Did they delete the article?

There is a difference between 50k transactions per second and 1000 transactions in real-world conditions. The number of transactions per second and the number of transactions that can be added to the block without problems, along with the average waiting time to ensure that the transaction is final, are numbers that differ.
The thing is that there will never be 50 000 real transactions!
The problem lies in validators' votes and RPC capacity, assuming 0.5 s and average voters efficiency you will get 30 000 votes and only 20 000 transactions, but that will be like something close to peak test environment.
As I mentioned for dot, the requirements for sol do start being absurd :
CPU 12 cores / 24 threads, or more
RAM 256GB or more
Internet service should be at least 1GBbit/s symmetric, commercial. 10GBit/s preferred
Actual transaction to vote ratio is around 20%, and in the current environment, more complex tx with more data will bring it probably down to 5-10%, it would even reach a point where more validators would trigger more clog than speed.