I'm not entirely sure what sort of courses you are planning to offer, but what I can tell you is that course material quality at traditional higher education providers (bottom tier) is appalling (and they know it). If you can get it right, that would be worth a lot more than $1m/year. At the moment, the most obvious bottleneck would be the relatively small context window of GPT models (<3k words). It is speculated that OpenAI plans to increase it to 24k words, which would make it a lot more feasible though.