Ask HN: How to achieve limited scalability with less cash burnout?

I have been working on an algorithm around text processing. As a POC usecase, I am looking at using news articles as input text to the algo. However, the effectiveness will be seen only if I can have large number of them (say around a million).

I would like to know the best way to get access to free / cheap cloud infrastructure with less cash burnout (CPU, disk, RAM) as the App will be intensive on memory and CPU when it processes 1 million news articles. The initial need may be around 2 4 core nodes with 8GB RAM each.

Alternatively, would it be more cost effective if I buy a custom made server(s) and keep it at home to validate initially and once I cross the Beta, and see some traction (usage, funding) , I start using the cloud.

Note that I am self funded.