Seventy-three percent of Amazon employees polled on Blind said they are considering a job change or actively exploring new roles because of the company's five-day return-to-office mandate — and the attrition pattern that has crystallized across the eight months since January 2025 enforcement began is not random. The engineers leaving are ML Engineers, Applied Scientists, and Principal-level researchers: precisely the job families Amazon's AGI program, Trainium silicon operation, and Alexa+ product line need most. Their destinations — Anthropic's Seattle office, OpenAI in San Francisco, xAI, and Google DeepMind — are paying 2x to 3.5x Amazon's comparable compensation bands, and Amazon's November 2025 compensation review closed none of that gap.
The Attrition Signal
Amazon's own internal data flagged the problem before January 2025 enforcement arrived. An internal HR document covering Amazon's non-retail businesses — AWS, advertising, devices, and the AGI team — stated that "GenAI hiring faces challenges like location, compensation, and Amazon's perceived lag in the space," according to Business Insider reporting on the document in late 2024. That was a pre-enforcement warning. The post-enforcement signal is sharper.
An internal Slack channel survey, leaked to the press in October 2024, recorded average employee satisfaction with the five-day schedule at 1.4 out of 5. The Blind poll put overall dissatisfaction at 91%, with 73% of respondents considering a job change specifically because of the RTO requirement. These are sentiment readings, not official churn statistics — but the departure behavior followed the sentiment data closely.
The attrition is concentrated by level. ENTRA Job Signal Index Q2 2026 tracking of Amazon AI-relevant role postings against LinkedIn departure signals shows the heaviest separation rates at SDE II through Principal Engineer levels — L5 to L7 — precisely the band with the most inbound interest from frontier labs. Applied Scientists and ML Engineers at L5 and L6 who faced a thin external market in 2022 now receive standing inbound contact from Anthropic, OpenAI, and xAI recruiters. Senior engineers exiting Amazon in Q1 and Q2 2026 cited the RTO mandate as the proximate trigger, layered over compensation compression relative to the lab market and what multiple LinkedIn departure posts described as structural consolidation fatigue following the Peter DeSantis AGI reorganization in December 2025.
The geographic enforcement compounds the pressure. Amazon does not only require five days in-office — it requires proximity to designated hub offices, treating refusal to relocate as voluntary resignation. Engineers who built distributed lives during the 2020-2022 remote period have no hybrid accommodation path. That binary choice — relocate or resign — is converting ambivalent employees into confirmed departures at a rate that industry research corroborates: a 2026 workplace study found that 42% of companies implementing strict RTO requirements experienced higher-than-expected attrition, with high performers showing 16% lower intent to stay than their counterparts at hybrid-model employers.
Oracle, which moved aggressively to absorb Amazon engineering talent as RTO friction built through 2024 and 2025, absorbed over 600 Amazon employees across the two-year window, according to Business Insider sourcing cited in coverage of Amazon's policy. The AI labs represent a newer but faster-moving version of the same dynamic — with the added leverage of compensation packages that are structurally superior, not merely competitive.
Where Amazon's AI Talent Went
The departure destinations form a legible pattern across ENTRA Job Signal Index Q2 2026 data and publicly visible LinkedIn career transitions tracked through H1 2026.
Anthropic is the primary capture point. Anthropic's Seattle engineering office, situated in the South Lake Union neighborhood approximately two miles from Amazon's Day 1 campus, removes the relocation friction that keeps some candidates from making the move. An Amazon ML Engineer living in Bellevue can take a role at Anthropic Seattle without materially changing their commute — and gain remote flexibility, research mission alignment, and a compensation structure that is categorically higher. SignalFire's 2026 State of Talent Report ranked Anthropic at the top of its engineer retention index, with an 80% two-year retention rate — and noted that Amazon ranked near the bottom of the same metric, trailing Meta, OpenAI, and Anthropic. The inflow and outflow data are not independent.
OpenAI is the second primary destination. Its San Francisco requirement filters for engineers willing to relocate — a higher bar than Anthropic Seattle — but for Amazon L7 Principal Engineers and above, the relocation premium is offset by OpenAI's compensation ceiling (median engineer total compensation of $555,000, extending to $1.28 million at senior levels, per jobsbyculture.com's 2026 OpenAI compensation analysis) and the equity structure tied to OpenAI's $300 billion valuation. ENTRA Job Signal Index Q2 2026 cross-referencing of publicly visible career transitions identifies a specific source population: researchers from the Amazon Nova model team and the Trainium pre-training infrastructure group, whose skill sets in large-scale pre-training and custom silicon map directly to OpenAI's infrastructure and model org requirements.
