The receipts are now large enough to audit.
Over the past three weeks, ENTRA Intelligence has tracked the September 2026 RTO enforcement quarter across four markets, fourteen employer cohorts, and four analytical instruments. The qualitative findings — a 5× attrition multiplier, a 67% upstream candidate filter rate, a structural talent bifurcation between flexible and mandate-enforcing AI employers — have already entered the public record. What remained, entering the final week of the month, was the financial model: not the multiplier, but the dollar figure it produces. Not the rate of attrition, but the cost per event.
That cost is $2.3 million — the estimated replacement and counter-offer burden the average AI employer enforcing a five-day mandate absorbs for every 10 senior ML engineers who resign in protest, as modeled across ENTRA's four-vector talent tax framework. It is not a theoretical number. It is the product of four compounding data series — attrition signal rates, backfill cost modeling, counter-offer compensation premiums, and hiring velocity degradation — each measured independently and combined into a single per-cohort figure.
This is the closing report of ENTRA's September RTO Wars coverage. It is also, by design, the only piece in the series that arrives with the full financial model assembled. The opening report (September 11) framed the enforcement quarter. This one closes it with an accounting.
The price of presence
The $2.3M figure requires four components, each independently sourced and each representing a distinct channel through which the talent tax accumulates.
Vector one: The attrition rate differential. ENTRA's Q2-Q3 2026 Recruiter Survey — n=134 technical recruiters and talent acquisition leaders; n=1,081 candidate sub-cohort across AI/ML roles — found a 23-point voluntary attrition signal gap between five-day mandate employers and remote-first peers. Five-day mandate employers registered 31% voluntary attrition signal in the senior IC cohort (L5 equivalent and above). Remote-first employers registered 8%. That differential — which holds across company size, market cap tier, and regional cluster — is the foundation from which every other vector in this model flows. A five-day mandate employer operating a 100-person senior AI team should expect to absorb approximately 23 additional senior departures annually versus a remote-first employer running an equivalent team.
The 31% figure is not uniform within the mandate cluster. ENTRA's n=1,081 sub-cohort reveals a clear seniority gradient: the departure signal runs highest at L6 and L7 equivalents (41% exit intent, consistent with earlier ENTRA analyses), lower at L5 (24%), and lower still at L4 (19%). The talent tax is regressive in the worst possible direction for the employer paying it. The engineers most expensive to replace are leaving at the highest rate.
Vector two: The backfill cost. ENTRA's September 2 analysis established the replacement cost ceiling for a departing L7 Principal ML Engineer at $400,000 — a composite of external recruiter fees ($80,000 to $130,000 at 20 to 25% of first-year total comp), internal hiring team investment ($12,000 to $24,000 for 80-plus hours at fully-loaded cost), productivity ramp deficit ($80,000 to $180,000 across a 3-to-6-month ramp to full output), and project disruption costs ($50,000 to $120,000 per senior departure). The L6 cohort within the same band averages $200,000 to $260,000 per replacement. The present model applies the seniority-adjusted cost structure to the observed departure distribution.
Because the 31% attrition signal skews toward the L6/L7 tier, the blended replacement cost per departure in the mandate-driven attrition pool runs higher than an all-seniority average would imply. ENTRA's weighting of the observed departure distribution — approximately 55% of departures at L6/L7, 30% at L5, 15% at L4 — produces a blended replacement cost of $185,000 per departure across the senior IC cohort. Ten departures: $1.85M in backfill alone, before counter-offer costs are accounted for.
Vector three: The counter-offer escalation premium. ENTRA's Q3 2026 Salary Survey — n=2,847 AI practitioners across 47 countries — provides the first cross-market quantification of the retention premium that five-day mandate employers must pay to hold senior ICs who have received competing offers from flexible-policy employers. The finding: five-day mandate employers are extending counter-offers at an average of 15 to 25% total compensation lift to retain engineers who have reached the offer-acceptance stage with a hybrid or remote-first competitor. At hybrid employers, the equivalent retention premium runs 5 to 8%.
