[nostrademons]

Your observation is true, your conclusion isn't necessarily. AI is different from other areas of computing in that data matters far more than code does. Current legal practice is that the data is owned by the organization that collected it, which means that to collect data on millions of users to train behavioral models, you need to have millions of users. If you have millions of users, you're probably a pretty big company.

[mac01021]

That depends to a large extent on what skill you want your AI to be able to learn.

Many applications from recommender systems, to market analysis to drug discovery probably require large datasets of the sort that can only be collected by a large business.

But if you want to learn control polices for robotic agents that navigate the physical world, driving cars, stocking shelves, etc. then I think the physical world is presenting everyone with approximately the same data set regardless of company size.

[seizethecheese]

My conclusion is that it's feasible, so by definition not necessarily true. An AI could be bootstrapped by scraping data, or using any number of large datasets out there. This isn't to say that large corporation don't have advantages.