In the last two years, the three leading public cloud service providers, AWS, Google and Microsoft did something that may compel us to reevaluate our assumptions about the business of cloud. From an IT service provider’s perspective, the argument for adopting a cloud delivery model is its significant cost structure (economic) scalability advantage over legacy IT managed service and hosting models. This tenet has sustained an almost unquestioned advocacy for public cloud as the ultimate IT service provider business model and the inevitable destination of enterprise and industrial IT workloads and applications.
So, why have many of the public cloud darlings decided to increase the useful life of their servers and other data center network equipment?
What is different? Why the change?
On Amazon’s Q4 2021 earnings, CFO Brian Olsavsky cited that the update to the useful life of their servers and network equipment was attributed to continued refinements in how their software runs on their hardware as the reason why the company further extended the useful life of its servers from 4 to 5 years and their network equipment from 5 to 6 years starting January 2022.
However, one could argue that the useful life of server and network equipment in a hyperscale data center would stay at the lower end of the 3-to-5-year range that IT organizations conventionally use to depreciate their data center assets. The average server utilization rate in a hyperscale cloud data center is around 65% versus 15% or lower for dedicated equipment running in enterprise data center environments.
“AWS’s infrastructure is 3.6 times more energy efficient than the median of U.S. enterprise data centers surveyed.” – 451 Research
A study by 451 Research commissioned by AWS showed that higher server utilization is a if not the key factor of the sustainability advantage that AWS’s cloud infrastructure has over enterprise data centers. According to the study, 61 percent of the advantage is due to more energy-efficient servers and higher server utilization.
Most of this sustainability benefit is attributed to higher asset utilization since the energy efficiency of a custom server will not be an order of magnitude better versus an OEM box. In other words, energy efficiency is not a matter of wear and tear. It’s more a matter of cost-efficient capacity and environment.
Higher utilization would suggest more wear and tear on the hardware running in cloud infrastructure versus comparable equipment on-prem or in an enterprise data center. This logic would apply whether the infrastructure is used for public cloud services, or servicing retail operations and corporate IT in the case of Amazon. Let’s not forget, cloud computing from the service provider’s perspective is all about sweating data center assets.
Moreover, the advancements in data center hardware and software technologies are not slowing. The rate of obsolescence of compute and network hardware that we can expect going forward will not likely changed materially. Moore’s Law continues to progress at the system level as novel methods of advance packaging and integration and hardware design innovations keep new generations of compute and networking moving faster and cheaper.
Could it be the silicon?
It’s no secret that the leading cloud service providers are investing in custom silicon. AWS introduced EC2 C7g instances based on their third-generation Graviton Arm-based processors to their customers in November 30 of 2021. While Amazon claims that the Graviton instance demonstrate 20 to 40 percent better performance per watt than its x86 counterparts, this should not materially extend the useful life of these servers as they will be ideally running an average utilization rate of 65% to maximize return on assets.
The introduction of DPUs (Data Processing Units) or what Intel dubs IPUs (Infrastructure Processing Units) might be a possible argument for prolonging the useful life of network equipment, but these technologies and system architectures will not necessarily reduce the stress on boxes if a core tenet of your business is to sweat your infrastructure. These system efficiencies would more likely reduce the rate of capital investment in new data center equipment, not extend useful life of data center assets.
Capacity and performance demands on the cloud infrastructure continues to increase and become at least as intense, not less intense. Furthermore, hardware and software running in hyperscale data centers are becoming increasingly purpose-built and application specific as many of the leading cloud players continue to diversify the kinds of workloads and software platforms their cloud platforms support.
It is more likely that improvements in the performance per watt of server and network equipment would reduce the incremental energy cost of a data center as it scales out its capacity, which would improve the cost of service, hence gross margin of the business.
It is important to note that Amazon is not the only “cloud company” extending the useful life of their servers and network equipment though they were the first of the big cloud players to do so in recent times. The technological and/operational innovation that is bringing about significant useful life improvements remains a mystery.
