Per the BigQuery docs, GCP charges $5 per TB of data read. Which works out to be $0.50 per query if said query reads 100 GB of data.
It feels expensive, but then again running and maintaining a big data platform is inherently very expensive. If you consider the fully loaded cost of a single big data engineer to maintain such a platform to be ~$250,000 (likely an underestimate if you want similar performance characteristics
to BigQuery [disclaimer: never used it myself, but I assume its performance is near-unbeatable]), that'd be ~500,000, 100 GB queries. Which makes a $0.50 query feel reasonable relatively speaking. GCP also sells dedicated "slots" (as they call it) which as I understand it is an abstraction over a CPU. If you buy said slots, the marginal cost of queries is $0, but you may be subject to queueing. No idea what a "slot" actually represents however.
A full scan rarely happens on BQ because of the nature of the columnar store. Try some public dataset like HN archive, and see how a query actually costs. You'll need very advanced (or stupid) query to read 100GB at once on BQ.
It's not for every use case, but Big Query is often a very stupidly cheap datastore. Query results get cached, and repeats don't incur a charge unless the data has changed.
It's not a datastore to power a crud app, or anything requiring frequent queries, but it's a great place to stash gobs of logs that you may need to query at some point. Or it's great for serverless batch workloads and is often cheaper in both time and money than firing up spark clusters or something similar to do the work.
Quite frankly, it's awesome. But sure, they do use it as a tool for lock-in, and for some cases it would be prohibitively expensive.
Incredible as in 'that's a great deal' or as in 'that seems a ripoff'?
I find 50c to read 100GB from disk, do useful work on it (including running javascript code or ML models if you are so inclined) and returning a result in seconds... pretty damn incredible.
A query reading about 100GB with one of the most advanced data warehouse systems with no operational overhead and integration into a major cloud environment costs $0.50.
The point of BQ is to allow you to perform queries which are ad-hoc and/or touch a significant fraction of the data. If you have a problem of that shape, then full column scans are not merely tolerable, they are optimal.
It feels expensive, but then again running and maintaining a big data platform is inherently very expensive. If you consider the fully loaded cost of a single big data engineer to maintain such a platform to be ~$250,000 (likely an underestimate if you want similar performance characteristics to BigQuery [disclaimer: never used it myself, but I assume its performance is near-unbeatable]), that'd be ~500,000, 100 GB queries. Which makes a $0.50 query feel reasonable relatively speaking. GCP also sells dedicated "slots" (as they call it) which as I understand it is an abstraction over a CPU. If you buy said slots, the marginal cost of queries is $0, but you may be subject to queueing. No idea what a "slot" actually represents however.
https://cloud.google.com/bigquery/pricing#on_demand_pricing