Hacker News
points by jgalt212 14 hours ago | 0 comments
> GCP has been an enormously successful 'bet' for Google over the past decade.

I disagree on this. It's not clear how much money they are actually making and how much revenues are shifted over from other business lines to GCP because GCP earnings growth is / was valued higher than other business lines by the markets.

[–] whakim 13 hours ago
I mean, $11B in quarterly operating income ($33B in revenue) is clearly quite a lot. The only reasonable argument here is that this includes hardware sales (i.e. TPUs), but it is still a lot even if you subtract that. I think many folks internalized the narrative that GCP was an also-ran many years ago (when that was demonstrably true), but haven't updated their priors now that revenue is almost the same as (and may soon surpass) Azure's.
[–] HDThoreaun 12 hours ago
Google has spent $132 billion on capex this year. Most of that money is going to GCP. We have to think about the hundreds of billions being invested to contextualize the profit numbers. $40 bill a year is not enough to justify the capex they’re partaking in, GCP still needs to grow a lot to be successful.
[–] whakim 9 hours ago
I don't think it's correct to view AI capex as simply booking profit for Google Cloud - theoretically almost every aspect of Google's business should indirectly benefit from those investments. Bigger picture though, the bet behind that staggering capex is future profits, not current profits (see: OpenAI, Anthropic). Even if you think that's a stupid bet, the evidence won't be in 2026 income, it'll be in future income.
[–] motionlessveloc 10 hours ago
The vast majority of that money is going to AI. Some part of it is for internal consumption, some of it is for resale to GCP customers, but it is not some sort of loss making subsidy to prop up GCP.
[–] HDThoreaun 9 hours ago
All the AI revenue is booked under GCP. This isnt any different from people saying openai and anthropic are selling a dollar for 50 cents
[–] spwa4 13 hours ago
That's the big criticism on Microsoft, not Google. On GCP this isn't the case, because Google does not have a huge install base of MS Office to give "free" 50% Azure credits (but paying the same) faking their cloud numbers.

GCP has a reputation on being the fastest compute and network available. Also the most bang-for-your-buck (on hyperscalers) ML infrastructure by far (because TPUs)

Amazon is "the default" (and most expensive). Works with everything.

Azure is "Microsoft cloud". It's not fast, cheaper but certainly not cheap, ...

But of course that should be interpreted as GCP being 10x the cost of on-prem hardware (for equivalent storage/performance). Azure 13x or so. And Amazon a solid 20x (mostly because none of their products deliver the performance advertised and because if you buy 1 thing of Amazon, they bill you for 20 different things). Even with the price hikes a server under your desk (not the same, I know, however, once you have 20 servers in 2 rooms in 2 buildings and kubernetes running the workloads it is the same in reliability)