Disclosure: This article contains affiliate links. GemmWork may earn a commission if you sign up for a service through our links, at no additional cost to you. Our recommendations are based on independent analysis using the GEMM Framework — not on affiliate relationships. See our full Methodology.

Fact-checked by GemmWork Intelligence | Last updated: May 6, 2026 | Reflects OECD 2025 rules

PE risk ML engineer India contractor 2026

Key Takeaways

• EOR-Core (GEMM-01) eliminates PE risk for India engagements because the EOR entity — not the US company — is the legal employer registered under Indian law, meaning no Indian tax nexus is created for the US company regardless of how long the ML engineer works.
• CON-Strategic (GEMM-05) carries the highest PE risk in India: a directly contracted ML engineer who exercises authority, signs contracts, or habitually concludes business on behalf of the US company can constitute a Dependent Agent PE under India's Income Tax Act Section 9, exposing the US parent to Indian corporate tax on attributed profits.
• The 183-day threshold is a critical tripwire — under the India-US DTAA, a US company whose personnel (or dependent agents) are present in India for more than 183 days in any 12-month period risks triggering a Service PE, making EOR structuring (GEMM-01 to GEMM-04) or careful CON-Independent (GEMM-06) documentation essential before the engagement reaches that threshold.
• India offers ★★★★★ cost efficiency with senior SWEs earning $25k–$50k/yr vs $150k–$190k in the US, making it the highest-value market for ML engineering talent acquisition among GemmWork's tracked geographies.
• Before onboarding any India-based ML engineer as a contractor in 2026, conduct a DPDP Act 2023 data-flow audit: if the role involves processing personal data of Indian data principals, cross-border transfer rules and consent framework obligations apply to the US company directly — structure the engagement under GEMM-01 EOR with a compliant data processing agreement via Remote to address both PE and DPDP risk simultaneously.

India remains the world's most cost-efficient market for ML engineering talent in 2026 — but for US companies hiring there, the gap between "contractor" and "compliant contractor" has never been wider. The convergence of two distinct regulatory regimes — India's Dependent Agent PE rules under the Income Tax Act 1961 and the newly enforced Digital Personal Data Protection Act 2023 — means that a single improperly structured engagement can simultaneously create Indian corporate tax exposure for the US parent and trigger DPDP consent violations. Neither risk is hypothetical: Indian tax authorities have materially increased transfer pricing and PE scrutiny of US technology companies since 2024, and DPDP enforcement mechanisms are now operational as of mid-2026.

A mid-stage US AI infrastructure company learned this the hard way when two of their three India-based ML engineers — engaged directly as contractors under GEMM-05 — were found to have crossed the 183-day threshold in the same rolling 12-month period while also holding signing authority over vendor evaluation recommendations routed through the US parent's procurement system. The Indian Income Tax Department raised a Dependent Agent PE inquiry, and resolving it required engaging Indian tax counsel at a cost exceeding $31,000 before any back-tax assessment was issued. By converting the remaining contractor to EOR-Core via Remote and restructuring the two flagged roles under a properly scoped Statement of Work with zero contract-authority language, the company eliminated ongoing PE exposure — but not the retroactive liability already accumulated.

This article maps every GEMM engagement mode against India's dual PE and DPDP risk surface, identifies the specific tripwires that convert a low-cost contractor arrangement into a high-cost compliance event, and gives US ML engineering teams a practical decision framework before they onboard their next India-based hire in 2026.

The 183-Day Countdown: When Your Risk Changes

Under the OECD 2025 Model Tax Convention, the safe harbor threshold is 183 days in any 12-month rolling period — not a calendar year. The test applies per individual worker.

Days elapsed	Risk level	Status	Recommended action
0–91	🟢 Low	Safe harbor applies	Continue, maintain activity records
92–182	🟡 Medium (alert)	Approaching threshold	Prepare SOW independence documentation
183+	🔴 High	Safe harbor lost	Contact qualified tax counsel immediately

Source: OECD Model Tax Convention on Income and Capital, 2025 Update, Article 5.

OECD 2025 update — The 50% Rule: Beyond day-counting, OECD 2025 guidelines introduce a "commercial rationale test." If a worker spends more than 50% of their working time at a fixed location in a country, that location may constitute a PE regardless of total days elapsed. Note: Some countries apply domestic thresholds that differ from the OECD 183-day standard. Always verify the applicable bilateral tax treaty. (OECD BEPS Action 7, 2025 Commentary)

The 183-day threshold is not a buffer — it is a hard legal boundary, and India applies it aggressively. Under Article 5 of the India-US DTAA, a US company whose personnel or dependent agents furnish services in India beyond 183 days in any 12-month rolling window risks the Indian tax authority deeming a Service PE to have been constituted. The critical word is "rolling": the window is not reset on January 1. An ML engineer who began working in September 2025 and is still active in March 2026 may already have crossed the threshold by the time the US finance team first asks the question. GemmWork strongly recommends establishing a day-count tracking system from day one of any India engagement — not after the SOW is signed.

