systems. The destination: one platform, Deel, handling everything. The result, as published in the Deel case study: a 50% reduction in payment processing time, five systems collapsed into one, and 70 workers onboarded without breaking compliance across either of Hugging Face’s legal entities.
In this article, you’ll learn the Hugging Face EOR and global payments case study — what they were trying to solve, why a multi-entity AI startup is uniquely difficult to pay, what Deel actually did differently, and the lessons any AI company hiring globally can take from it.
Hugging Face Timeline
Clément Delangue, Julien Chaumond, and Thomas Wolf found Hugging Face in New York City. The company is named after the 🤗 emoji and launches as a chatbot for teenagers.
Angel round of $1.2M led by Betaworks kicks off operations and early product development.
$4M seed round expands the team and accelerates product development.
After open-sourcing the model behind the chatbot, the company pivots from consumer chatbot to ML platform. Series A of $15M led by Lux Capital follows.
$40M Series B (Addition) funds enterprise expansion. The Transformers library becomes the standard interface for the open-source AI community.
$100M Series C (Lux Capital) accelerates the model-hub strategy.
$235M Series D in August led by Salesforce Ventures with Google, Amazon, and Nvidia values the company at $4.5 billion. Total raised crosses $395M.
With a workforce split across U.S. and French legal entities plus a subsidiary in Canada, Hugging Face is paying contractors through five different systems, individually each month. The administrative load becomes unsustainable.
Hugging Face migrates global contractor payments to Deel. According to the Deel case study, this centralizes five systems into one, cuts payment processing time by 50%, and onboards 70 workers onto the new platform.
Why a Multi-Entity AI Startup Is Uniquely Hard to Pay
Most “global payments” case studies are about companies hiring across many countries through a single home entity. Hugging Face’s problem was the inverse: a single workforce being paid through multiple home entities. That distinction matters because it determines which problem you’re actually solving.
Hugging Face is a New York-based startup with subsidiaries in Canada and France, as documented. That structure exists for two reasons familiar to any AI company:
1. The founders are French. Clément Delangue grew up in La Bassée in northern France, and Thomas Wolf and Julien Chaumond are also French. A meaningful share of the engineering team is in France — including some of the most senior research talent. A French subsidiary isn’t optional; it’s where a chunk of the IP gets built.
2. The U.S. is where AI customers and investors live. Hugging Face raised from Salesforce, Google, Amazon, and Nvidia: all U.S. companies. The commercial center of gravity sits in New York, and the U.S. parent entity is what investors fund.
That dual U.S./French structure creates a payments problem most startups never face. Every contractor payment has to route from the correct entity (U.S. or French) to comply with intercompany transfer pricing rules, French URSSAF social contributions, U.S. 1099 reporting, and FX exposure on EUR/USD conversions.
A French contractor paid from the U.S. entity creates tax problems; a U.S. contractor paid from the French entity creates different ones. Multiply that by 70 contractors and five different payment tools, bank transfers here, Wise there, a separate platform for invoice management, another for tax forms and a CFO’s monthly close becomes a multi-day reconciliation exercise.
This is the problem Deel was hired to solve. Not “we need to hire in a new country”, but “we need one system that respects our existing multi-entity structure and stops the monthly fire drill.”
The Real Problem: Five Systems Was Worse Than No System
Deel’s study is unusually direct about what wasn’t working: “Before Deel, Hugging Face used five different tools to pay their workforce, handling all payments individually each month.” That sentence buries five distinct operational failures inside one description.
1. Individual monthly payments at scale don’t compose. Paying 70 people manually, one transaction at a time, every month is roughly 840 discrete payment actions per year. Each one is a chance for a typo in an account number, a wrong currency, or a missed payment cycle.
2. Five systems means five reconciliations. At month-end, finance had to pull reports from five separate tools, normalize them into one chart of accounts, and reconcile against bank statements. For a public-quality finance function at a Series D company, this is unsustainable.
3. Compliance lives in the gaps between systems. When a French contractor’s tax status changes, which of the five systems updates? When a 1099 needs to be generated for a U.S. contractor, which system has the authoritative invoice history? The answers were “it depends” — and “it depends” doesn’t pass an audit.
