The +28 percent figure at the centre of this index is not a cost-of-living differential. It is a premium — a systematic, role-specific compensation uplift that employers pay to attract AI talent willing to come in four to five days per week, documented across 1,840 verified AI practitioners in ENTRA's Q3 2026 Salary Survey. At the top of the index, a Foundation Model Researcher sitting in a frontier lab in San Francisco earns a median total compensation of $978K annually. The remote equivalent — same seniority, same role, different location — earns $760K. The gap is $218K, and it is widening. This is the role-level anatomy of the RTO Wars: not a policy story, but a compensation map.
That $218K delta is not an accident of negotiation. It is a structural feature of how frontier labs have responded to the return-to-office moment. When companies reinforced SF-preferred postures in early 2026 and extended attendance requirements through Q3, the market signal to competing labs was immediate: the engineers you want most will go to the lab that pays them most to accept the commute. RSU grant structures are the mechanism. ENTRA Salary Survey Q3 2026, cross-referenced against Levels.fyi verified offer data for 2,200-plus AI practitioners, shows that on-site Foundation Model Researchers receive equity grants averaging $580–640K annually versus $420–460K for remote-equivalent researchers at the same labs. The RTO premium is not being buried in perks. It is being written directly into offer letters.
Three findings from the data
Finding 1: The frontier lab ceiling effect. The top three roles — Foundation Model Researcher (92), ML Research Scientist (88), and AI Safety Researcher (84) — are all concentrated at a small number of SF and London employers. The premium is structural at those labs because the labs themselves have decided it is: in-office requirements are maintained not primarily for collaboration benefits but as a signal about research culture and team cohesion. Engineers who accept those requirements are granted larger equity packages as direct acknowledgment that they are foregoing remote market alternatives. The market is pricing that concession explicitly. GPU Kernel Engineers (83) and Senior ML Engineers at Big Tech (82) complete the AAA/AA tier on similar logic — hardware-proximate roles at employers with the leverage and the willingness to pay for physical presence.
Finding 2: The discretionality gradient. The roles scoring highest are not necessarily the roles with the least remote execution capability — they are the roles where the employer has the most leverage to require on-site presence and has chosen to exercise it. AI Infrastructure Engineers (79) score high partly because their work benefits from hardware proximity, but also because employers know these engineers will accept location requirements in exchange for premium equity access to frontier-scale compute. Prompt Engineers (56) and AI Quality Analysts (46), by contrast, have fully remote-executable work and employers who lack the leverage to require otherwise — no premium emerges because no mandate is being enforced. The index reveals that RTO premiums are not distributed by technical necessity alone; they are distributed by employer power in the labour market.
Finding 3: The Transparency Paradox. September 2026 marks the first month in which a majority of US states have active pay transparency requirements on the books. Yet ENTRA Job Signal Index Q3 2026 analysis of 52,000-plus AI job postings finds that fewer than 31 percent of in-office-required postings include an explicit location premium disclosure — meaning that the RTO compensation gap documented in this index is systematically absent from the public record of offers being made. Candidates negotiate it. CHROs know it. But the market as a whole cannot price it efficiently because the premium is not published. This is the mechanism by which companies enforcing RTO mandates maintain information asymmetry: they price the premium privately into the offer while advertising only the base. Pay transparency legislation has not yet closed this gap. The data says it needs to.
The mid-table mechanics
AI Product Managers (76) and Computer Vision Engineers (74) sit in the A tier on the back of different premium mechanics. For PMs, the in-office premium reflects proximity to research and roadmap cycles at companies with strong in-person engineering cultures — the equity grant differential, at a median +$88K, is real but below the research tier. For CV Engineers, the dispersion is the story: robotics-adjacent CV roles generate premiums above $180K at p90, while enterprise API deployment engineers see barely $40K. Aggregated, the blended A-tier score obscures what is, in practice, a bimodal role distribution — with the physical AI segment generating some of the highest premiums in the index, and the enterprise API segment generating almost none.
The BBB tier — AI/ML Solutions Architects (68) through Data Scientists with ML focus (61) — represents the region of the market where RTO premiums are real but employer leverage is genuinely limited. Solutions Architects' client-side co-location requirements create premiums more attributable to client contract terms than employer mandate. MLOps Engineers (63) score here because the in-office requirement, where it exists, is a cultural artefact of engineering team norms rather than a functional necessity — modern MLOps tooling has simply made the location dependency obsolete.
