Optiver puts 30–40% of its 950 engineers on platform work — and says better AI models now get more focus than lower latency
The Pragmatic Engineer's 2026-08-11 deep dive on the proprietary trading firm reports ~2,200 employees including ~950 engineers and ~1,000 traders/researchers, €4.5B ($5.1B) trading income and €1.7B ($1.95B) profit in 2025, and 10M+ trades daily across 100 exchanges. The structural outlier is platform investment: 30–40% of engineers, versus a typical 15–20% at tech companies. Stack is C++ for latency-critical paths, Python for modeling, Rust emerging in research and orchestration, VHDL/SystemVerilog for FPGAs, on custom Linux kernels with Kafka, Postgres (they upstreamed nanosecond-precision timestamp work), Databricks and GitHub Actions on bare metal. The headline shift: option-repricing 'retreat' systems went from seconds a decade ago to nanoseconds, and now the platform team's newest work is an AI gateway and MCP hosting platform for agent workflows.
↳ Follow the thread