There is an easy exit strategy for Google (and Amazon), and Google has already done it for Gmail: put a reasonable cap on how much users are allowed to use (storage for Gmail, # of queries for Alexa) and charge for anything on top of that. As server-side hardware costs decrease, wait for that "billion dollar curse" to asymptotically approach zero over the course of a few years.
There’s probably a 1% of users who use more than the other 99% added up and quotas cut down on that.
But in spite of having thousands of email messages, Gmail wasn't taking up much room at all.
Is storage really that expensive at Google scale that this would constitute a real cost saving measure after engineering overhead of implementation and PR backlash?
Likely multiple stages downhill, from multiple site redundancy to single site but multiple hosts to multiple drives on a single host to single drive but still copied over if the drive reaches some scheduled EOL before of failure. Perhaps there's even a very last stage, available as long as that one drive is still powered up.
And when you consider that hard drives have stagnated in price over the past few years, YouTube is doomed if it does not delete unprofitable videos. Unless prices to host content goes down, the whole thing becomes a sort of ponzi.
The suggested idea, which Google might already do, is to keep the low quality videos in less redundant storage on the theory that they can be re-encoded from the more durably stored original video if the underlying storage fails.
Not necessarily, is storage(and not retrieval) that much of a cost? and if so is youtube's revenue/storage cost actually decreasing? and if it is would the backlash for deleting people's videos worth the reputational cost? especially when you have an entire industry around small people trying to get big, the majority of which will never get there.
Youtube even tried putting a limitation on non-premium users watching 4k video and pulled back on it, can't imagine something as drastic as deleting videos would be worth it.
I think the article has a point, but is overselling it as bigger than it actually is.
I've found this varies. I've had to make a few new gmail accounts lately for various reasons, and only once was forced to add a phone number.
Koomey's Law says compute costs will be ~1/16th / ~7% in a decade.
IIUC, storage prices & bandwidth reduce at slower rates.
"Approaching 0" and "a few years" seem a little hyperbolic.
Probably worse, Koomey's law is an observation based on the effects of Dennard scaling (constant power density while doubling transistor count due to scaling), and this pretty much ended over a decade ago due to no longer being possible to scale transistors in a uniform way.
Power improvements will continue due to other one off process improvements, and due to other factors such as architectural improvements or moving more things into hardware, but the trend is not going to be as pronounced or consistent as it was before.
Compare top CPUs from 2012 and current top CPUs
I've opened cpubenchmark and compared those CPUs:
AMD Ryzen Threadripper PRO 5995WX
and
Intel Core i7-3970X @ 3.50GHz (I didnt know how to find something better)
First one has 98k points, second one has 8k points.
It's giant gap.
even pretty normal (consumer) CPU like Intel Core i9-13900K has 60k~ points.
I did not say transistor count, I said power density due to transistor scaling. There is a difference, transistor counts have continued to increase since 10 years ago for other reasons, but it no longer comes with the implicit power density and switching speed improvements of dennard scaling underlying the original moors law projection. It also doesn't have the same road ahead of it since transistors are no longer uniformly scaling, and dies can't grow exponentially.
> First one has 98k points, second one has 8k points.
Why do you think that is?
What are "points", how have CPUs changed since 2012? and How did 2012 CPUs change since 2002? i.e what are the qualitative differences between these periods that are made opaque by "points", and which are key in indicating improvement projections.