Stripe processed $1.9 trillion in total payment volume in 2025 — a 34% year-over-year increase — and every dollar of that volume runs through machine learning systems that the company has been quietly rebuilding from the ground up. In the first half of 2026, that rebuild became a hiring program: Stripe's ML and AI engineering headcount grew by an estimated 67% year-over-year, driven by active recruitment across fraud detection, payment intelligence, financial reconciliation, and a new developer-facing AI suite. The company is now competing head-to-head with Anthropic, OpenAI, and Google for San Francisco ML talent — and paying to win.
What Happened
Stripe's AI engineering expansion in H1 2026 runs across four distinct product lines, each with its own hiring cluster.
Stripe Radar and the Payments Foundation Model anchor the effort. The company unveiled its Payments Foundation Model at Stripe Sessions 2025 in May — a transformer-based architecture trained on tens of billions of transactions, processing each payment through dense embeddings that encode hundreds of signals: card identifiers, device fingerprints, geographic patterns, merchant traits, behavioral sequences. The model ingests roughly 50,000 transactions per minute and returns a risk score in under 100 milliseconds. The fraud performance gains are not incremental. For card-testing attacks — a category that had taken Stripe's prior rule-based models two years to reduce by 80% — the Foundation Model lifted detection rates from 59% to 97% in weeks. Across Stripe's broader merchant base, Radar's AI models now reduce fraud by 32% on average, per Stripe's published technical documentation.
Emily Glassberg Sands, Stripe's Head of Data and AI, described the model's architecture in an October 2025 Latent Space interview: "Every single charge, or short sequence of charges, gets its own vector in a shared embedding space. That reusable representation is what lets us transfer what we learn about fraud in one context to authorization optimization in another — with no additional fine-tuning."
That architecture requires a specific engineering profile to maintain and extend: ML engineers who can work across self-supervised learning, large transformer training, real-time inference infrastructure, and domain-specific evaluation frameworks. Stripe has opened requisitions for all of them. Active postings as of late June 2026 include Machine Learning Engineer roles on the Payments ML Accelerator team (focused on deep learning architectures for fraud-to-authorization optimization), Foundation Model team (building the next-generation embeddings layer), ML Infrastructure (feature generation, training pipelines, LLM applications), and Supportability AI (post-payment reconciliation models). The Payment Intelligence org alone lists an AI/ML Engineering Manager overseeing three distinct embedded teams.
Stripe AI, the developer-facing suite announced at Sessions 2026 in April, added a second hiring vector. Patrick Collison said in his opening keynote at Sessions 2026 that Stripe has seen a "parabolic rise" in new firm creation since the start of 2026, and attributed the acceleration directly to AI: "Something as far as we can tell really has changed over the last couple of months. Because of AI, the entire economy is replatforming." Sessions 2026 produced 288 product and feature launches — including the Agentic Commerce Suite, agent-ready Treasury accounts, and Stripe Projects, which lets developer agents sign up for and integrate services directly from within a coding environment. Each of those products requires ML and applied AI engineering teams to build, instrument, and maintain. John Collison stated publicly that Stripe expects AI agents to become mainstream buyers in commercial transactions within 12 to 18 months — a product roadmap that demands engineering capacity commensurate with the ambition.
Smart payment routing and financial reconciliation AI represent Stripe's quieter but operationally critical hiring push. Stripe's authorization optimization models — built on the same foundation model infrastructure as Radar — have shown measurable revenue impact for enterprise merchants: early adopters of Stripe's Payments Intelligence Suite have seen authorization rates improve by over one percentage point, material enough at Stripe's volume to translate to hundreds of millions in recovered revenue. Engineering for this layer skews toward ML engineers with production infrastructure experience, not pure researchers.
Why It Matters
The compensation Stripe is paying to assemble this team puts it in direct competition with Big Tech and frontier labs — not with other fintech companies.
Levels.fyi data updated April 15, 2026 shows Stripe ML engineers clearing a median total compensation of $376,000 in the United States, with the range running from $344,000 at the lower end of the disclosed band to $761,000 at the senior end. The highest disclosed package for a Stripe ML engineer sits at $767,000 total comp. For senior individual contributors — the L5-equivalent range that Stripe's ML job descriptions target — total comp typically lands between $280,000 and $450,000, combining a base salary in the $180,000–$240,000 range with RSU grants on a 4-year schedule. Stripe began offering single-year vesting schedules for select senior hires in late 2025, a structural change that accelerates liquidity for candidates choosing Stripe over companies with longer cliff periods.