The gap — 15 to 25% versus 5 to 8% — is the mandate's market price. An engineer who can move to a hybrid employer and receive essentially equivalent technical opportunity is extracting a 7-to-17-point premium to stay at a five-day shop. At an L6 base salary of $340,000 to $380,000, that premium runs $51,000 to $95,000 in additional annual total compensation. Across a 100-person senior AI team with a 31% attrition signal, the counter-offer spend to retain even a fraction of at-risk engineers compounds quickly against the backfill cost of the departures the counter-offers fail to prevent.
The $2.3M model. For every 10 senior ML engineers who reach the resignation stage at a five-day mandate employer, ENTRA estimates the following distribution of outcomes: approximately 6 of the 10 received counter-offers before resignation was finalized, at an average 20% TC lift — those counter-offers were paid and the employees left anyway (sunk cost of $408,000 at a $340,000 average base for the six, annualized to the average tenure extension of 4.2 months before departure); the 10 backfill searches at blended senior-IC cost ($185,000 per position, weighted to actual role distribution of the resignation pool) add $1.85M; the counter-offer sunk cost adds $408,000; and the total lands at approximately $2.3 million. This is a conservative model: it excludes knowledge reconstruction costs — the hardest line item to quantify and, at the Principal level, frequently the largest.
Vector four: The velocity penalty. ENTRA's Job Signal Index Q3 2026 — tracking 47,000+ AI role postings across all mandate tiers — finds a median time-to-close of 51 days for five-day mandate roles versus 28 days for hybrid roles in the same function and seniority band. The 23-day gap compounds the backfill cost model: each additional day a senior AI role sits vacant represents both an ongoing productivity deficit and a recruiting market signal that filters qualified candidates upstream. At a role level, the velocity penalty translates to an additional $46,000 to $92,000 in opportunity cost per unfilled position (23 days × $2,000-$4,000 estimated daily productivity value for an L6 ML engineer). Across a 10-position backfill cohort, the velocity penalty adds $460,000 to $920,000 to the model — a range ENTRA excludes from the base $2.3M figure to maintain conservatism, but which represents the realistic upper-bound adjustment.
The aggregate model, then, is not simply a departure-and-backfill calculation. It is a four-vector compounding structure in which the attrition rate determines how frequently the cost event occurs, the backfill cost determines the event's floor value, the counter-offer premium represents the sunk investment in failed retention, and the velocity penalty extends the cost duration of every open position. Mandate-enforcing AI employers are paying all four simultaneously.
The geography of RTO divergence
The financial model described above is a global average. Its components vary materially by region — shaped by different legal frameworks, compensation architectures, and cultural baselines for what office attendance means. To deploy the $2.3M model correctly requires understanding how each geography adjusts the inputs.
United States: the mandate cluster and its discontents. The hardest mandates in the global AI labor market are here, and the sharpest financial exposure follows. Amazon's five-day requirement — badge-enforced, incorporated into Q3 2026 performance reviews — covers approximately 8,000 AI and ML engineers across AWS Bedrock, Trainium silicon development, the AGI lab organization, and enterprise AI tooling. Its ENTRA RTO Talent Risk score is 33, B tier. xAI operates a five-day Austin mandate with no hybrid accommodation for ICs, scoring 28, B tier — the lowest in the ENTRA cohort.
Against this cluster, Levels.fyi Q3 2026 verified offer data (n=4,100+) shows the compensation bifurcation that the flexibility differential has produced: Anthropic's L6 research track now runs $480,000 to $740,000 in total compensation, versus a Google equivalent L5/L6 research scientist at $390,000 to $580,000 and an Amazon Principal Applied Scientist (L6) at $360,000 to $540,000. The Anthropic-to-Amazon gap at the senior research tier has widened from approximately 18% in Q4 2025 to 28% in Q3 2026, with the gap driven not by Anthropic raising its floor but by Amazon's inability to compete for the 67% of the qualified candidate pool that has already filtered out five-day mandate employers.
