One parallel that's unclear to me is with vendor lock-in. Over the last decade, a positioning argument shared by data analytics and infrastructure-adjacent SaaS startups has been that customers prefer third-party systems so they don't have to lock into a single cloud vendor. Will this parallel be true with foundational AI companies? My hunch is that this isn't as valuable as a selling point for AI startups since the marginal cost of switching foundational cloud platforms is much lower.
It's a good question. Most startups that I'm paying attention to today aren't yet trying to bring in primary data storage — that's still with the SaaS vendors. But you can imagine that "I spent 12 months improving the quality of this system." will be pretty compelling a while down the road? I don't think most companies are doing any true learning yet, but they probably will be soon.
Folks at leading labs say “the model is the product,” which, when you look at it, is pretty dumb
Totally agreed.
One parallel that's unclear to me is with vendor lock-in. Over the last decade, a positioning argument shared by data analytics and infrastructure-adjacent SaaS startups has been that customers prefer third-party systems so they don't have to lock into a single cloud vendor. Will this parallel be true with foundational AI companies? My hunch is that this isn't as valuable as a selling point for AI startups since the marginal cost of switching foundational cloud platforms is much lower.
It's a good question. Most startups that I'm paying attention to today aren't yet trying to bring in primary data storage — that's still with the SaaS vendors. But you can imagine that "I spent 12 months improving the quality of this system." will be pretty compelling a while down the road? I don't think most companies are doing any true learning yet, but they probably will be soon.