A decade ago I worked on building AI systems that made (legitimate, paid) use of the Twitter “firehose”. At that time more than 99% of the data was garbage. It’s worse now. The value then was largely in two areas: historical trends (something like Google trends), and breaking news; and only the latter was really that interesting. I doubt it’s a high value data source scraped in bulk; it could have value in a much more targeted approach. Seems unlikely to require the addition of “large numbers of servers … on an emergency basis”.