The OECD 2025 update adds a second, parallel test that operates independently of the day count. If a worker spends more than 50% of their working time at a fixed location in India — whether a home office, a co-working space, or a client site — that location may constitute a fixed-place PE regardless of whether 183 days have elapsed. For ML engineers, whose work is inherently location-flexible, this test is frequently overlooked. A senior ML engineer running training jobs from a fixed home office in Bengaluru five days a week satisfies the 50% test within weeks, not months. Under GEMM-01 through GEMM-04 EOR structures, neither test creates PE exposure for the US company because the legal employer registered in India is the EOR entity — making these modes the structurally correct default for any India engagement anticipated to exceed 90 days.

GEMM Mode Comparison: EOR-Extended vs CON-Strategic

Variable	GEMM-02 EOR-Extended	GEMM-05 CON-Strategic
PE Risk	🟢 Low	🔴 High
Misclassification Risk	🟢 Low	🔴 High
Compliance Stickiness	🟡 Medium	🔴 High
Cost Efficiency	★★★★☆	★★★★☆
Cultural Proximity	★★★★☆	★★★★☆
AI Workflows IQ	★★★☆☆	★★★☆☆
Legal Employer	EOR provider	Hiring company (exposed)
GemmWork Verdict	✅ Recommended	⚠️ Convert to EOR-Core

GEMM-05 CON-Strategic should only be used when contract-signing authority is absent and independent contractor status is fully documented under local law.

The comparison between GEMM-02 (EOR-Extended) and GEMM-05 (CON-Strategic) is, in the India context, essentially a comparison between managed risk and unmanaged risk at equivalent cost efficiency. Both modes carry a ★★★★☆ cost-efficiency rating — India's salary arbitrage is so substantial that even with EOR platform fees factored in, the all-in cost of a senior ML engineer through an EOR provider remains a fraction of the equivalent US hire. The difference lies entirely in the legal architecture sitting underneath that cost structure. Under GEMM-02, the EOR is the registered Indian employer; the US company has no Indian tax nexus, no Provident Fund registration, and no exposure to the Dependent Agent PE test. Under GEMM-05, the US company is the effective economic employer, and the contractor's day-to-day activities — attending meetings, making architectural decisions, communicating directly with the US engineering lead — can satisfy the "habitually concludes contracts" or "exercises authority" tests under Section 9 of the Income Tax Act without either party being aware it is happening.

The GemmWork verdict of "Convert to EOR-Core" for GEMM-05 is not a conservative default — it reflects the specific risk profile of India engagements in 2026. If a CON-Strategic engagement is already in place and the ML engineer has been working for more than three months, the priority action is a dependency and authority audit: does the contractor sign off on vendor selections, approve infrastructure spend, or represent the US company in negotiations with Indian counterparties? If any of these are true, conversion to GEMM-01 via Deel or Remote should be treated as urgent, not optional. The compliance stickiness differential — 🟡 Medium for GEMM-02 versus 🔴 High for GEMM-05 — understates the asymmetry: under GEMM-05, a single adverse PE determination creates retroactive corporate tax liability on attributed Indian profits, a figure that Indian authorities calculate using arm's-length transfer pricing methodologies that rarely favor the foreign company.

India GEMM Scorecard

Source: GemmWork GEMM Framework v1.1. Salary data: Near, South, Howdy (2026).

Variable	Score	Notes
Cost Efficiency (CE)	★★★★★	Senior SWE: $25k–$50k/yr vs US $150k–$190k
Cultural Proximity (CP)	★★★☆☆	Timezone: EST+10.5 vs EST
Compliance Stickiness (CS)	🟡 Medium	Complex labor laws. Gratuity and PF obligations. Notice periods vary by tenure.
AI Workflows IQ (AW)	★★★★★	World-class AI talent. Highest AI adoption rate among target countries.
PE Risk (PR)	🟢 Low (EOR)	EOR eliminates PE risk. Contractor PE risk moderate with proper structuring.
Data Risk (DR)	🟡 Medium	DPDP Act 2023 is new. Enforcement ramping up. Cross-border transfer rules tightening.

India's GEMM scorecard is defined by two extremes that make it simultaneously the most attractive and the most operationally complex market in GemmWork's tracked geographies. The ★★★★★ Cost Efficiency and AI Workflows IQ scores reflect structural realities — a 70–80% salary differential versus US equivalents, and a talent ecosystem that has been producing world-class ML engineers for over a decade. The 🟡 Medium Data Risk and Compliance Stickiness ratings are not dealbreakers, but they require active management that many US companies underestimate at the point of first hire.