Microsoft is the Big Tech outlier. Its 2-to-3-day hybrid model, combined with AI org headcount growth from approximately 4,200 to 6,800 employees in six months, represents the most direct empirical counterargument to the mandate productivity thesis available in the US market. Microsoft did not enforce a five-day mandate. Microsoft grew its AI organization 62% in six months. The correlation is visible; the causal mechanism — a larger addressable talent pool, lower counter-offer expenditure, faster time-to-close — is consistent with the four-vector model above.
Hugging Face operates from the opposite end of the policy spectrum and is winning the talent arithmetic accordingly. Fully distributed across 47 countries, no attendance requirement, asynchronous-first operating model: ENTRA RTO Talent Risk score 91, AAA. The direct competitive consequence for mandate employers is a 3× qualified-applicant ratio advantage for Hugging Face on equivalent senior IC roles, per ENTRA Job Signal Index Q3 2026 analysis.
Middle East: the tax-free absorption model. The UAE and Saudi operating environments present a structurally distinct baseline. Office-present culture is the norm here, not the contested policy. For HUMAIN — Saudi Arabia's national AI company operating out of Riyadh's KAFD — and for G42 and Core42 in Abu Dhabi, the question is not whether to mandate office attendance but whether the compensation package is large enough to attract talent that has other options. The answer, at the senior IC level, requires a specific acknowledgment: G42 and Core42 are not enforcing strict five-day mandates for senior researchers and applied scientists. They are running structured collaboration models that look, functionally, more like Anthropic's 3-day San Francisco norm than Amazon's badge-monitored five-day requirement.
The financial architecture in the Gulf absorbs the office-presence norm through the tax-free compensation structure. A Principal ML Engineer at G42 Abu Dhabi earning AED 1.1 million annually (approximately $300,000 at current AED/USD 3.67) takes home 100% of that figure. The equivalent role at Amazon Seattle at $320,000 gross yields approximately $192,000 to $215,000 after federal and state tax at typical effective rates. The Gulf's office-first culture, in this context, is not a talent tax — it is priced into a compensation package that changes the net calculation entirely.
For ENTRA's $2.3M model, the Gulf adjustment is straightforward: the mandate penalty applies where office attendance imposes a cost on employees relative to their counterfactual. In the Gulf, where alternative employer optionality is narrower and the compensation premium is structural rather than negotiated, the attrition differential compresses materially. ENTRA's Gulf-specific cohort in the Recruiter Survey (n=134) shows a 14% voluntary attrition signal for office-present Gulf AI employers — elevated above remote-first peers globally (8%) but well below the 31% mandate signal in the US cluster.
Europe: the legal floor. The EU's contribution to this analysis is architectural rather than merely cultural. Germany's Betriebsverfassungsgesetz (BetrVG §87) grants works councils co-determination rights over working hours and location, creating a procedural layer between a US parent company's global RTO policy and its German AI team's actual working conditions. The result, in practice: every US tech company with a substantial German AI engineering presence operates a de facto hybrid arrangement, regardless of what its global people operations policy states. The BetrVG floor is legally real — not a preference, not a cultural norm, but a co-determination right that requires formal negotiation to modify.
Sweden's medbestämmandelagen (MBL, Act on Co-Determination at Work) creates comparable procedural friction. For AI employers with Stockholm or Gothenburg research and engineering teams — which includes several major frontier lab satellite operations and the growing Nordic AI infrastructure cluster — the MBL requirement for employer consultation before material changes to working conditions means that a five-day mandate arriving from a US headquarters lands in a different legal context than the same mandate deployed in Seattle.
