Amazon has spent H1 2026 executing an AI hiring campaign that has largely escaped the coverage volume directed at OpenAI or Anthropic — and that relative quiet is itself the strategy. AWS generated $37.6 billion in Q1 2026 revenue, up 28% year-over-year, the fastest growth rate the segment has posted in 15 quarters. Behind that number is an org restructure, a consumer AI product launch, and an 11,000-person intern and engineering hiring push that positions Amazon as the largest AI delivery machine in the US market — even as its frontier model ambitions remain in early formation.
What Happened
The structural headline is the org redesign Andy Jassy announced on December 17, 2025. Rohit Prasad, who had led Amazon's AI function and overseen the Nova model program since 2023, left the company at year-end. Jassy replaced the dispersed AI leadership structure with a unified organization combining the Nova and frontier model research teams, Amazon's custom silicon division — Graviton, Trainium, and Nitro — and its quantum computing group, all reporting to Peter DeSantis, a 27-year Amazon veteran who had previously run AWS infrastructure. DeSantis reports directly to Jassy.
Within that structure, Pieter Abbeel — the UC Berkeley robotics researcher who joined Amazon in 2024 through its acqui-hire of Covariant — leads frontier model research inside the AGI organization. The AGI SF Lab, seeded in December 2024 with personnel from Adept and led initially by Adept co-founder David Luan, lost its founding director in February 2026. Luan departed less than 18 months after joining, saying he was leaving "to cook up something new" (GeekWire, February 24, 2026). Four of the five Adept co-founders have now exited Amazon: Erich Elsen left after four months and is now a principal research scientist at Databricks; Augustus Odena and Maxwell Nye both departed after roughly a year and are now research scientists at Meta. Adept co-founder Kelsey Szot and agent model training lead Bryan Silverthorn remain. The lab's mission — building AI agents that operate across software, browsers, and real-world interfaces — is unchanged, now operating inside Abbeel's frontier research remit.
The geographic split is deliberate. The AGI SF Lab operates out of San Francisco. The Nova model and silicon teams are anchored in Seattle and Bellevue — specifically the Bellevue campus, not Redmond. Amazon's Palo Alto offices carry applied ML and research roles, including Annapurna Labs ML Systems positions that posted for 2026 entry-level hires this spring. The result is a three-node structure across the US AI talent map.
On the consumer side, Amazon made Alexa+ available to all US customers in February 2026, pricing the product at $19.99 per month (free with Prime). Panos Panay, who joined Amazon from Microsoft in 2024 to lead Devices and Services, has described the rebuild as "infusing Alexa with AI" in a way that represents a platform shift, not a feature upgrade. The Alexa org pulled forward hiring of product managers, applied scientists, and inference engineers through Q1 2026 to support a full-release timeline that compressed roughly six months faster than originally scheduled.
Campus hiring reflects the scale of the build. AWS CEO Matt Garman announced at the company's "What's Next with AWS" event that Amazon is staffing one of the biggest software engineering intern classes in its history for 2026 — approximately 11,000 offers — with AI and machine learning roles representing the densest cluster of that demand. The announcement followed the elimination of roughly 30,000 corporate positions across late 2025 and early 2026. The net-negative headcount story on the surface conceals a deliberate sectoral reallocation underneath: Amazon finished Q1 2026 with approximately 1.57 million total employees, roughly flat year-over-year, while its AI-specific hiring accelerated.
Compensation bands have moved. Levels.fyi self-reported data through mid-2026 shows Amazon's Machine Learning Engineer total compensation ranging from $176,000 at L4 to $409,000 at L6, with a median MLE package across all levels of $265,000. Applied Scientist packages run from $245,000 at L4 to $653,000 at L7, with a median of $335,000. For L6 Machine Learning Scientists specifically, the band sits at $350,000 to $400,000 in total comp. Those figures sit below what frontier labs pay at comparable seniority — Levels.fyi self-reported data through May 2026 shows frontier-lab senior ML engineers clearing $470,000 to $630,000 median total comp — but Amazon has adjusted its RSU grant structures to narrow part of that gap. Because Amazon's equity vesting back-loads into years three and four, larger initial grant sizes are the primary lever recruiters are using to compete for senior ML candidates with faster-vesting lab offers.
The Anthropic relationship adds a dimension no other Big Tech company can replicate. Amazon has committed a cumulative $13 billion to Anthropic across multiple tranches, with up to $20 billion more contingent on commercial milestones (CNBC, April 20, 2026). Anthropic's commitment to deploy up to 5 gigawatts of Amazon Trainium capacity has made AWS the primary compute substrate for Claude training and inference. More than 100,000 customers now run Claude models on AWS. AWS's AI revenue run rate crossed $15 billion in Q1 2026 — Bedrock processed more tokens in Q1 FY2026 than all prior years combined, with 170% customer spend growth quarter-on-quarter.