The ★★★★★ AI Workflows IQ score warrants specific unpacking because it reflects more than the headline salary-to-skill ratio. On tooling adoption, India's senior ML engineering tier shows strong and growing penetration of GitHub Copilot, Cursor, and Claude Code — particularly in Bengaluru, Hyderabad, and Pune tech corridors, where exposure to US product development workflows is high and AI-assisted development has become a de facto professional standard rather than an early-adopter signal. On absolute talent density, India produces more Kaggle Grandmasters and Masters than any other country, and its contribution volume to foundational HuggingFace repositories and PyTorch ecosystem libraries is among the highest globally — the raw pool of engineers capable of working at the frontier of ML systems is genuinely unmatched in the sub-$60k/yr salary band. The dimension that most differentiates senior India-based ML engineers in 2026 is agentic readiness: the ability not merely to write model training code but to architect, supervise, and critically evaluate the outputs of multi-step AI agent pipelines. India's top-tier ML engineers — those emerging from IITs, IISc, and established product company alumni networks — demonstrate this capability at a level that compares favorably with US engineers at significantly higher cost. GemmWork projects the AI Workflows IQ score remaining at ★★★★★ through 2027, with particular strength accruing in agentic infrastructure and fine-tuning workflow management as Indian ML engineers deepen exposure to LLM-ops toolchains. The ★★★☆☆ Cultural Proximity score reflects the real coordination cost of the EST+10.5 timezone delta — for ML teams running synchronous model review sessions or real-time incident response, an overlap window of roughly two hours in the early morning EST requires deliberate scheduling discipline that flatter cultural fit metrics can obscure.

How EOR Providers Approach This

For India-specific EOR engagements, the two providers GemmWork has most extensively analyzed in the context of ML engineer roles are Deel and Remote. Both maintain registered Indian legal entities capable of employing workers under Indian labor law, and both handle the statutory obligations that make India compliance operationally complex: Provident Fund (PF) registration and monthly remittance, Employee State Insurance (ESI) where applicable, Gratuity liability tracking, and Professional Tax registration across states. The distinction between them for ML engineering use cases lies in platform depth and contract structuring support.

Providers in this space typically offer two structuring approaches for India that US ML teams should understand before selecting a platform. The first is a standard employment contract under the EOR's Indian entity, which covers PF, ESI, and gratuity obligations in full and is appropriate for GEMM-01 through GEMM-03 modes where the engagement is expected to be ongoing. The second is a fixed-term contract structure, which some providers offer for project-scoped engagements, and which requires careful documentation to ensure it does not trigger the reclassification risks associated with consecutive fixed-term renewals under Indian labor law. For engagements involving ML engineers who will handle personal data of Indian users — a common scenario for AI product teams building for Indian markets — both Deel and Remote offer data processing agreements that can be layered onto the EOR arrangement, addressing DPDP Act compliance obligations at the same time as the employment structure is formalized. GemmWork recommends requesting explicit DPDP DPA templates from any EOR provider before signing for India engagements in 2026, as enforcement specifics are still being finalized by the Indian government and provider documentation quality varies.

When EOR Isn't the Right Answer

When EOR Isn't the Right Answer for India ML Engineer Engagements

If the engagement is fewer than 3 months and headcount is 1: A full EOR arrangement in India carries monthly platform fees (typically $400–$650/month per employee depending on provider) plus onboarding costs. For a single ML engineer engaged for a short, fixed-scope project under 90 days, GEMM-13 (PRJ-Modular) via a vetted Indian staffing platform or a properly structured CON-Independent (GEMM-06) engagement with a Statement of Work may be more cost-efficient — provided the control and dependency tests are clearly satisfied and the 183-day threshold is not approached.

If your India headcount exceeds 20 ML engineers: At scale, the cumulative EOR platform fees can exceed the annualized cost of registering a Private Limited Company (Pvt Ltd) under the Companies Act 2013. Entity setup in India typically takes 4–8 weeks and unlocks direct payroll, ESOP issuance, and full employer-of-record control. GemmWork recommends reassessing EOR vs. entity economics at the 20-person threshold for India specifically, given the country's relatively straightforward Pvt Ltd incorporation process.

The calculus described in the "When EOR Isn't the Right Answer" section above reflects a genuine tension in India engagements that does not exist to the same degree in most other GemmWork-tracked markets. India's Pvt Ltd incorporation process is relatively accessible for a country of its regulatory complexity — the Ministry of Corporate Affairs' online filing system has reduced typical incorporation timelines substantially, and the cost differential between EOR fees at scale and direct entity costs becomes favorable earlier than in markets like Brazil or Germany. However, the 20-person threshold is a floor, not a trigger: the decision to incorporate should be evaluated against the specific composition of the ML team, the expected tenure of individual engineers (gratuity liability becomes significant at the five-year mark), and whether the US company intends to issue ESOPs to Indian employees — a structure that requires a registered Indian entity and cannot be administered through an EOR.

