its not the size but what you do with it
having worked at length for the two of the hyperscalers, I have two cents to chuck in.
First, these public clouds are stitched together with LOTS of duct tape. One bomb dropped in Virginia or Dublin (or other catastrophic events like non-optimised bad code emanating from AI) will take out the digital world as we know it. In my case, I figure about 80% of the apps (including bank and gov apps) will simply cease working. We, as a collective, have placed too many of our eggs in the public cloud basket.
Two: if one really dives deep into what makes cloud tick, it is the management plane that lets one spin up VM's, containers, Serverless, really popular databases (looking at you Postgres, not Oracle :-)) etc with security/networking options supported on a defined set of h/w. It isn't that hard for the open source tribe or an entrepreneurial vendor (ideally, a few vendors) to create management plane that will run on the certified hardware with appropriate virtualisation innovations. Enabling and keeping up with security is the hard part, the way I see it. Heck, there is a lot of money to be made in subscriptions for companies brave enough to can do that. Pity that OpenStack didn't quite catch on and become that plane but it definitely can be done.
It doesn't take $100B to compete with these hyperscalers, most of the spend is for capacity that is absorbed by a few large providers (think Netflix, Cloudflare, Youtube etc). Enterprise and Gov only use a fraction of the capacity stood up/investment being made, so it is a much smaller spend for sovereign Enterprise cloud.
However, this is not true for GenAI workloads where one requires lots of GPU investment that these hyperscalers are making. but that business model continues to mystify me. Google recently breathlessly announced that their infrastructure served 500 trillion tokens a month. If one applies the 70/30 rule, where 30% of the tokens are paid for, say at an average price of $10 per million tokens, that barely works out to revenues of $20b a year for an investment of $80B a year in GenAI infrastructure that most of these hyper-scalers seem to average. We know that with most personal plans, no one is paying anywhere close to that, so it seems to be a loss leader for everyone. The only one making out like a bandit is Jensen Huang and crew, having gone from somewhere around $22B revenue in 2023 to $150B+ in annual revenues now.
to (mis)quote the bard: Verily, I hath witnessed a spectacle so witless, mine very soul hath packed its belongings and fled mine body. T'is as though reason itself hath tripped over a turnip and perished.
Biting the hand that feeds IT
