Page 2 of 3
Why Your Cloud Expenses Are Rising: Blame Cloud-flation
Everyone is talking about inflation lately. In hindsight, inflation was inevitable. Our civilization is just coming out of a global pandemic where our supply chain ground to a sudden halt, governments have printed an unprecedented amount of money, and we are now in the middle of a war that is interrupting the global flow of commodities. This all matters because energy is the most important input to all modern manufacturing processes. We have not yet absorbed the full impact of the Russian invasion from an economic perspective, and it will likely take months.
What we know so far is that as of April, the annual inflation rate in the U.S. stands at 8.3 percent, only down slightly from 8.5 percent in March. To put this into historical perspective, that is the highest inflation rate since January of 1982. Many readers of this article have never seen inflation of this magnitude.
Has inflation impacted cloud prices and availability of resources? Absolutely. Cloud-flation is a reality and the technology industry will need to learn to deal with the new realities of cloud computing economics.
Let’s take a look at a few practical examples of cloud-flation; from there we can examine the economic and political forces driving price changes.
Google Cloud Drops an Inflationary Bomb
Google just recently announced a significant (and sometimes alarming) increase in cloud storage costs -- between 25 to 50 percent. Costs for operations (API calls to the storage data plane) are increasing a staggering 100 to 400 percent, with the largest increases showing up in multi-region coldline storage.
Data transfer costs for object storage (egress charges) have gone from free to two cents per GB, even if data stays in the same continent. Two cents doesn’t sound like a lot, but to put it in perspective, a typical high-volume microservices architecture could be impacted by hundreds or even thousands of dollars per month. This change is so important that I will quote the entirety of the change from the announcement:
Reading data in a Cloud Storage bucket located in a multi-region from a Google Cloud service located in a region on the same continent will no longer be free; instead, such moves will be priced the same as general data moves between different locations on the same continent. [Emphasis added]
Customers have until October 1, 2022, to change their usage patterns before the new pricing kicks in.
Supporters of the change may argue that Google is moving to pricing that is in line with competitors. Alternatively, one may reason that it makes sense to charge for data transfer from the U.S. East Coast to the West Coast because that is a real cost Google must absorb. I will take a contrarian position: If Google were able to continue charging lower prices without impacting profitability and gross margins, offering a sustained discount over its competitors, it would do so to win market share. This is not a company that will miss earnings projections for one of its fastest growing divisions due to unexpected inflationary pressures. This is a forward-thinking enterprise that is getting ahead of the inflationary curve.
AWS Spot-Instance Price Drift
Amazon Web Services, the most dominant public cloud platform in the world, has a feature called spot instances. It allows customers to rent EC2 instances (computers) at a fraction of their list price.
Spot instances have an interesting pricing characteristic. They provide a heavy discount, sometimes 80 percent or more, but they also come with an important caveat. AWS reserves the right to terminate and reclaim these instances with just two minutes’ notice. Many cloud customers cannot deal with this chaos and simply stay away from spot instances, even though they offer an amazing value proposition.
To avoid these unwanted interruptions, a typical customer strategy is to leave the price “floating” at market rates so if demand increases, rather than reclaiming the spot instance, AWS can simply increase the price that customers pay, avoiding a service interruption.
Data from AWS itself shows that prices fluctuate and have risen significantly since the beginning of 2022. Why is this the case? We can surmise, based on the laws of supply and demand, that AWS is having a hard time replenishing supply while demand is still extremely strong.
Why is supply constrained? Massive supply chain interruption has created a semiconductor shortage, which has affected many areas of the economy including automotive production and raw computing power for data centers (aka cloud). Where do most chips come from? The supply chain is not as diversified as you may believe. (We will discuss supply chain issues in greater depth later because they have become a strategic point for geopolitics.)
SaaS Providers Feel the Pinch
Software-as-a-service (SaaS) providers are particularly focused on cost reduction and margin improvement in 2022. This makes sense; in an economic downturn, everyone will be looking for additional areas where they can improve profit margins. The area of SaaS likely most impacted is the freemium pricing model where providers give customers a basic service at no cost and charge for premium features. These providers have to be careful about rising cloud costs in relation to their overall gross margin. Indeed, my team went through several architectural iterations for our raw data and time-series storage and cross-zone microservice communication.
I believe we will see SaaS providers increase prices over the next 12 months, and, in fact, we’ve already started to see this play out with larger companies that have dominant market share. For example, Adobe just recently raised the price of Creative Cloud between 3.8 and 6.3 percent, depending on the end-user plan. We may also see price increases in the form of feature movement. Imagine that a provider has three plans (free, basic, enterprise). It would be easy for these companies to move a few features from free to basic, which is just a price increase in disguise.
Moore's Law and Cloud Pricing
Now that we anticipate inflation in cloud computing, let’s try to understand the larger macro trend that has been happening with computing power over many decades -- essentially since the advent of the microprocessor.
Most readers are familiar with Moore’s Law -- which is more an observation than a law -- first articulated by Gordon Moore in 1965. It says that the number of transistors in a dense integrated circuit (IC) doubles about every two years. If we double the number of transistors in an IC, we can extrapolate that computing power -- as measured in millions of instructions per second (MIPS) -- must also double. All things being equal, when we double MIPS, we lower the cost of computers, and that is what we have observed over time.