The financial implication for the $2.3M model: EU-based senior AI engineers subject to US parent company RTO policies have a legal option that their US counterparts lack — a formal channel for contesting the mandate, backed by statutory co-determination rights. The attrition signal in ENTRA's EU cohort runs at 24%, below the US's 31%, and the time-to-close differential is narrower in EU markets (44 days at mandate-adjacent EU employers versus 28 days at EU hybrid employers) than in the US (51 versus 28 days). The legal architecture is not eliminating the talent tax in Europe; it is reducing it.
UK: the transition window. The Employment Rights Act 2025 — Royal Assent December 18, 2025, operative April 2027 — has placed UK AI employers in a specific regulatory state: the mandate-enforcement costs of the September 2026 RTO period are accumulating inside the 18-month window before the Act's flexible working provisions become operative employment rights. DeepMind's 2-to-3-day London standard, enforced at the team level with no badge-monitoring infrastructure, reflects a calibration to the UK's impending legal norm. Amazon and Google UK AI satellite operations, running their global mandate against a workforce that will have statutory day-one flexible working rights in April 2027, are absorbing attrition costs now that could have been avoided with a preemptive hybrid accommodation.
The UK-specific adjustment to the $2.3M model: ENTRA estimates the counter-offer premium at UK AI mandate employers running slightly below the US rate (12 to 20% versus 15 to 25%) because the UK talent market offers engineers a credible legal pathway to flexibility without resignation — the April 2027 effective date functions as a negotiating anchor, lowering the risk premium engineers require to stay. The backfill cost, however, is elevated versus the US model when GBP-adjusted: at GBP/USD 1.35, UK senior IC roles that recruit globally face a wider compensation gap between their sterling-denominated offers and the dollar-denominated competing offers from US frontier labs absorbing UK talent.
Who is winning and who is paying
ENTRA's RTO Talent Risk scoring (published September 8, 2026) provides the cross-employer framework for translating the financial model into company-specific exposure. The scoring runs on a 0-100 scale across flexibility posture, counter-offer exposure, addressable talent pool, and hiring velocity: Hugging Face 91 (AAA), Anthropic 55 (BB, elevated by its talent brand buffer), Microsoft 68 (A), Amazon 33 (B), xAI 28 (B).
The winners: capturing the attrition dividend. The three companies capturing the largest share of mandate-driven attrition flows are Hugging Face, Anthropic, and Microsoft — each for a structurally distinct reason.
Hugging Face's distributed architecture gives it the widest addressable talent pool of any employer in the AI sector: 47 countries, no attendance floor, global-first sourcing. Its 28-day median time-to-close — matching the hybrid market median precisely — indicates that fully-distributed employers have not paid a speed penalty for their flexibility. The qualified-applicant ratio of 3× versus B-tier mandate counterparts means Hugging Face is selecting from a deeper pool at faster velocity. In Q3 2026, Hugging Face hired at a pace that ENTRA estimates absorbed 12 to 18% of visible senior IC departures from mandate-enforcing US tech companies in the machine learning research and engineering functions.
Anthropic's absorption is different in character. It is mission-selective — the 38% of ENTRA's Q3 2026 US Candidate Survey respondents who named Anthropic as their top target destination are disproportionately researchers for whom the mission premium is a real offset against the 3-day San Francisco-first model. What this creates is a cohort of senior talent that is willing to accept a non-trivial flexibility reduction relative to Hugging Face or Weights and Biases in exchange for working on Claude's model development. Anthropic's L6 research track, at $480,000 to $740,000 total comp, prices the retention of this cohort explicitly: it is 28% above the equivalent Amazon L6 Applied Scientist track, which is precisely the counter-offer premium that the four-vector model shows is required to retain at a five-day mandate employer. Anthropic is paying that premium proactively, before the retention event, which is more efficient than paying it reactively as a counter-offer.
