Bandwidth contracts can be defined as fixed-cost billing for 95th +/- percentile of usage per period with charges of peak bursts exceed x% of period, without charging for bytes transferred at all; before S3, this was typical, and it’s still available from providers today. Under that model, free user rate limits ensure that their peak possible bandwidth bill from concurrent unpaid users fits within their (cheap and fixed) contractual limits, while paid users pay for a different endpoint whose bandwidth contract is usage-billed and/or has a higher 95th percentile rate and/or has a higher peak burst cap so that rate limits aren’t necessary. This hybrid approach lets them provide both {cheap, reliable, slow} free service, and {costly, reliable, fast} paid service at a much lower total cost than if they went with one or the other approach exclusively. (Elsethread, a suggestion that they switch to torrents would provide {cheap, unreliable, fast} service, completing the “pick two” trifecta.)