In fact, this topic has been gradually discussed in China. For startups, spending on the basic cloud is still slightly affordable, but more and more PaaS and SaaS services have become a problem. When the cloud expenditure gradually exceeds the stage that the company's profits can bear, it will be found that the enterprise has been in a dilemma and has fallen into the quagmire of cloud computing.
This paper also puts forward some suggestions. Enterprises should consider the issue of cloud repatriation when going to the cloud, that is, leave a way out for themselves and list the expenditure of cloud computing as an important KPI assessment indicator. The CEO-CFO-CTO of the company should fully consider All-in cloud computing while keeping a clear head.
Of course, different enterprises have different goals, and cloud computing is an irresistible trend anyway, so the translator has a question. When cost gradually becomes the pain point in the era of cloud computing, is it just the inevitable pain in the era of computing or the weakness of cloud computing itself?
There is no doubt that cloud computing is one of the most important changes in the history of computing platforms. Although IT is still in its infancy, cloud computing has affected hundreds of billions of dollars of enterprise IT expenditure, and public cloud expenditure is still growing rapidly at a rate of more than 654.38+000 billion dollars per year. This transformation is driven by a very strong value concept-that is, infrastructure can be purchased and used immediately, which meets the needs of large-scale business development-and improves the efficiency of enterprise operation and economy. In addition, cloud computing can also help foster innovation, because companies can free up resources to focus on new products and growth without investing too much energy in IT facilities and technologies.
However, as the experience of cloud computing industry matures and we see a more comprehensive life cycle of cloud computing from the company's economic situation, it is increasingly obvious that although cloud computing can obviously realize its promise (cost saving) in the early stage of the company's development, with the expansion of the company's scale and the slowdown of its growth, the pressure on its profit margin may begin to exceed its income. Because this change happened in the late stage of the company's development, it is difficult to reverse, because it is the result of years of focusing on the development of new functions, rather than the optimization of infrastructure. Therefore, rewriting or major restructuring in order to significantly improve efficiency may take several years and is usually considered impossible.
Now, people are increasingly aware of the long-term cost impact of cloud computing. As the cost of cloud computing increases, the total revenue cost (COR) or the cost of goods sold (COGS) obviously begins to increase. Some companies have taken dramatic choices, such as "repatriating" most workloads (such as Dropbox), returning the workloads originally deployed on the cloud to the local area, or adopting hybrid clouds (such as CrowdStrike and Zscaler) in some cases. Those companies that have already done so have released obvious cost-saving reports in their financial reports: in 20 17, Dropbox detailed in its S- 1 file that due to the optimization and transformation of its infrastructure, it saved a huge sum of 75 million US dollars in the two years before listing, most of which needed to be transferred back from the public cloud.
However, considering the heavy workload of this behavior and the dominant and unique industry mainstream saying that "the cloud is great". Most companies find it difficult to justify moving workloads out of the cloud. That's true, but we need to consider the wider impact. ) because the calculation method will change when evaluating the market capital scale of potential losses-we propose it in this paper. Income growth (usually) slows down with the increase of scale, and short-term efficiency is increasingly becoming the key determinant of open market value. With the growth of scale (usually) slowing down, the recent efficiency has become an increasingly critical determinant of open market value. The excess cost of cloud computing reduces the profit rate and seriously affects the market value.
However, the significance of this article is not to discuss repatriation. This is a very complicated decision, which varies from company to company and has a wide range of influences. On the contrary, we must first understand the extent to which cloud computing depresses the market value, so as to help formulate the decision-making framework and manage the infrastructure when the company expands.
Our analysis highlights the value that can be gained through cloud optimization-whether through system design and implementation, re-architecture, third-party cloud efficiency solutions, or shifting workloads to dedicated hardware. This is a very counterintuitive assumption, which gives a popular saying about local deployment of cloud PK in this industry. However, it is obvious that apart from short-term savings, when you consider the impact on market value, large-scale companies can prove that almost any level of work can help keep cloud costs low.