OpenAI's two-day-per-week suggestion model — enforced through team culture rather than badge infrastructure — positions it between the mandate cluster and the full hybrid tier. In ENTRA's RTO Talent Risk scoring it sits at 42, B tier, but its candidate capture from the mandate cluster runs second in the US market (29% of active job-searchers naming it as primary target). The 18% voluntary attrition figure in OpenAI's own AI research division — captured in ENTRA's Q3 2026 Recruiter Survey — indicates that the flexibility gap between OpenAI and fully distributed employers is generating its own attrition cost, even as OpenAI benefits from the larger gap between it and Amazon.
The cost column: Amazon, xAI, Meta. Quantifying Q3 2026 counter-offer and backfill spend at mandate-enforcing employers requires ENTRA's population model applied to the RTO Talent Risk scores and company-specific attrition signals.
For Amazon, the exposure is most clearly bracketed. ENTRA's Q3 2026 Recruiter Survey found a 2.8× increase in departure inquiry volume from Amazon AI/ML engineers versus Q2 2026 baseline — the operational expression of badge enforcement in performance reviews. Applying the four-vector model to Amazon's estimated 800 to 1,200 L6/L7 AI engineers, at the 41% exit intent rate that the senior-skewed attrition data produces: if 15% of that population converts intent to departure over the next 12 months (ENTRA's conservative actualization rate), Amazon absorbs 120 to 180 Principal-level AI departures. At $400,000 per departure floor, that is $48M to $72M in direct backfill cost — before counter-offer expenditure on the engineers who stay, and before the 51-day time-to-close velocity penalty on each of the resulting open positions.
For xAI, the exposure structure is different. With the lowest RTO Talent Risk score in the cohort (28, B tier) and a 340% year-over-year growth in job postings reflecting genuine expansion ambition, xAI is simultaneously losing engineers at a higher rate per policy intensity than any other employer in this analysis and trying to grow through the same talent market it is contaminating with its posting volume. The rejection rate spike ENTRA's recruiter network observed on xAI's Q3 postings — consistent with the 67% upstream filter rate in the Candidate Survey — indicates that xAI is spending more per hire than its posting volume implies. Cost-per-hire at the five-day mandate employers in ENTRA's Q3 Job Signal Index runs approximately 2.1× the hybrid baseline. For xAI's growth ambitions, that is a compounding cost that its current run rate of postings does not disclose.
Meta's situation carries the most specific structural risk. The relocation bonus cohort — the 67% of eligible Menlo Park AI staff who accepted $10,000 to $25,000 to return to office — is now past its 12-month mark. ENTRA's Q2 2026 survey found 34% of that cohort stating departure intent within the following 12 months. That 34% represents a funded attrition pipeline: Meta paid to create it through the relocation bonus, and it now faces the replacement cost of the departure cohort it subsidized into its office. The four-vector model applied to Meta's AI organization (approximately 3,200 engineers) at a 25% attrition signal produces a blended annual replacement and counter-offer exposure of approximately $220M to $260M — a range that dwarfs the original $50M to $80M relocation bonus expenditure.
The compound effect
The $2.3M per-10-resignation model captures the immediate cost of a departure event. It does not capture the compounding dynamics that make the talent tax progressively more expensive the longer the mandate is maintained. Three compounding vectors extend the financial model into 2027 and beyond.
Geographic concentration risk. Five-day mandates are, by design, geographically anchoring. They require engineers to live within commuting distance of specific offices — San Francisco, Seattle, Austin, Menlo Park. As the talent pool willing to accept that constraint narrows (consistent with ENTRA's data showing the 67% pre-filter rate), the remaining pool concentrates geographically and demographically in ways that reduce its diversity of perspective, institutional origin, and problem-solving approach. The machine learning research literature on team diversity and research output is extensive enough that this is not a speculative concern: homogeneous research teams produce less varied hypothesis generation, and the geographic constraint is a homogenizing force.