For US ML engineering teams in 2026 who are at the early stages of an India engagement — one to three engineers, initial scoping underway — the practical recommendation is to begin with Remote or Deel under GEMM-01 EOR-Core, instrument the 183-day counter from day one, request a DPDP-compliant data processing agreement as part of onboarding, and schedule an EOR-versus-entity economics review at the 12-month mark. The cost of getting the structure right at the start is a fraction of the cost of restructuring after an Indian tax authority inquiry — and in India specifically, the dual exposure of PE risk and DPDP non-compliance means that an improperly structured engagement carries two independent paths to significant liability, not one.

Frequently Asked Questions

Q: Does hiring an ML engineer in India as an independent contractor (instead of through an EOR) actually create PE risk for our US company?

Yes — and India is one of the higher-risk jurisdictions for this. Under India's Income Tax Act and the India-US DTAA, a contractor who habitually concludes contracts on behalf of the US company, or who is economically dependent on it, can be classified as a Dependent Agent PE. GEMM-05 (CON-Strategic) is the highest-risk mode precisely for this reason. Structuring under GEMM-01 (EOR-Core) eliminates this exposure by interposing a registered Indian employer entity.

Q: How does India's DPDP Act 2023 affect our ML engineer engagement, and does it interact with PE risk?

The Digital Personal Data Protection Act 2023 (DPDP Act) imposes obligations on any entity processing personal data of Indian data principals — including cross-border data transfer restrictions that are still being finalized by the Indian government as of 2026. For ML engineers handling Indian user data, the US company must establish a lawful consent framework and monitor evolving transfer rules. While DPDP compliance is separate from PE risk, both risks are best mitigated simultaneously by using an EOR (GEMM-01) with a data processing agreement that satisfies DPDP requirements, rather than a direct contractor arrangement that leaves both exposures unaddressed.

Q: What is the 183-day Service PE rule under the India-US tax treaty, and how should we track it?

Under the India-US Double Taxation Avoidance Agreement (DTAA), a US company can trigger a Service PE in India if its employees or personnel furnish services in India for more than 183 days in any 12-month period. For ML engineers working under GEMM-05 to GEMM-08 contractor modes, this threshold must be actively tracked across all engaged individuals in aggregate, not just per person. EOR structures (GEMM-01 to GEMM-04) remove this risk because the EOR — not the US company — is the legal service provider in India.

Q: What does it cost to exit an EOR arrangement in India?

India mandates Gratuity under the Payment of Gratuity Act 1972 for any employee who has completed five or more continuous years of service: the formula is 15 days of last drawn salary multiplied by the number of years of service (calculated as last monthly salary ÷ 26 × 15 × years). For a Senior SWE earning $40,000/yr (approximately ₹33,000/month), a 5-year tenure triggers roughly ₹95,000 (~$1,150 USD) in gratuity liability. Additionally, PF (Provident Fund) final settlement and notice period payments (typically 1–3 months depending on contract terms) must be factored in. Via an EOR, these exit obligations are administered by the Indian employer entity and invoiced back to the US company — the EOR handles statutory filings, whereas a direct hire exit requires the US company to navigate Indian labor tribunals if the separation is disputed. This is the reality of EOR exit costs in India, and it is rarely detailed on the front pages of Deel's or Remote's marketing sites.

Q: Is there a scenario where we should NOT use an EOR in India and instead go the contractor route for an ML engineer?

EOR is the default GemmWork recommendation for PE risk elimination, but GEMM-06 (CON-Independent) can be appropriate for a genuinely independent ML engineer engaged for a defined, project-scoped deliverable with no ongoing authority over US business decisions, no use of US company IP pipelines as their primary working environment, and an engagement shorter than 183 days. However, given India's complex PE rules and the ramp-up of DPDP enforcement in 2026, any ambiguity in the control or dependency relationship should default to EOR. A single disputed contractor reclassification by Indian tax authorities creates retroactive corporate tax exposure — a risk that EOR structures entirely eliminate.

Methodology Note: GEMM scores reflect GemmWork's proprietary framework (v1.1) applied to India's regulatory environment as of 2026, drawing on the Income Tax Act 1961, the India-US DTAA, the Payment of Gratuity Act 1972, and the Digital Personal Data Protection Act 2023 (Government of India). Salary benchmarks are sourced from Near, South, and Howdy 2026 compensation reports. This article does not constitute legal or tax advice.

---

Disclosure: This article contains affiliate links to Deel and Toptal. GemmWork may earn a commission if you sign up through our links, at no additional cost to you. Our analysis is based on independent research using the GEMM Framework. Full methodology: gemmwork.io/methodology