Google Cloud Platform has MASSIVE markups compared to equivalent service providers, either from "the big three" or smaller service operators. These smaller service operators have far less margin to work with, yet manage to not only break even but to turn a profit! Either GCP has some product that is a metaphorical black hole consuming all of their excess revenue, or Google generally is spending far too much money doing the same thing that everyone else is doing.
One of the things I've noticed GCP prides themselves on is uptime, and I'm wondering whether they're doing something like _doubling_ their costs to have seven nines of uptime for the entire platform, which practically doesn't matter for anyone other than a handful of clients (healthcare, government, military, etc)
They are in a huge growth phase. Any rate-based service in a hyper-growth mode is going to lose money as they make capital and personnel investments.
I used to run a pre-O365 large exchange shared service that topped off at ~200k users. When we were growing 50% year over year, we were “losing” a fortune depending on how to measure it. You want to do that, as making a profit today may mean losing the lifetime value of a long term customer.
I think Google has a good thing going. AWS is the incumbent, but has its own legacy to deal with. Microsoft dangles carrots for Azure, but the Microsoft sales culture will start beating customers over the head with the carrots.
> Microsoft dangles carrots for Azure, but the Microsoft sales culture will start beating customers over the head with the carrots.
I reckon a good deal of Oracle sales force now works at AWS. Thomas Kurian, GCP CEO, is ex-Oracle and has likely hired more in leadership positions. MSFT would do fine, if not better, with their in-house sales team.
> I reckon a good deal of Oracle sales force now works at AWS. Thomas Kurian, GCP CEO, is ex-Oracle and has likely hired more in leadership positions.
Sorry just want to clarify: did you mean to say "reckon a good deal of Oracle sales force now works at GCE", so a typo? Or are you saying that a lot of the Oracle sales staff now work at AWS (even though Kurian who is ex-Oracle works at GCE)?
Not critiquing, just a honest question since I couldn't tell if it was a typo or not.
No problem. I did indeed mean that per my estimates, AWS sales is likely bustling with ex-Oracles; while GCP may have them too, since the CEO himself was, by some accounts, #2 at Oracle.
that could be a choice in the accounting department not to separate R&D and operational expenses when accounting for software engineers. Many engineers bristle at such accounting measures.
"The increase in operating loss in both periods was driven by an increase in total expenses of $5,103 million from 2019 to 2020…Operating expenses increased primarily due to compensation expenses (including SBC), largely driven by an increase in headcount. Additionally, data center and other operating costs increased in both periods."
This may not be a bad financial investment if the effect is to centralize internal google infrastructure investments under the "cloud" banner e.g. If google requires a Tier-1 cloud service for their operations but cannot use an alternative cloud for strategic reasons.
It's possible that the figures have been shifted, bent, and twisted to report a loss on what could actually be their biggest profit, as you so question.
I think they may be investing a lot on future cloud infrastructure. they're pouring a lot in AI chip design which will dominate in 5+ years specially when cars become fully self-driving.
If they put things like TPU development and TPU advertising costs (Google Colab) into the Google Cloud budget I can easily see that being responsible for a notable part of this loss.
The fixed costs (chunky engineering & product salaries) are ~ the same no matter how much revenue you have. Just take a napkin and do the math thinking through each part of the cost structure v AWS.
Isn’t it safe to say that GCP is a perfect place for Google to take losses from a tax perspective? They can just keep building new DCs, upgrading gear, creating partnerships, and buying customers. I feel like they could make it profitable any time they like, and they certainly did last quarter.
That's an interesting perspective. Also, since Google runs several of its services on its own cloud platform [1][2], I wonder if they are just moving data center and similar CapEx costs into their GCP budget, thus removing those cost line items from the rest of Google's quarterly and annual earnings results.
> Ciauri quit his post at Salesforce in September 2019 after 10 years as executive veep president and GM EMEA, and set about trying to build a regional team to lead Google's Workplace and Cloud Platform business where it's playing catch-up with AWS and Microsoft.