The impact on cloud economics
So, how do these accounting changes impact cloud economics? They don’t. It’s just accounting. Cloud economics are what they are and continue to evolve. With flattening (and in some cases declining) margins on an unadjusted basis, it does seem that cloud service providers have reached the limits of data center profit margin optimization despite more power-efficient compute and networking equipment.
From an accounting perspective the change in useful life makes a cloud business look more profitable when compared to the previous baseline for depreciation. Though the extension of useful life will have minor impact on the gross margin of a cloud business, they will have outsized bearing on operating margins where depreciation expense are accounted for.
To give you a feel for the scale of the benefit, Amazon saw a $2.7 billion benefit to Amazon’s operating profits in fiscal year 2020 (or 11.5% of the company’s annual operating income) while Microsoft and Alphabet benefited 2.7 billion USD and 2.6 billion USD respectively during their respective 2021 fiscal year.
The impact of these accounting changes will vary depending on the cloud business type and mix of services. From an operating margin perspective, IaaS and operational cloud infrastructure operating costs will look lower given the relatively capital-intensive cost structure of these businesses. Operating margins for SaaS and PaaS business models will tend to be less impacted since these business models are more software-oriented business.
From example, ServiceNow’s SaaS cost structure looks more similar to a licensed software business with average gross margins of 85 percent compared to the much lower gross margins of an IaaS-heavy managed IT infrastructure business such as Rackspace which clocks in at around 35 percent. This explains the paltry 100 basis point impact on ServiceNow’s gross margin versus the 11.5% impact on Amazon’s operating profit from the change in useful life of data center equipment.
If you are curious, PaaS businesses, such as Microsoft’s Azure, have gross margins that range in the 55 to 70 percent range. AWS is an interesting case. It is largely an IaaS business with a growing PaaS aspect as it pushes and expands its portfolio of database, AI, and IoT platform services, tools, and software. As such we see that these accounting changes benefit AWS’s operating margins more than Microsoft which has a comparatively lower mix of IaaS in its portfolio of cloud businesses.
In the end, these are not trivial sums. According to Amazon’s 2022 annual report, the recent changes to useful life of servers and network equipment will impact their 2022 operating income by an additional $3.1 billion in 2022 on top of benefits realized beginning in their fiscal year 2020. It will be interesting to see if the other cloud players follow suit.
These accounting changes should prompt enterprise technology leaders, IT and telecom service providers, and IT vendors to think differently about the cloud giants and the cloud computing narrative that has dominated the discourse of the future of computing for the last decade. Has cloud computing as we knew it and know it reached an inflection point?
Enterprise CIOs will want to revisit their workload migration decision models and cloud strategies based on the new financial benchmarks that cloud players are setting. They will want to work with their IT vendors to determine how they can squeeze additional useful life out of their existing and incoming compute and networking assets with advanced and emerging cloud-native software and data center technologies. CIOs will want to work with their CFOs to adjust budgeting assumptions for infrastructure spend based on a rethinking of depreciation schedules for server and network assets.
For telecom operators that are looking at the edge cloud opportunity will want to keep an eye out on this trend from a benchmarking perspective. The economics of “edge” are changing the economics of cloud as the concept of cloud computing moves outside of the hyper scale data center. These are economics that network operators are quite familiar with and arguably good at.
It will be important for operators to ensure they are accounting for their edge cloud cost structure on the same basis as the big cloud players who are competing (or coopeting) in the emerging edge cloud opportunity lest they lose the narrative. More importantly, they will want to consider the implications of the cloud-native modernization of their network and MEC infrastructures on their future competitiveness in the edge cloud.
Finally, it is important of the ICT industry and Wall Street analysts to be mindful of the impacts of these accounting changes on their view on the margin trends of cloud service providers as their maturity plateaus. Has the thesis changed? Is the old thesis still valid? After all, cloud is all about the economics of scalability. We should not lose sight of this because of changes in accounting practices.