When Dropbox started its infrastructure optimization plan in 20 16, they saved nearly $75 million in two years by transferring most of the workload from the public cloud to "lower-cost, customized infrastructure in co-located facilities" directly leased and operated by Dropbox, and. The gross profit margin of Dropbox increased from 33% in 20 15 to 67% in 20 17. They pointed out that this is "mainly due to our infrastructure optimization and income growth. ("This is mainly because our infrastructure has been optimized, and … our income has increased during this period. ")
But this is just Dropbox. Therefore, in order to spread the potential savings of cloud repatriation to a wider range of companies, Thomas Dullien, a former Google engineer and co-founder of cloud computing optimization company Optimyze, said that it is estimated that 654.38 billion US dollars of public cloud expenditure can be repatriated every year, which is equivalent to about half of the total cost of ownership (TCO), including server racks and real estate.
The exact savings obviously vary, but several experts we interviewed agree with this "formula": the repatriation result is one-third to one-half of the cost of running the same workload in the cloud. In addition, the engineering director of a large consumer Internet company found that the price of the public cloud list may be 10 to 12 times the cost of operating its own data center. Discounts driven by usage commitment and quantity are common in the industry, and this multiple can be reduced to single digits, because cloud computing usually drops by 30-50% when it is committed to use. However, the operating cost of AWS is still about 30% of these discounts, and the active R&D budget means that potential companies will save more costs due to the return of funds. The performance improvement brought by managing your own hardware may bring more benefits.
In all our conversations with different practitioners, this model is very consistent: if you operate on a large scale, the cost of the cloud can be at least twice that of your infrastructure.
Considering the percentage of cloud computing in the total revenue cost (COR), the 50% savings brought by cloud computing are very meaningful. Based on the benchmark test of public software companies (those companies that disclose their cloud infrastructure expenditures), we find that contract expenditures account for 50% of COR on average.
The actual expenditure as a percentage of COR is usually even higher than the promised expenditure: a private software company worth $65.438+0 billion told us that their public cloud expenditure accounts for 865.438+0% of COR, and "cloud expenditure accounts for 75% to 80% of revenue cost, which is very common in software companies". Du Lin (noticed from his work at Google and now Optimyze), companies tend to be conservative in estimating the scale of cloud computing, because they are afraid of spending too much, so they only promise the baseline load. Therefore, according to the rule of thumb, the promised expenditure is usually 20% lower than the actual expenditure ..... The elasticity is two-way. Some companies we interviewed reported that their cloud spending exceeded the promised forecast by at least 2 times.
If we extend these benchmarks to a wider range of software companies that use public clouds for infrastructure construction, we roughly estimate that the cloud bills of the top 50 listed software companies total $8 billion (which shows a certain degree of cloud expenditure in their annual documents). Although some of these companies have adopted a hybrid cloud approach, public and local deployment (which means that the proportion of cloud expenditure in COR may be lower than our benchmark), our analysis balances this point, assuming that the committed expenditure is equal to the overall actual expenditure. According to our conversation with experts, we assume that cloud repatriation can reduce cloud spending by 50%, thus saving $4 billion in recycling profits. For large public software and consumer Internet companies that use cloud infrastructure, this number may be much higher.
Although the estimated net savings of $4 billion are amazing in itself, if converted into unlocked market value, this figure is even more staggering. Since all companies are conceptually valued at the present value of their future cash flows, the total net savings in these years can create a market value far exceeding $4 billion.
How much more? A rough indicator is to look at the valuation of the additional gross profit in the open market: high-growth software companies that are still burning money are often valued according to the gross profit multiple, which reflects the assumption of the company's long-term growth and profit rate structure. (The income multiple usually refers to also reflects the company's long-term profit rate, which is why even after the growth rate adjustment, the income multiple of the business with higher gross profit margin tends to increase). But these two capitalization multiples can be used as a heuristic method to estimate the market discount of the company's future cash flow.