The remote floor
At the B tier, the index becomes an argument about leverage more than compensation design. AI Trainers and RLHF Specialists (48), Junior ML Engineers (47), AI Quality Analysts (46), and AI Talent Acquisition Specialists (43) are all roles where employers either cannot enforce meaningful on-site requirements or have not found it necessary because the cost of doing so exceeds the benefit. The RLHF Specialist result is the most strategically significant: this is a role that frontier labs built entirely on distributed infrastructure. The median on-site differential of $8K is cost-of-living adjustment, not premium — and the 13/30 Office Attendance Premium score is the honest arithmetic of that reality.
The floor of the index — AI Talent Acquisition Specialists at 43 — carries a structural irony worth naming. The professionals who negotiate RTO premiums on behalf of employers with AI candidates receive no material premium themselves. Their work is remote-native, their leverage is low, and the organisations they recruit for have not found it necessary to incentivise their physical presence. ENTRA's data cannot resolve that irony. It can only document it.
How we ranked
The Top 20 AI Roles by RTO Compensation Premium Index — September 2026 is scored across 5 dimensions:
- Office Attendance Premium (30 pts) — Percentage salary uplift for in-office versus fully remote workers at equivalent seniority. (Source: ENTRA Salary Survey Q3 2026, n=1,840 AI practitioners; Levels.fyi Q3 2026 in-office vs. remote offer split, n=2,200+ verified offers; Glassdoor September 2026 remote vs. hybrid pay differential data)
- TC Lift for In-Office (25 pts) — Total compensation gap including equity, where RSU grant differentials are the dominant driver at frontier lab and Big Tech roles. (Source: Levels.fyi Q3 2026 offer corpus with location tag, n=2,200+; ENTRA Salary Survey Q3 2026 equity module; Anthropic, OpenAI, and Google DeepMind public band disclosures Q2–Q3 2026)
- Remote Pay Parity Gap (20 pts) — Inverse measure: higher scores indicate remote workers are more penalised relative to in-office equivalents. (Source: ENTRA Salary Survey Q3 2026 pay parity module; Glassdoor September 2026 differential data; LinkedIn Talent Insights Q3 2026 posting salary band analysis)
- In-Office Demand Heat (15 pts) — Growth rate of in-office-required postings for each role between Q2 and Q3 2026. (Source: ENTRA Job Signal Index Q3 2026, 52,000+ AI postings; LinkedIn Talent Insights Q3 2026 in-office filter)
- Role Discretionality Score (10 pts) — Inverse of remote execution capability: low discretionality means employers can mandate in-office more credibly, hardening the premium. (Source: ENTRA Q3 2026 practitioner survey self-reported remote productivity equivalence, n=1,840; ENTRA Job Signal Index Q3 2026 location-requirement language analysis)
Data window: ENTRA Salary Survey Q3 2026 (July 1 – September 5, 2026, n=1,840); ENTRA Job Signal Index Q3 2026 (April 1 – September 1, 2026, 52,000+ postings); Levels.fyi Q3 2026 verified offers with location tag (n=2,200+); Glassdoor September 2026 pay differential data; LinkedIn Talent Insights Q3 2026
Sample size: 1,840 verified AI practitioner responses (ENTRA Salary Survey Q3 2026); 52,000+ AI job postings (ENTRA Job Signal Index Q3 2026); 2,200+ Levels.fyi verified offers with location tag; Glassdoor September 2026 role-filtered pay differential dataset
Limitations:
- Private company equity is estimated from public funding rounds, comparable benchmarks, and Levels.fyi disclosed offer data; actual vested RSU value for frontier lab roles may differ materially from tabulated TC gap figures, particularly where post-financing valuation has moved significantly since the grant date
- The Remote Pay Parity Gap dimension relies on self-reported location and compensation data; respondents at companies with explicit remote pay adjustment policies are likely over-represented relative to the broader AI workforce, which may inflate gap scores for roles concentrated at policy-transparent employers
Inquiries about methodology: methodology@entracareers.com
What to watch in Q1 2027
The December 2026 round of annual compensation reviews at frontier labs will be the first to incorporate a full twelve months of enforced RTO premium data into RSU refresh structures. ENTRA will watch for whether the on-site equity differential widens further — a sign that employers are leaning into the premium mechanism — or begins to compress, indicating that the talent market has adjusted expectations and the premium is being absorbed into base salary rather than disclosed as a location uplift. The Transparency Paradox is likely to partially resolve by mid-2027 under state pay transparency enforcement pressure: once the location premium must be published, the information asymmetry that currently favours employers in RTO offer negotiations will shift toward candidates. When that happens, the roles at the top of this index will face the most recalibration — because the premium they carry has, until now, been the most opaque.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.