4. Customizable contracts didn’t survive the handoff. Hugging Face has strict legal and HR policy requirements typical for a company holding the open-source AI community’s trust. As Anna Tordjmann, Chief of Legal at Hugging Face, put it: “Deel gives us the ability to customize agreements and amendments to fit our strict requirements, which is invaluable.” The legacy stack didn’t.
5. The workforce was growing faster than the operations team. With $235M of Series D capital and a $4.5B valuation, Hugging Face was hiring aggressively. A payments stack that already strained at 70 workers was going to break at 200.
The Selection Process: What Hugging Face Needed in a Global Payments Platform
Hugging Face appears to have evaluated against five criteria:
1. Multi-entity support out of the box. Any platform that required Hugging Face to flatten its U.S./French/Canadian entity structure was a non-starter. The new system had to route each payment from the correct entity automatically.
2. Customizable contracts and amendments. This is the criterion Anna Tordjmann named explicitly. A legal team with strict policies cannot use a vendor whose contract templates are take-it-or-leave-it.
3. Single platform for the full payment lifecycle. Contract → invoice → payment → tax form → audit log, all in one tool. Not five.
4. Global currency and country coverage. Deel’s platform supports payments in 150+ countries and 200+ currencies according to Deel’s own documentation. For an AI company recruiting researchers wherever they live, country coverage is the table-stakes filter.
5. Compliance automation. Automated 1099 and W-8BEN generation, audit-ready records, and built-in compliance for both U.S. and French regulatory regimes — without Hugging Face’s legal team manually filling forms.
Deel won on all five. The migration began.
How the Migration Actually Worked: Hugging Face's Payments Consolidation Approach
The mechanics of the migration wasn’t captured in the study, which is typical of vendor-authored case studies, and one of the reasons independent retellings like this one matter. From the documented outcomes and Deel’s general onboarding patterns, the consolidation appears to have followed a recognizable four-stage pattern any AI startup can replicate.
1. Worker classification audit
Before migrating any payments, Hugging Face would have needed to classify each of its 70 workers correctly: who is a contractor vs. an employee, which entity each one belongs to, what currency they’re paid in, and what tax forms apply to their jurisdiction. This is the step that gets skipped most often and where most consolidations leak compliance risk. With contractors spread across multiple countries and two home entities, getting this right is non-negotiable.
2. Contract standardization with policy preservation
This is where Anna Tordjmann’s quote becomes operationally important. Hugging Face had to migrate from heterogeneous legacy contracts (some generated in the legacy systems, some custom-drafted) into Deel’s contract templates without losing the strict legal and HR requirements that made the legacy contracts work for them. Deel’s customizable agreements and amendments made this possible; a less flexible platform would have forced Hugging Face to either weaken their legal standards or maintain a parallel contract layer outside the system.
3. Phased payment cutover
Migrating 70 workers off five payment systems doesn’t happen in one weekend. The realistic pattern is a phased cutover: pay the next monthly cycle through both old and new systems for a subset of workers, validate that net pay arrives correctly, then expand. Hugging Face’s documented outcome: successful onboarding of 70 workers implies the phasing worked.
4. Entity-aware payment routing
With Deel live, every payment now routes from the correct U.S. or French entity automatically. The intercompany transfer pricing implications, FX conversion, and tax form generation all happen inside one system instead of being reconciled across five.
The Results
The headline outcomes:
• 50% reduction in payment processing time. What used to take Hugging Face’s finance and legal teams a multi-day monthly exercise now takes half as long.
• Five systems centralized into one. The full contract-to-payment-to-tax-form lifecycle lives on Deel.
• 70 workers onboarded to the new platform without disruption to pay or contracts.
• Customizable agreements preserved. Hugging Face’s strict legal and HR requirements survived the migration intact, they didn’t have to lower their bar to gain operational simplicity.
The qualitative outcome matters too. As Anna Tordjmann summarized: “The Deel platform was a game-changer for us. It’s very easy to use and has streamlined our entire payment process.” For a legal team at a Series D AI company, “game-changer” is a stronger endorsement than the published metrics suggest on their own.
What Tools Does Hugging Face Use Alongside Deel?
Hugging Face’s operational stack is a snapshot of what a modern AI company actually runs on. The model-hosting infrastructure runs on Hugging Face’s own platform (naturally). Secrets management runs on Infisical for centralized credential sharing across engineering. Global contractor payments and contract management run on Deel. The collaboration stack is the standard remote-AI-company kit: GitHub for code, Slack for chat, Notion or Google Workspace for documents.
