This paper adds to a growing literature on the willingness to pay (WTP) for internet services. We surveyed 5,200 respondents across four demographically diverse United States (US) states, and developed a curve-fitting WTP estimation technique that builds on existing conjoint analysis models. We find that users are willing to pay an extra $1.13 per month for a 1 megabits per second (Mbps) faster internet and $45.52 per month for better connection quality. We document a strong non-linearity: a given speed increment generates the most value for users with the slowest internet but confers negligible value on users with the fastest internet. Specifically, improving speed from 1 Mbps to 25 Mbps creates 2.32 times more value than improving from 25 Mbps to 100 Mbps. A similar non-linearity arises in terms of internet quality. We report large WTP discrepancies by income levels and employment status. The results corroborate that internet subsidies are most impactful if they first assist users with the poorest internet access. In this sense, subsidizing the right population could be as important as – if not more important than – raising funds to subsidize internet access.