This is an illustrative scenario, not an account of a specific client engagement. Consider a data scientist at a Lagos fintech company. For most of the work — pandas dataframes, sklearn models, exploratory analysis — a laptop is sufficient. But for heavier tasks — training XGBoost on millions of rows, running SHAP explanations on complex models, or anything requiring GPU acceleration with PyTorch — the default is Google Colab.
Colab is free up to a point. But session disconnections, memory limits, and the latency of uploading datasets to the cloud become constant friction. The requirement is a local machine that handles the full range of the workload.
The Build
Working back from that workload, the spec lands here:
- AMD Ryzen 9 7950X (16 cores — pandas operations, cross-validation loops, and parallel feature engineering all use CPU cores)
- 128GB DDR5 RAM (entire datasets in memory — no sampling)
- NVIDIA RTX 4070 Ti Super 16GB (CUDA for PyTorch, RAPIDS for GPU-accelerated dataframes)
- 2TB NVMe PCIe 5.0 (fast I/O for large CSV and Parquet file reads)
- Windows 11 Pro with WSL2 (full Linux environment within Windows for her Python toolchain)
What Changes
The 128GB of RAM is the decision that changes the work: entire datasets sit in memory, so you analyse the full dataset rather than a sample. No session timeouts, and no waiting on uploads before a job can start.
For a fintech workload there is a second argument that has nothing to do with speed: the data stays in your environment. No cloud uploads, no privacy questions to answer about where client records went.