The pattern worth noticing: each tool owns one job. Deel owns global payments and contracts. Infisical owns secrets. Hugging Face’s own platform owns model collaboration. There’s no “one tool to rule them all”, there’s a deliberately composed stack where each system is best-in-class at its category.
That’s a useful contrast to the GitLab case study, where the goal was consolidation onto one vendor for one category. Both companies converged on the same principle: one vendor per category, no overlap, no fragmentation within categories.
What Other AI Companies Can Learn From Hugging Face's Payments Strategy
Hugging Face’s 70-worker payment problem is closer to the median AI startup’s reality than GitLab’s 1,500-employee EOR problem. Most AI companies will face exactly this scenario, a small but growing global workforce, multi-entity legal structure driven by where founders and capital sit, and a payments stack that quietly accumulates tools until month-end becomes a fire drill.
1. Multi-entity is the rule for AI startups, not the exception. If your founders are non-U.S. and your investors are U.S., you’ll end up with at least two home entities. Plan your payments architecture around that from the start, not after the fact.
2. Contractor-heavy workforces need contract flexibility more than payroll automation. The Hugging Face story isn’t really about payments — it’s about contracts. Anna Tordjmann’s quote specifically calls out customization as invaluable. AI companies hire specialized researchers with bespoke arrangements (advisor equity, IP carve-outs, NDA depth). A payments platform that can’t accommodate custom contract amendments will get rejected by your legal team within weeks.
3. Don’t wait until 70 workers to consolidate. At 10–15 contractors, multiple payment tools feel like flexibility. At 70, they feel like a tax. The earlier you consolidate, the cheaper the migration. Hugging Face’s 50% time reduction is a benefit they could have captured years earlier.
4. Currency and country coverage matter more than UI. A platform with 150+ country and 200+ currency coverage gives you optionality on every future hire. A platform optimized only for “U.S. and Europe” will become a constraint the first time you recruit a researcher in Singapore, Brazil, or Nigeria.
5. Centralization pays back in audit-readiness, not just time savings. The “50% reduction in payment processing time” is the headline, but the durable benefit is having one auditable system instead of five. Series E and beyond, that matters more than the time savings — it’s the difference between a clean financial close and a deferred audit opinion.
Conclusions
The Hugging Face story is the AI-startup version of a problem every globally distributed company faces eventually: at some point, the tools you adopted reactively need to be replaced by tools you select deliberately. The trigger is usually pain, a finance team that can’t close the month, a legal team that can’t validate compliance across five systems, a CEO who realizes the next 70 hires will break what the first 70 barely survived.
For Hugging Face, the answer was Deel, a single platform that respected their multi-entity structure, preserved their strict legal standards through customizable contracts, and collapsed five tools into one. The 50% reduction in processing time is the headline number. The harder-to-measure win is what Anna Tordjmann named directly: a payment process that’s now easy. For a legal team running point on global contractor compliance at a $4.5B AI company, “easy” is the highest praise a vendor can earn.
If you’re an AI company at a similar stage, multi-entity, contractor-heavy, growing fast: the takeaway is simple: pick the payments and global hiring platform deliberately and early. Our guide to the best Employer of Record services in Asia compares Deel against Remote, Rippling, Oyster, and Multiplier on exactly the criteria Hugging Face used: multi-entity support, contract flexibility, currency coverage, and compliance depth. And if you’ve narrowed to two finalists, the Deel vs Multiplier comparison breaks down the trade-offs in detail.
Frequently Asked Questions
Hugging Face is an AI infrastructure company, not a payroll company. Building proprietary multi-entity global payments tooling, with French URSSAF compliance, U.S. 1099 generation, and 200+ currency support — would have consumed engineering resources better spent on their core product. Outsourcing to Deel let them get the outcome without the build cost.
Yes — arguably more easily than from larger case studies. Hugging Face’s pre-Deel state (five payment tools, 70 workers, multi-entity structure) is exactly where most Series A–B AI startups end up by accident. Consolidating earlier, at 15–30 workers, is dramatically cheaper than consolidating at 70+.
Hugging Face used five different tools to pay their workforce before consolidating onto Deel.