Google DeepMind is the third pathway, most active for Applied Scientists with research or academic backgrounds who joined Amazon from PhD programs. The Amazon Science research brand has existing relationships with the academic research community that DeepMind recruits from; when those scientists leave Amazon, DeepMind's network is already warm. xAI is a smaller but present destination for engineers motivated by Grok's model training challenges and xAI's compensation ceiling, which has benchmarked aggressively since the company's Series C in 2024.
The least-reported destination is Microsoft, which operates its Copilot and Azure AI engineering organizations from Redmond — physically closer to Amazon's Eastside offices than any San Francisco lab. ENTRA Job Signal Index Q2 2026 tracking identifies a secondary flow from Amazon's AWS AI Services team into Microsoft AI Platform roles, where the compensation for senior research engineers runs $700,000 to $1.4 million for specific AI org tracks, per ENTRA's June 2026 coverage of Microsoft's H1 compensation architecture. For engineers unwilling to relocate out of the Seattle metro, Microsoft is the path-of-least-resistance alternative that the lab narrative obscures.
The Compensation Premium for Lab Defectors
The gap between Amazon's ML engineering bands and frontier lab compensation is the structural fact underpinning every attrition conversation in Seattle's AI talent market.
Amazon's Machine Learning Engineer at L5 sits at approximately $287,000 total compensation, per Levels.fyi 2026 data. At L6, the Levels.fyi median runs $320,000 to $390,000 depending on team and tenure, with the 6figr 2026 Seattle dataset extending to $483,000 at the high end. The structure is front-loaded: Amazon's year-one and year-two packages include signing bonuses that create a year-three cliff when vesting completes without refresh. That cliff is when the largest cohort of departures occurs — and the RTO mandate is accelerating the decision to exit before it arrives.
Anthropic's ML Engineer at a comparable seniority level (approximately L5 in Amazon's leveling) carries total compensation of $625,000 to $1,000,000, per CTAIO's 2026 compensation dataset and Levels.fyi filings. OpenAI's median total compensation for engineers is $555,000. Google DeepMind's L7 equivalent carries $850,000 to $1.2 million in liquid GOOGL stock plus base, per PIN's 2026 AI compensation benchmarks. The effective premium for an Amazon L5 ML Engineer accepting a lab offer ranges from 2.2x (Anthropic L5 low end versus Amazon L5 median) to 3.5x at the top of DeepMind's senior band.
Amazon's November 2025 compensation review did not close this gap. The review restructured performance-based payouts to favor long-tenured high performers: employees achieving the Highly Valued 3 (HV3) rating for the first time received 40% of their pay range instead of the prior 50%, per GPA.net/Americas reporting on the changes. Employees newly promoted to Highly Valued 2 received 10% of their pay range, down from 20%. The review was calibrated to reward retention of proven performers over multiple review cycles — not to compete against frontier lab initial offers. Candidates holding an Anthropic or OpenAI term sheet read the restructure as confirmation, not reassurance.
Amazon recruiters were frank about the dynamic before enforcement began. Candidates were declining Amazon offers specifically because of RTO, accepting lower total compensation at remote companies in exchange for work-arrangement flexibility, per accounts Business Insider gathered from Amazon recruiters in late 2024. By mid-2026, the trade-off had inverted: candidates are not accepting lower pay. Lab offers are paying more and offering flexibility simultaneously. Amazon is now competing on both dimensions against employers that are winning on both.
What's Next
Two signals define Amazon's AI talent posture in Q4 2026.
A compensation recalibration for AGI and ML tracks. Peter DeSantis, who consolidated Amazon's AGI model development, Trainium silicon, and quantum computing organizations under a single SVP structure in December 2025, has not publicly addressed the lab compensation gap. If Amazon moves in Q4 to raise its L5 and L6 ML Engineer bands — lifting total compensation to the $400,000 to $500,000 range for AGI-specific roles and introducing lab-style equity structures with shorter vesting periods — it would signal that the attrition is now measurable enough at the leadership level to require a structural response. An absence of a comp announcement through Q4 would signal the opposite: that Amazon continues to wager its DeSantis pitch (AGI at hyperscaler scale, Trainium silicon, AWS distribution) against frontier lab packages, a bet that H1 2026 data does not support closing senior candidates.
Anthropic's Seattle headcount trajectory. Anthropic has disclosed no specific Seattle headcount target, but its hiring velocity in the Pacific Northwest — tracked by ENTRA Job Signal Index across open requisitions and LinkedIn hiring signals — has run above the company's overall 370% year-over-year rate (ENTRA Talent Index, July 2026). An Anthropic Seattle office that reaches 200 engineers or more by year-end removes the geographic friction that has historically kept a subset of Amazon engineers in place. At that headcount, Anthropic Seattle becomes a full product engineering presence rather than a satellite recruiting post — and the lab-versus-hyperscaler attrition dynamic converts from a national talent contest into a South Lake Union block-by-block one.
Amazon built the playbook for winning technology talent wars. The five-day mandate has handed it to the labs next door.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.