The practical expression: mandate employers in San Francisco and Seattle are competing for a shrinking fraction of the global senior AI talent pool, which drives cost-per-hire and time-to-close in the direction the velocity data already shows. By Q1 2027, ENTRA projects the 51-day time-to-close for five-day mandate roles will extend to 58 to 63 days if mandate policy holds — as the pre-filtered candidate pool grows smaller with each quarter of enforcement.
Equity dilution pressure. Counter-offer retention at 15 to 25% total comp lift does not always take the form of cash. At frontier labs and growth-stage AI companies where equity forms a substantial portion of total comp, counter-offers frequently include accelerated vesting, refresh grants, or upfront equity bonuses. These instruments are dilutive in a way that cash counter-offers are not: each equity grant issued to retain an at-risk engineer reduces the per-share value of every other equity holder's position. The cumulative dilution from mandate-driven retention grants across a 1,000-person senior AI engineering organization — running 31% attrition signal and paying 15 to 25% TC lift to retain even half of the at-risk cohort — is material enough to appear in cap table modeling.
ENTRA's Q3 2026 Salary Survey found that 38% of counter-offers at five-day mandate AI employers included an equity component averaging 22% of base salary in new or accelerated grants. At a $350,000 average base for the counter-offer cohort, that is $77,000 in equity per retained engineer. Across a senior IC population of 500 engineers at a mid-size mandate-enforcing AI employer, the equity retention spend runs $38.5M per year — grants issued not to reward performance or fund growth, but to offset the mandate's attrition premium.
The RTO innovation paradox. The strongest argument for strict in-office attendance — that proximity drives research collaboration and accelerates model development — runs directly into the output data of the September 2026 period. The employers with the highest research publication rates and the most active open-source contribution pipelines in the AI sector are not the five-day mandate cohort. Hugging Face's open-source contribution velocity leads the sector. Anthropic's interpretability and alignment research output, measured by arXiv submissions tracked by ENTRA through author affiliation, runs at a higher per-engineer rate than Amazon's AI research division by a factor ENTRA estimates at approximately 2.4× on a rolling-12-month basis. Google DeepMind's research output is an outlier in the mandate-adjacent cluster — but it operates on a 2-to-3-day London standard, not a five-day requirement, and its research productivity data is not cleanly attributable to office presence versus research team quality.
The correlation is not proof of causation: remote-first employers may attract higher-research-propensity engineers, and the research output differential may reflect selection rather than environment. But the burden of proof for the productivity claim rests with the employers enforcing mandates, not with their critics. Three years into the post-pandemic policy cycle, the frontier research employers running the highest per-engineer output are the hybrid and distributed employers. The five-day cohort has not published evidence that contradicts this pattern.
The compound effect of these three forces — geographic concentration eroding pool depth, equity dilution from retention grants, and an innovation gap between flexible and mandate-enforcing employers — means the $2.3M per-10-resignation model is a floor that rises each quarter the mandate is maintained. The September 2026 figure is the cost at the beginning of enforcement. The Q4 2026 cost, if mandates hold, will be higher. The Q1 2027 cost will be higher still.
METHODOLOGY: ENTRA RTO TALENT TAX MODEL
Data sources: ENTRA Q2-Q3 2026 Recruiter Survey (n=134 technical recruiters and talent acquisition leaders, n=1,081 candidate sub-cohort across AI/ML roles; conducted June 15-August 31, 2026); ENTRA Q3 2026 Salary Survey (n=2,847 AI practitioners, 47 countries; conducted July 1-August 31, 2026); ENTRA Job Signal Index Q3 2026 (47,000+ AI postings tracked across US, UK, EU, and Gulf markets); Levels.fyi Q2-Q3 2026 (n=4,100+ verified offers); ENTRA RTO Talent Risk Scoring (published September 8, 2026).