On April 28, 2026, Amazon formalized a separate and strategically distinct partnership: OpenAI's GPT-5.5, GPT-5.4, and Codex models landed on Amazon Bedrock one day after Microsoft's exclusivity arrangement with OpenAI expired. The arrangement is part of a $38 billion, seven-year strategic partnership AWS and OpenAI signed in November 2025. The effect on Amazon's competitive positioning in enterprise cloud AI is direct: AWS can now offer customers OpenAI and Anthropic models through a single managed interface, while Azure rebuilds its model distribution strategy. "The opportunity ahead of us is enormous, and the most exciting part is that this is not something in the future — it's starting right now," Sam Altman said in a recorded video at the Bedrock launch event.
Why It Matters
Amazon is running a structurally different talent play than the frontier labs, and the distinction matters for benchmarking the US AI hiring market at the H1 2026 mark.
OpenAI and Anthropic are competing for researchers who want to work directly on the model stack — pretraining, alignment, evals, inference optimization. The talent they need is rare, the comp is extraordinary, and the hiring volumes are in the hundreds. Amazon's 11,000-engineer campaign is not competing in that market. It is building the delivery and application layer that sits on top of it — the inference engineers, the Bedrock integration specialists, the agent tooling developers, the Alexa product scientists — at a scale that only AWS can absorb.
But Amazon is not conceding the frontier. The AGI SF Lab, despite the Adept co-founder attrition, is still the company's clearest statement that it intends to build foundational model capabilities internally, not just distribute other labs' work. Abbeel's appointment as frontier research lead is a credible signal to the research community — he is a published robotics and machine learning researcher with a CV that earns attention. The lab is recruiting what it describes publicly as "a few dozen passionate, talented people" from disciplines including physics, math, and quantitative finance, not just AI engineers. That framing is deliberate: it signals a research culture, not an engineering execution org.
The comp reset at the L6 and L7 bands is an implicit acknowledgment that Amazon lost mid-senior ML engineers to Anthropic, Google DeepMind, and Scale AI at a rate that became operationally costly in 2024 and 2025. The RSU grant adjustment is a quiet acknowledgment that the prior equity package was not retaining the specific profile of engineer the new org requires: people who can work at the interface of model deployment, Trainium silicon, and AWS product. Garman said publicly at the "What's Next with AWS" event that Amazon is hiring developers "at the fastest rate in years," while noting that expertise in "authoring Java code" will be less valuable than AI tooling fluency — a public signal to candidates about what Amazon wants.
The David Luan departure from the AGI SF Lab is the one cautionary data point in an otherwise expansion narrative. Luan was the anchor hire that gave the SF lab credibility in the agent research community. His exit, less than 18 months in, sends a signal that the market reads as leadership instability — in a talent environment where founding-team continuity is a primary variable researchers use to evaluate whether a lab is worth joining. Amazon has not publicly addressed that signal. The lab will need to demonstrate research output, not just organizational structure, to fully restore its recruiting position among the agent research cohort Luan had assembled.
What's Next
Three things to watch in H2 2026:
1. Whether the DeSantis unified org produces model output. Peter DeSantis is an infrastructure executive, not an AI researcher. The organizational logic of combining Nova models, Trainium silicon, and quantum computing under a single infrastructure-oriented leader is internally coherent. But the research community evaluates labs by published work and benchmark performance, not org charts. The first major Nova release under DeSantis's tenure, expected in H2 2026, is the signal event for whether the consolidation produced a research organization or an engineering delivery function.
2. Whether the Anthropic and OpenAI partnerships create internal talent tension. Amazon's $13 billion Anthropic investment and its $38 billion OpenAI partnership make Amazon simultaneously the primary compute and distribution partner for the two frontier labs that its own AGI org is competing with for talent. Amazon's investment in Anthropic came with no non-compete or talent-sharing agreement — meaning the two organizations are actively competing for the same San Francisco research community despite the capital relationship. Garman has described the dynamic as "complementary," but every senior ML engineer in SF who receives offers from both Amazon's AGI lab and Anthropic is living the tension directly.
3. The Alexa+ retention test. The engineers and applied scientists who rebuilt Alexa from scratch and shipped Alexa+ to 300 million deployed devices are now, post-launch, the most portable profiles in the US consumer AI talent market. Their experience — shipping a live, production AI agent at consumer scale — is exactly what every frontier-adjacent startup and AI product company in the US is recruiting for. At Amazon's current L4-to-L6 MLE bands, those profiles command 20–40% premiums in the outside market. Whether Amazon can retain the Alexa+ core team through H2 2026 is the retention question that matters most for its consumer AI roadmap.
Amazon entered H1 2026 as the quietest large-scale AI hiring machine in the US market; the H2 org is structured to make that quiet unsustainable.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.