…
> Google Cloud reported an operating loss of $5.6bn in 2020, compared to an operating loss of $4.645bn in the prior year. The company did not publish the regional breakdown. In its 10K SEC filing [PDF] for the annual results, Google states:
> "The increase in operating loss in both periods was driven by an increase in total expenses of $5,103 million from 2019 to 2020…Operating expenses increased primarily due to compensation expenses (including SBC), largely driven by an increase in headcount. Additionally, data center and other operating costs increased in both periods."
This is about a 21%, or $955mn, increase in operating expenses from 2019 to 2020 with Chris Ciauri in charge during most of that time.
> …total revenue at Google Cloud grew 46.4 per cent year-on-year to $13.059bn.
If that $955mn operating expense is responsible for the $13.059bn revenue increase then each $1 increase in operating expense, assuming a linear relation, is responsible for $13.67 in increased revenue.
GCP will do fine. I worked as a contractor at Google in 2013 and learned to love their internal infrastructure, and in many ways GCP is similar. I preferring GCP but most of my customers work on AWS, so Google does have an uphill climb.
(disclosure: i work on gcp)
i asked this question to one of the people in my orientation and he said that if you don't believe what they say, one real thing they're doing is hiring bunches of new people to work on gcp. also, gcp runs on the same underlying infrastructure as the rest of google, so were i to guess just keeping it running would be relatively cheap.
What I witnessed was a feast and famine cycle. Senior leaders were always ravenous for headcount. In good times they would get it, and allocate it to increasingly dubious ideas. In bad times they would be forced to lay off even more than they had hired. Everyone from the dubious ideas, plus some of the maintainers of business-critical stuff, would be let go. The remaining engineers would (rightly) complain that they were understaffed and bereft of vital institutional knowledge. As times got better the gaps would be backfilled. But the company never had the restraint to stop there, it would just go right back into the cycle.
Add to that a not-invented-here / not-invented-by-me problem. A platform with some warts probably could have been fixed by those who understood it. But those people were gone. So the team genuinely needed larger headcount than before the layoffs, in order to rewrite the thing.
If I had to root-cuase it, it's that senior leaders get no points for delivering outsized results with a small contingent of engineers; their prestige and leveling is all about the headcount growth occurring beneath them. So there's no pushback on headcount growth, until there is.
If a company is going to go bust in 6-12 months and the metrics wouldn't warant further investment. Then the senior leadership may opt to increase the burn rate to run out of money in 3-6 months with the hope that the extra headcount will solve the deathly problems coming up at the end of the year.
At the very least it can help drum up interest from investors based on "look at the new X we shipped, or look at how strong our team is"
Hiring in the US may be a big expense, but it's not a commitment to much beyond the WARN Act requirements.
Of course, there's the issue that when a company has lost the trust of internet cynics, there's not much that company could say to convince them of a commitment. maybe a public, long term contract with some organization of consequence... but you can always get out of those by paying fines or malicious compliance.
We're moving our software to the cloud, and the people in power had dismissed Google before discussions about which provider to go with even began, simply due to their reputation for killing off their products with short notice.
Probably a bit reactionary, but I'm not going to stick my neck out for Google on that one.
I heard that they're moving some of their core products onto GCP, like parts of YT. If this trends keep going up over the next 2~3 years, maybe you don't have to worry too much about GCP itself, though there still might be smaller deprecations.
While I personally believe the fears of GCP being deprecated are overblown, the reason you give isn't actually true. Internally, Google is all for deprecating products with a ton of usage if it's inconvenient to maintain. That's why all internal tools got a new UI in the last year or two, because they deprecated the UI framework while everyone still depended on it.
Most of the deprecated Google products have either alternatives or generate no direct revenues. GWT is just a library in the second category, replaceable with a few hundred SWE years. Unlike that, GCP and YT are making real revenues in the order of billions. This is what "core product" means.
Google Cloud Platform has MASSIVE markups compared to equivalent service providers, either from "the big three" or smaller service operators. These smaller service operators have far less margin to work with, yet manage to not only break even but to turn a profit! Either GCP has some product that is a metaphorical black hole consuming all of their excess revenue, or Google generally is spending far too much money doing the same thing that everyone else is doing.
One of the things I've noticed GCP prides themselves on is uptime, and I'm wondering whether they're doing something like _doubling_ their costs to have seven nines of uptime for the entire platform, which practically doesn't matter for anyone other than a handful of clients (healthcare, government, military, etc)