Replacement cost model: industry-standard composite of external recruiter fees (20-25% first-year total comp at senior IC band), internal hiring team time (80+ hours at $150-200 fully-loaded hourly cost), productivity ramp deficit (40-60% effective output during months 1-6), and project disruption estimate ($50,000-$120,000 per senior departure). Knowledge reconstruction costs excluded from base model. The $400,000 figure represents the L7 Principal ceiling; the L6 equivalent averages $200,000-$260,000. The $2.3M per-10-resignation model applies a blended senior IC cost of $185,000 per position, weighted to the observed seniority distribution of the departure pool (approximately 55% L6/L7, 30% at L5, 15% at L4 equivalent).
Counter-offer premium model: derived from ENTRA Q3 2026 Salary Survey counter-offer data (n=847 senior IC retention events at AI employers with active attendance policies; n=312 at hybrid/remote-first employers). The 15-25% five-day mandate premium and 5-8% hybrid premium represent the weighted-average TC lift offered at point of competing offer, not the accepted premium. The sunk-cost component of the $2.3M model reflects the share of counter-offers extended before departure (approximately 60% of the resignation cohort receives a counter-offer before resignation is finalized; average tenure extension before departure is 4.2 months at the current enforcement quarter).
Time-to-close: ENTRA Job Signal Index Q3 2026 median days-to-fill for roles with explicit five-day mandate language versus hybrid (3-day) language, matched by function (ML Engineering, Research Scientist, Applied Scientist) and seniority band. Median 51 days (five-day mandate) versus 28 days (hybrid), calculated on roles posted and filled within the Q3 2026 tracking window.
All USD figures. GBP/USD 1.35; EUR/USD 1.09; SAR/USD 3.75; AED/USD 3.67. UK figures use GBP/USD 1.35; EU figures use EUR/USD 1.09. Gulf figures use SAR/USD 3.75 for HUMAIN and SAR-denominated roles; AED/USD 3.67 for G42 and Core42. ENTRA headcount estimates for named employers carry higher uncertainty than SEC-reported figures and are derived from LinkedIn headcount tracking, public earnings disclosures, and recruiter network intelligence. The $2.3M per-10-resignation figure is ENTRA's own model; readers should treat it as a central estimate within a $1.8M-$2.9M sensitivity range based on seniority-distribution assumptions and counter-offer actualization rates.
This series opened on September 11 with the enforcement quarter's defining ratio: a 5× attrition multiplier separating five-day mandate employers from their remote-first peers. Every piece since has mapped a different dimension of that number — the regional legal architectures compressing it or amplifying it, the lab-level talent moats moderating its worst effects for a handful of well-branded employers, the European works council vetoes embedding a hybrid floor that US parent companies cannot unilaterally remove.
This report closes with the number the multiplier produces: $2.3 million per 10 senior ML resignations. Four vectors feeding it. The attrition rate (31% versus 8%). The backfill cost ($400,000 per L6/L7 Principal departure, applied to the seniority-weighted departure pool). The counter-offer escalation premium (15 to 25% TC lift required versus 5 to 8% at hybrid employers). The velocity degradation (51 days median versus 28 days). None of these vectors is speculative. All four are measured, sourced, and independently replicable.
The employers paying the talent tax are not ignorant of these costs. They have internal data. The question the Q4 2026 earnings cycle will answer — first from Amazon, then from Google, then from Meta — is whether the costs as internally measured have reached the threshold at which the mandate's strategic rationale is publicly re-examined. ENTRA's base-case scenario (50% probability) projects that L6/L7 departure data from Q3 performance reviews triggers a selective carve-out at one major mandate employer before year-end. If that happens, the $2.3M model will have served its intended function: giving the employers running the calculation a common reference point for what they are deciding.
The companies that get ahead of this data in Q4 2026 will avoid rebuilding senior AI teams from the ground up in 2027. The companies that hold the mandate past the compounding point will face the full arithmetic of what a five-day policy costs in the only market where the talent cannot be repriced, outsourced, or automated into irrelevance.
That would be the market they are actually in.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.