Among the 50 listed software companies we analyzed, the average multiple of the total enterprise value and the gross profit of 202 1 year (based on the data when CapIQ was released) was 24-25 times. In other words: for every dollar of gross profit saved, the market value will increase by an average of 24-25 times the net cost. (Assume that the saved capital is the depreciation cost after deducting the increased capital expenditure (if relevant)).
In other words, only these 50 companies, it is estimated that the extra gross profit of $4 billion will generate an extra market value of $654.38+000 billion. In addition, because the gross profit multiple (relative to the free cash flow multiple) is used to assume that the increase of gross profit is also related to the increase of some operating expenses, this method may underestimate the impact of the annual net savings of $4 billion on the market value.
For a specific company, its impact may be higher, depending on its specific valuation. To illustrate this phenomenon, we take the infrastructure monitoring of service company Datadog as an example. When the company released its financial report, the transaction price was close to 40 times of the estimated gross profit in 20021year. In its S- 1, it was disclosed that the total consumption commitment of Amazon cloud service for three years was $225 million. If our annual commitment to Amazon's cloud services reaches $75 million-assuming that 50% or $37.5 million of it can be recovered through cloud repatriation-this means that the company will gain a market value of about $654.38+0.5 billion every year just by the promised expenditure reduction!
Although these rough analyses are never perfect, the directional trend is clear: the market value of large listed software companies is dragged down by cloud computing and hundreds of billions of dollars in costs. If it is extended to a wider range of enterprise software and consumer Internet companies, this figure may exceed $500 billion-assuming that 50% of the total cloud expenditure is consumed by large technology companies, these companies are expected to benefit from cloud repatriation.
For business leaders, industry analysts and builders, it is too expensive to ignore the impact on market value when making long-term or even short-term infrastructure construction decisions.
What should we do next? On the one hand, starting to remove workloads from the cloud is a major decision. For those who don't plan ahead, the necessary rewriting seems impractical, so it is impossible. Any such work requires a strong infrastructure team, which may not be fully established. All this requires the establishment of a professional knowledge and ability beyond its own core system, which will not only distract attention, but also slow down the growth rate. In addition, cloud computing retains many advantages, such as the ability to respond on demand and a large number of existing services to support innovative projects and new geographical areas.
On the other hand, we outlined a phenomenon in this article, that is, the cost of cloud computing "gained the upper hand" at a certain moment, locking in the market value of hundreds of billions of dollars, and these market values are now caught in such a paradox: if you haven't started using cloud computing, you are crazy; If you continue to use cloud computing, you are crazy (if you don't start from the cloud, you are crazy; If you stay up there, you are crazy. ) 。
So how can enterprises get rid of this paradox? As mentioned earlier, we did not provide any reasons for cloud repatriation. On the contrary, we pointed out that infrastructure expenditure should be a primary measure. What does this mean? Companies need to optimize as early as possible, often and sometimes even outside the cloud.
Although there is still a lot to say about the change of thinking mode and best practices, especially recently, there are some factors that may help enterprises cope with the rising cost of cloud computing.
Take cloud expenditure as KPI. The purpose of making infrastructure a part of the first-class measurement standard is to ensure that it becomes the KPI index of the enterprise. Take Spotify's Cost Insights as an example, which is a self-developed tool to track cloud spending. By tracking cloud spending, the company's engineers can get all the permissions of cloud spending, not just the patents of the financial team. Ben Schaechter used to work in Digital Ocean and is now the co-founder and CEO of Vantage. He observed that they not only saw companies in the whole industry put cloud cost indicators together with core performance and reliability indicators in the early stage of the business life cycle, but also, "developers who were accidentally troubled by cloud computing bills became more and more savvy and expected their teams to be more strict in cloud computing spending. 」
Motivate the right behavior. Empowering engineers with KPI data of infrastructure can help the team raise awareness, but it will not consider changing the way of working. A well-known industry chief technology officer told us that in one of his companies, they provide short-term incentives similar to sales (SPIFFs), so that any engineer who saves a certain amount of cloud expenses by optimizing or shutting down the workload will get a cash (although the saved funds continue to appear, the company's return on investment is still high). He added that this method-basically "connecting people who create problems with people who can solve them"-is actually cheaper because it can bring a return of 10% to the whole company and reduce the total expenditure of $3 million in just six months. It is worth noting that the CFO of the company is the key figure to support this non-traditional model.