The density of modern computer chips is measured by the size of a single transistor in nanometers (nm) -- one-billionth of a meter. Our current modern transistor and chip manufacturing process is down to 5nm as of 2020, and there is hope it will shrink to 3nm by 2023. Historically, the semiconductor industry has made great progress in chip density. For context, back in 2001, the process stood at 130 nm.
To fight inflation, we must continue to see this trend evolve. Density has to improve and we must make computers more productive, more efficient, and less power hungry.
The Semiconductor Supply Chain
If we stop observing the doubling of transistor density every two years, we could see continued significant inflation. What are some of the potential reasons that could lead to this change?
For many years, the U.S. and other western nations have been far too complacent, farming out critical technology processes such as chip manufacturing to places like Taiwan. Factors such as labor costs and impact to local environments have made it more convenient and profitable for semiconductors to be made elsewhere.
What happens when the semiconductor supply chain stalls, as it did during COVID-19? Fewer servers are available in the cloud, and as a result, prices rise.
However, the risk is much greater than a temporary slowdown in semiconductor shipments. The problem is often depicted as consumers unable to get the latest tech-heavy new car, but this is just a short-term effect. The longer-term effect of supply chain concentration comes from our inability to innovate. To fight inflation, the semiconductor industry must continue to increase transistor density, and this only happens with serious R&D investments by industry leaders.
Why Taiwan Is Important
Taiwan Semiconductor Manufacturing Company Limited (TSM) is the world's largest semiconductor foundry, and perhaps the largest company in Taiwan. In 2021, that country enjoyed a more than 66 percent share of the global chip making market, a statistic that has likely only grown since. TSM plans $40 to $44 billion in capital expenditures in 2022, up from roughly $30 billion in 2021. CapEx will help TSM expand production capacity, which will help to ease supply constraints.
More important, TSM is one of the largest investors in the future of semiconductor manufacturing processes and technology. It is planning to commercialize the 3-nanometer chip this year and is ahead of most other manufacturers.
Here’s where geopolitics comes into play. Taiwan has become the crown jewel of the region and a major exposure for the U.S. and its western allies. China strongly believes that modern-day Taiwan is a part of the People's Republic of China and is becoming increasingly emboldened. Put simply, if China were to seize control of Taiwan today, China would control the single major source of innovation that drives computing power’s cost.
It’s important to point out that China has been able to observe the devastating series of global sanctions imposed on Russia because of its invasion of Ukraine, as well as the effects on trade and overall GDP. Although China has likely put the brakes on any type of move towards military action to capture Taiwan, we must stay vigilant. Depending on what China does, we could see a disruption to the continuous doubling of semiconductor density (i.e., Moore’s Law).
Energy as a Key Input Cost
Computers are power hungry. They need significant amounts of electricity to run and cool them. Energy prices are rising, partially due to the impact from the Russian invasion of Ukraine. If natural gas used for heating becomes scarce, people still need to heat their homes for the winter, and electricity is the only fallback plan.
Although electricity and natural gas are not fully exchangeable resources, we only get energy from a finite number of sources. The cost of electricity will escalate along with other energy sources; Europe already increased its price by 20 percent between 2021 and 2022. These costs are soaring more than the inflation rate.
Even as the prices of natural gas and oil continue to rise, several members of the EU are cutting back on electricity production by systematically shutting down their nuclear power capabilities. For example, Germany is planning to shut down all in-country nuclear power by December 31, 2022. We have not replaced all of that power generation capacity with green renewable energy, so Germany will need to increase imports and turn to burning fossil fuels, ultimately driving up the cost of electricity even further. Germany is one example from many countries that consider nuclear power to be an unsafe form of power generation. Spain and Switzerland plan to phase out nuclear power by 2030, and Italy has already done so as of 2022.
Finding a Solution
How does global cloud-flation recede? Technology has been historically responsible for driving costs down and human productivity up. We can only get back to that point by seeing that Moore’s Law continues its path into the future. That can only happen with a diversified investment in semiconductor innovation, and that all has to happen while Taiwan remains part of the free, global, and open market.
It takes about two to five years to add a semiconductor manufacturing plant. Luckily, we may have that time if investments start immediately. Intel recently announced that it would invest $100 billion in semiconductor manufacturing in Ohio. That’s roughly double what Taiwan Semiconductor expects to spend in 2022. Interestingly, Intel only has $30 billion in working capital on its balance sheet as of the end of 2021. The U.S. federal government may step in to help, although we have not seen many details. Regardless, production is expected to rise by 2025.
Taiwan Semiconductor and Samsung are also increasing their investments. Samsung plans to spend a whopping $150 billion by 2030. Both companies can help alleviate supply issues over the next three to four years, but not immediately.
In tandem, 3-nanometer (and smaller) density chips need to come online soon to help drive higher MIPS and lower costs overall. The world's appetite for computing power will only grow over the next two to five years. We need both manufacturing capacity and faster chips to beat cloud-flation.
We also need to get our energy house in order. We need to invest in renewable energy and existing sources such as nuclear power must be reconsidered in light of our current geopolitical situation. We cannot afford to shut down nuclear power plants while cutting off Russian natural gas without viable replacements.
These are massive challenges, and Moore’s Law may fail in the end. As a global society, we have a path to green. Bold steps need to be taken. At CAST AI we are doing our part to reduce waste and cloud spend through automation.