A fantastic write up in ArsTechnica about how AI hasn't been profitable for Big Tech. A WSJ article cited says Microsoft loses around $20 per user per month on GitHub Copilot. In fact, the service has operated at a loss despite having more than 1.5 million users. The $10 a month flat fee doesn't cover the average user cost of $20; worse still some power users can consume as much at $80 a month.
It's a huge loss leader for Microsoft who is jockeying with other tech companies for market share. The priority, at this point of the cycle, is to prove AI's value and to integrate it into as many workflows as possible. The bet is that the service will become sticky and increase switching costs once customers buy into the platform.
It's a huge gamble that has paid off in the past. Much like cloud, however, it is inevitable that prices will go up at some point. If you are planning to incorporate AI or ML For a product, consider the long-term costs carefully.
As Gergely Orosz points out in Pragmatic Engineer newsletter, cost estimates are difficult for AI powered features without actual usage numbers. If you build a brand-new app this will be particularly challenging. He rightly points out that costs are going down but not quick enough that early AI products won't burn a lot of cash. In the meantime, everyone is competing for market share among developers and coding tools.
It's a huge loss leader for Microsoft who is jockeying with other tech companies for market share. The priority, at this point of the cycle, is to prove AI's value and to integrate it into as many workflows as possible. The bet is that the service will become sticky and increase switching costs once customers buy into the platform.
It's a huge gamble that has paid off in the past. Much like cloud, however, it is inevitable that prices will go up at some point. If you are planning to incorporate AI or ML For a product, consider the long-term costs carefully.
As Gergely Orosz points out in Pragmatic Engineer newsletter, cost estimates are difficult for AI powered features without actual usage numbers. If you build a brand-new app this will be particularly challenging. He rightly points out that costs are going down but not quick enough that early AI products won't burn a lot of cash. In the meantime, everyone is competing for market share among developers and coding tools.