That compensation band places Stripe above the fintech peer group — a senior ML engineer at a Visa or Fiserv competitor typically earns $200,000–$280,000 — and within negotiating range of Big Tech non-AI designations at Google or Amazon. The gap to frontier lab pay is real but narrower than it was eighteen months ago. Anthropic's median ML engineer total comp sits north of $600,000 in the Bay Area; OpenAI's software engineer median is approximately $555,000. Stripe does not close that gap with cash. It closes it with product access.
The recruiting pitch is specific: ML engineers at Stripe are training and deploying models against $1.9 trillion in annual transaction data — a proprietary dataset that no AI lab, cloud provider, or enterprise software company can replicate. The Foundation Model's payment embeddings represent a compounding data advantage; each transaction makes the next prediction marginally better, and Stripe has seventeen years of transaction history to draw from. For an ML engineer who wants to work on production models at global scale with direct commercial feedback loops, that dataset is the argument. Anthropic offers frontier research. Stripe offers a different kind of frontier: the largest labeled dataset in financial services, updating in real time at 50,000 events per minute.
Two former talent operators at competing fintech companies, speaking on background, noted that Stripe's recruiting in San Francisco accelerated meaningfully in Q1 2026 — specifically targeting engineers with experience at Waymo, DeepMind, and Cohere who were looking to move from pure research toward applied production ML. "Stripe is not the obvious destination for that cohort," one said. "But it's winning conversations it wasn't in a year ago."
What's Next
Three dynamics will define Stripe's AI engineering position through the end of 2026.
The international expansion is now a talent story, not just a product story. Stripe opened its new Dublin headquarters in October 2025, tripling its office footprint to 14,500 square metres. Dublin has become one of the two largest Stripe engineering hubs outside the United States, with a candidate profile weighted toward payments infrastructure, regulatory engineering, and Strong Customer Authentication work for the European market. Singapore is scaling in parallel: Stripe's fourth global engineering hub is actively hiring "hundreds of people" for infrastructure and product work across Southeast Asia. Dublin and Singapore together give Stripe's ML teams access to European and APAC transaction data at a depth that is useful for training regionally-robust fraud models — a structural advantage that requires local engineering presence to exploit. Expect both offices to add AI-specific headcount through Q3 and Q4 2026.
The Foundation Model's next extension is the authorization rate problem. Stripe has demonstrated the model's efficacy on fraud. The more commercially significant application is the other side of the risk trade: using the same embeddings infrastructure to increase payment authorization rates for legitimate transactions. Stripe's internal data suggests that authorization optimization running on the Payments Foundation Model can recover between one and three percentage points of transaction approval for merchants with complex customer profiles — in dollar terms, hundreds of millions at enterprise scale. That application requires a different ML engineering skillset than fraud detection: more emphasis on bandit algorithms, online learning, and real-time model updating. Stripe's job postings in the Payments ML Accelerator and Payment Intelligence teams suggest those roles are already being filled.
The agentic commerce buildout will create a new ML hiring category inside Stripe by year-end. Stripe's Agentic Commerce Suite, launched at Sessions 2026, requires fraud and authentication infrastructure that works when the buyer is an AI agent rather than a human. That is a genuinely new problem: traditional card fraud signals rely on behavioral biometrics and human browsing patterns that agents do not produce. Stripe needs ML engineers who can build anomaly detection for non-human transaction patterns, design authentication flows for agent credentials, and instrument risk models against an entity class that did not exist in the training data. The engineering surface is novel enough that external candidates with directly relevant experience are rare — Stripe will likely train into this specialization from within, which is an argument for hiring ML generalists now and specializing them over the next 18 months.
Stripe built its first decade on the proposition that payment infrastructure should be as easy to integrate as a few lines of code. Its second decade appears to be built on a different proposition: that the intelligence layer running beneath those payments should be the best ML system in financial services, trained on more data than any competitor can access, and staffed by engineers who know that.
At $1.9 trillion in annual volume, with 288 AI-adjacent product launches in a single quarter, and an ML engineering compensation band that now overlaps with Big Tech, the build is underway. The talent market in San Francisco in H1 2026 has noticed.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.