I get the impression that people mix up things that are logically distinct, even if they might overlap. For instance, one reader says criticizes the failure to use big data at the Fed   (see this post for a rejoinder) and in a closely adjoining comment, contrasts the GDPNow series run by the Atlanta Fed and the absence of a global counterpart at the Board. Let’s define some terms, so people who are inclined toward confusion become less confused.
Nowcasting: Reichlin et al. (2010) “… define nowcasting as the prediction of the present, the very near future and the very recent past. Crucial in this process is to use timely monthly information in order to nowcast key economic variables, such as e.g. GDP, that are typically collected at low frequency and published with long delays.” For instance, the Atlanta Fed’s GDPNow uses monthly data to nowcast GDP, which comes in at a quarterly frequency. If you look at the input series, you’ll notice that they are pretty much from conventional releases from the US government, Fed, or surveys like those from the ISM. I would not characterize GDPNow as a use of big data (although some might disagree). For an application, see here.
If you’re going to purport to know something about nowcasting, you might want to know something about MIDAS (application in an Econbrowser post here).
Big Data: “Big Data refers to data sets of much larger size, higher frequency, and often more personalized information.” This definition is going to encompass an enormous set of types of analyses, so any specific list will be arbitrary. However, one can get a feeling for what counts, by referring to instances cited on Econbrowser: OECD Weekly Tracker (using GoogleTrends), using GoogleTrends, pandemic impacts (using Google mobility trends), Delta variant (Google, Apple), and for Asia over the pandemic. For the 15 months of the pandemic, the Dallas Fed maintained a “mobility and engagement index”, used in this post. Finally, an application of big data to China was in a guest post by Laurent Ferrara (via QuantCube); this was a nowcast of quarterly consumption, so was an application of both nowcasting, and big data.
Enormous data sets and/or lots of variables, you’ll need something to figure out which data to use and which not to use. Lasso is one of those things you need to understand if you’re going to go down this route. (To me the latest version of neural nets, but I’m sure there are subtleties I’m eliding here.)
If one thinks these are things not being considered for use at central banks, investment bank research departments, and international organizations, even if they are not being reported for public consumption, I think that person is confused.
(As for whether the Fed has considered using things like satellite data for forecasting / nowcasting as Mr. Kopits queried, see e.g., Hakkio and Nie (2018), Nie and Oksol (2018), Clark et al. (2020). Whether they maintain a series that’s updated over time, I can’t say (and they won’t circulate until it’s “ready for prime time”, even if they were to use it internally). I mean, let’s assume Mr. Kopits uses spreadsheets; is he willing to publicize all of them?