An ecommerce company receives several gigabytes of customer data daily and trains a demand-forecasting model. They need to run inferences once per day. Which inference type is most suitable?

Batch inference is most suitable because the company processes a large volume of data (several gigabytes daily) and only needs to run inferences once per day. This allows for efficient processing of data in large chunks at scheduled intervals, which aligns perfectly with the requirement for daily forecasting. Asynchronous inference is similar to batch but often implies a more immediate, though not real-time, response for individual requests, which isn't the primary need here. Real-time inference is designed for immediate, low-latency predictions on individual data points, which is overkill and inefficient for daily, large-scale processing. Serverless inference describes a deployment model where the infrastructure scales automatically, but it doesn't specify the inference type (batch vs. real-time) and can be used for either.