Adyen (ADYEN) Trains AI Model on 51 Trillion Tokens, Tackling Fraud

Adyen (ENXTAM:ADYEN), the Amsterdam-based global payments giant, has revealed a groundbreaking AI initiative: the development of its Transaction Foundation Model (TFM), trained on a massive 51 trillion tokens. This model aims to revolutionize fraud detection and payment optimization by analyzing raw payment sequences rather than relying on traditional feature engineering.

The scale of the dataset is extraordinary—200 billion payments, each with sequences of up to 256 tokens, creating a training corpus over twice the size of the 22 trillion tokens used for Llama 4. By uncovering behavioral patterns within transaction histories, Adyen’s TFM enhances the company’s ability to spot fraud and optimize payment authorizations. Testing revealed a more than 200 basis point improvement in performance compared to previous tabular models, particularly as training scaled to hundreds of millions of payments.

To support TFM, Adyen overhauled its infrastructure. Legacy PySpark pipelines, which took four to six hours for data preparation, have been replaced with a modern Python-based system using Ray. This single streaming pipeline allows preprocessing, tokenization, and training to scale independently across multiple nodes. This update not only boosts reliability but also enables engineers to run five times more experiments per week, accelerating innovation.

Strategic Timing Amid Market Challenges

This announcement comes at a crucial time for Adyen. In February 2026, the company faced a sharp 15% drop in its share price after soft payment volume overshadowed its revenue growth, with analysts questioning its 2026 guidance of 20–22% net revenue growth. For context, Adyen processed €745 billion in transactions in H2 2025 but missed market expectations, contributing to investor concerns.

The TFM project aligns with Adyen’s strategy of embedding intelligence directly into its single-stack payments architecture, which differentiates it from competitors. By leveraging its global transaction data across 400 payment methods, 143 countries, and 230 currencies, the company aims to improve authorization rates while minimizing fraud and chargebacks—key metrics in the payments industry.

Why It Matters for Traders

Adyen’s push into large-scale AI could bolster trust in its platform, particularly among high-volume merchants dependent on secure, efficient payment processing. If TFM delivers on its promise to enhance fraud detection and approval rates, it could improve merchant retention and expand Adyen’s market share. These operational gains may help offset investor skepticism around its growth trajectory.

However, traders should watch closely for updates on TFM’s impact on key metrics like fraud loss rates and authorization uplift. Adyen’s AI initiative also highlights its broader commitment to innovation in a sector increasingly reliant on machine learning enhancements. This could position the firm as a leader in payments intelligence, potentially reversing its recent share price decline over time.

Adyen’s next big moment is likely to be its mid-2026 financial report. If the company can demonstrate measurable improvements from TFM, it may provide a significant catalyst for its stock price, which has been under pressure. For now, the market will be watching to see if this AI bet translates into tangible financial results.

Image source: Shutterstock