Optimization, optimization, optimization. When evaluating the value of any enterprise, the most important factor is the cost of selling goods, or the cost of selling goods-what is the distribution cost of every yuan earned by an enterprise? Segment, a customer data platform company, recently shared how they reduced infrastructure costs by 30% (traffic increased by 25% in the same period) by incrementally optimizing infrastructure decisions. There are many third-party optimization tools that can provide rapid gains for existing systems. According to our experience, the range is 10-40%.
Consider early repatriation. Cloud is cheaper and better in the early stage of the company's development, and more expensive in the later stage. The existence of this cloud paradox does not mean that companies must passively accept and not make plans. Make sure your system architect is aware of the possibility of repatriation as soon as possible, because by the time the cloud cost starts to catch up with or even exceed the revenue growth, it will be too late. In the early stage, even moderate or more modular architecture investment (including architecture that can move the workload to the best location without being locked) will reduce the work required to migrate the workload in the future. The popularity of Kubernetes and the containerization of software that makes workloads more portable are partly a response to the fact that companies don't want to be locked into a specific cloud environment.
Incremental repatriation. There is no reason why you can't repatriate in an incremental way and in a hybrid cloud way (if this is the right choice for your business). Here, we need more nuances than wishful thinking: for example, repatriation may only make sense for a subset of resource-intensive workloads. You don't have to be all on the cloud or all repatriated. In fact, in many companies we interviewed, even the most radical companies to restore workload still have 10% to 30% or more in cloud computing.
Although these suggestions are mainly aimed at SaaS companies, people can do other things. For example, if you are an infrastructure provider, you may want to consider the option of shifting costs-such as using customers' cloud points-so that the costs will not appear in your account. The entire ecosystem needs to consider the cost of cloud computing.
* * *
It is easy to understand how the industry got to this point: the cloud is the perfect platform to optimize innovation, agility and growth. In an industry driven by private capital, profit rate is often a secondary issue. This is why new projects tend to start in the cloud, because companies give priority to the speed of functional development rather than efficiency.
But now, we know. The long-term impact has not been well understood-which is ironic, because more than 60% companies regard cost saving as the primary reason for moving to cloud computing! For a new startup or a new project, the cloud is an obvious choice. Of course, it is worthwhile to pay even a moderate "flexibility tax" for the flexibility provided by the cloud.
The problem is that large companies-including those that reached a certain scale when they started their businesses-have been trapped in the cloud (and their ability to save themselves after being trapped in a quagmire). Interestingly, one of the most frequently mentioned reasons for the early mobile cloud-a large amount of capital expenditure in advance (CAPEX)-no longer needs to be repatriated. In the past few years, alternatives to public cloud infrastructure have developed significantly. These infrastructures can be built, deployed and managed entirely through operating expenses (OpEx) rather than capital expenditures.
Please note that although some of the figures we share here seem large, our assumptions are actually conservative. The actual expenditure is often higher than the promised expenditure, and we have not considered the flexible pricing based on time. The actual drag on the market value of the whole industry may be much higher than expected.
Will the 30% profit currently enjoyed by cloud service providers eventually change the seriousness of the problem through competition? Unlikely, because most of the current cloud computing expenditures are monopolized by three companies. There is a dramatic irony here: Amazon, Google and Microsoft, which represent a huge market of $5 trillion in this industry, have been hit by competition. Their high profit margin is driven by operating their own infrastructure to some extent, which enables them to reinvest more products and talents while supporting the stock price.
Therefore, under the balance of hundreds of billions of dollars, this paradox may be solved in one way or another: public clouds either start to give up profits or start to give up workloads. In any case, perhaps the biggest opportunity in infrastructure at present lies between cloud hardware and unoptimized code running on cloud hardware.