I am curious about the search performance for a database without indices.

Of course it depends on the queries and how you’re using the data. If the machine learning algorithms are reading all of the data, you don’t need indices. But if you want to subset the data, then having indices can be important.

There is a way to deal with large amounts of data, without indices. You can throw large computing clusters at the data to support search.

But here again, it depends on how large the data is. If you have N processors, but the data grows, in the end you have a fixed N divided an increasing S (for size). With large enough amounts of data, N can lose its significance.