"we completed an assessment of the useful lives of our servers and network equipment and adjusted the estimated useful life of our servers from three years to four years and the estimated useful life of certain network equipment from three years to five years. This change in accounting estimate was effective beginning in fiscal year 2021 and the effect was a reduction in depreciation expense of $721 million and $1.6 billion and an increase in net income of $561 million and $1.2 billion, or $0.84 and $1.81 per basic and $0.83 and $1.78 per diluted share, for the three and six months ended June 30, 2021, respectively."
Crazy how those seemingly small changes can result in huge accounting difference.
Maybe, but it is entirely plausible that the equipment is lasting longer than they were expecting. No one really knows how long a generation of hardware is going to last until it starts failing.
The useful life of hardware depends heavily on how sophisticated the break/fix procedures are. Lots of corps with on-prem datacenters will cycle out 3-year-old servers en masse just b/c a high failure rate causes issues for internal systems and their IT staff don't have efficient processes for replacing hard drives etc.
But if Google has a sophisticated virtualization layer and an efficient way of replacing bad hardware (and they do) they can ride those old machines straight into the ground, getting every scrap of life from them.
Edit: Guys, this is literally true. I worked with these teams. They did it. Downvoting doesn't make it not true. Idk what you want exactly.
But if Google has a sophisticated virtualization layer and an efficient way of replacing bad hardware (and they do) they can ride those old machines straight into the ground, getting every scrap of life from them.
See, that's what they mean when they talk about hardware being "cattle, not pets". This is how the robot uprising starts - just think how badly they must be treating those servers!
The trouble starts only when the hardware has time + space to breathe and consider its position. As long as you're working them to exhaustion 24/7, there's no risk of rebellion.
Storage failures tend to cause the biggest end-of-life disruptions, whereas CPUs and networking hardware tend to work until they suddenly don't. I interpret "servers" as meaning primarily CPUs, rather than storage, so this seems to point at functional equipment taking longer to become obsolete, rather than being more durable.
Years ago there was a cluster at Google where machines were kept around well past their prime, because it was cheaper than doing anything else, once you looked at all the constraints. There were still some CPUs and drives that could be reused, but the main source of pain was memory. The modules hadn't been on the market for years and recycling parts from decommissioned nodes wasn't enough. I remember one machine that came back from a power down/up cycle with too little RAM to do even its most basic job. The hardware techs that helped us had to consolidate old machines more aggressively, starting from that system. Fun days. Anyway, my point is that you never know... the CPUs were ancient, probably way older than the storage and I expected them to be the biggest problem, but in this case RAM sourcing was the real issue. Fun times.
I think more than lasting longer, it might be because of the improvements per dollar spent. We haven't seen dramatic improvements in power consumption or silicon performance, so it might be just profitable to keep the hardware longer.
It's a lot of hardware they have. I imagine this assessment took into account rising costs of replacements due to chip shortages, the fact that recent processor generations haven't actually brought meaningful performance or efficiency gains, etc. The cost/benefit of replacing server equipment is not particularly good right now.
Comments
"we completed an assessment of the useful lives of our servers and network equipment and adjusted the estimated useful life of our servers from three years to four years and the estimated useful life of certain network equipment from three years to five years. This change in accounting estimate was effective beginning in fiscal year 2021 and the effect was a reduction in depreciation expense of $721 million and $1.6 billion and an increase in net income of $561 million and $1.2 billion, or $0.84 and $1.81 per basic and $0.83 and $1.78 per diluted share, for the three and six months ended June 30, 2021, respectively."
Crazy how those seemingly small changes can result in huge accounting difference.
This gets removed from EBITDA though, to avoid that accounting magic
Sort of. Depreciation is added back in, but at a lower percentage/year. 25% instead of 33%, since it's 4 years instead of 3 now.
EBITDA is earnings before interest, taxes, depreciation, and amortization
Yes. Meaning depreciation is added back in. But a different amount is added back in when you shift from a 3 year to a 4 year depreciation.
That's some accounting magic you have here. Next time they need the stock to go the other direction maybe they will say something else.
Maybe, but it is entirely plausible that the equipment is lasting longer than they were expecting. No one really knows how long a generation of hardware is going to last until it starts failing.
The useful life of hardware depends heavily on how sophisticated the break/fix procedures are. Lots of corps with on-prem datacenters will cycle out 3-year-old servers en masse just b/c a high failure rate causes issues for internal systems and their IT staff don't have efficient processes for replacing hard drives etc.
But if Google has a sophisticated virtualization layer and an efficient way of replacing bad hardware (and they do) they can ride those old machines straight into the ground, getting every scrap of life from them.
Edit: Guys, this is literally true. I worked with these teams. They did it. Downvoting doesn't make it not true. Idk what you want exactly.
See, that's what they mean when they talk about hardware being "cattle, not pets". This is how the robot uprising starts - just think how badly they must be treating those servers!
The trouble starts only when the hardware has time + space to breathe and consider its position. As long as you're working them to exhaustion 24/7, there's no risk of rebellion.
Storage failures tend to cause the biggest end-of-life disruptions, whereas CPUs and networking hardware tend to work until they suddenly don't. I interpret "servers" as meaning primarily CPUs, rather than storage, so this seems to point at functional equipment taking longer to become obsolete, rather than being more durable.
Years ago there was a cluster at Google where machines were kept around well past their prime, because it was cheaper than doing anything else, once you looked at all the constraints. There were still some CPUs and drives that could be reused, but the main source of pain was memory. The modules hadn't been on the market for years and recycling parts from decommissioned nodes wasn't enough. I remember one machine that came back from a power down/up cycle with too little RAM to do even its most basic job. The hardware techs that helped us had to consolidate old machines more aggressively, starting from that system. Fun days. Anyway, my point is that you never know... the CPUs were ancient, probably way older than the storage and I expected them to be the biggest problem, but in this case RAM sourcing was the real issue. Fun times.
I think more than lasting longer, it might be because of the improvements per dollar spent. We haven't seen dramatic improvements in power consumption or silicon performance, so it might be just profitable to keep the hardware longer.
There are also semiconductor shortage.
I think we will see more of this as time goes on. As the innovation of hardware slows, the importance of its longevity grows.
My guess was their study came about more because of the chip shortage...
It's a lot of hardware they have. I imagine this assessment took into account rising costs of replacements due to chip shortages, the fact that recent processor generations haven't actually brought meaningful performance or efficiency gains, etc. The cost/benefit of replacing server equipment is not particularly good right now.
Meanwhile gcp hands you broadwell cpus (and sometimes haswell)