The math still doesn't add up...
200 inverters X 25 panels each X 3 analog inputs per panel = 15,000 inputs. *confused*
If the purpose of collecting the data is view-only live data and/or reporting, you could possibly just dump the readings into a SQL database and retrieve what you need to display using a simple web-based viewing system. This may help reduce the costs and increase flexibility of display and reporting. Also, you could divide the system easily into sub-sections, with each section sending its own data to the database independently of the others. Database queries could be then used to generate the live data and reporting per customer requirements.
Data storage and archival must be well designed for a dataset of this magnitude, in order to achieve reasonable performance and desired archival length.
If you are capturing 15,000 analog data points and dumping them to a database, you would probably need about 20 bytes per database row (2 bytes for analog data, unique ID (bigint): 8 bytes, timestamp: 8 bytes, input type indicator: 2 bytes)
For 15,000 samples, that's 300kB. If you sample once per minute, that's ~13GB per month, and 648 million data rows.
If you are actually capturing 1 million analog data points, you would be generating 20MB of data per minute, or 864GB per month. (43.2 billion database records.)
SQL Server's table length (# of rows) is limited only by disk size, but I'm sure at those numbers it would take some beefy hardware to retrieve any kind of useful subset for reporting purposes. Distributing the dataset across multiple servers may be more viable.
Good luck!