One thing to remember with the benchmarks is that when evaluating a cloud provider, the benchmarks are useless if talked about with absolute numbers or with regards to hardware specs. Because they're so horizontally scalable, it's more a question of average cost per compute operation or cost per (peta) byte.
Both AWS and Google clouds are perfectly capable of running Twitter, and running any software that Twitter uses, even if the number of machines is different. The only "benchmark" applicable is actually the total negotiated cost of the machines required to get the job done.
So it's not about being able to do things on Google cloud with fewer processors or less RAM or faster hard drives - Google was willing to give them a lower total cost of ownership for reasons known only to Google.
Do not assume that Google (or anybody else) will give you a similar preferential deal. Ignore the "benchmarks".
I don't think it's purely a measure of cost per compute operation. That assumes you would run the same software on any cloud using plain VMs or a service that slightly abstracts them. That may be true in this case since Twitter will continue to use Hadoop/Spark, but sometimes you can get a real advantage from switching to a service that only one cloud provider offers. For instance, someone in these comments pointed out that Twitter already migrated their anayltics workload to BigQuery. Evaluating BigQuery vs Redshift vs something else is not as clear cut as cost per compute operation.
If you want to check which cloud is best for you, run your application on all of them, measure your average+median+p95 cost per whatever (tweet/post/reader/dollar of revenue/ticket/?). Then factor in which platform is easier to work with, has better tooling and community support, because until you hit a certain scale your time will always be more expensive.
After you do all that, just start with Heroku and Postgres with some AWS Lambda, then move to ALB + AWS Fargate + RDS|Aurora / DynamoDB as you get bigger, then to NLB + ECS with a cluster of 20-80 On Demand and Spot Instances, then to a cluster of ARM instances on Spot.
If you need to build your own datacenter at that point, you'll know. And you would have built a 12Factor app to make all of the above work, so migrating will be easy.
Both AWS and Google clouds are perfectly capable of running Twitter, and running any software that Twitter uses, even if the number of machines is different. The only "benchmark" applicable is actually the total negotiated cost of the machines required to get the job done.
So it's not about being able to do things on Google cloud with fewer processors or less RAM or faster hard drives - Google was willing to give them a lower total cost of ownership for reasons known only to Google.
Do not assume that Google (or anybody else) will give you a similar preferential deal. Ignore the "benchmarks".