It is time for our annual bulletin to CAPEX obsessives (previous updates from 2016, 2017, 2018, 2019 as well as earlier installments). The hashtag for CAPEX 2020 is “#bonkers”.
The three hypercloud companies – Amazon, Google, and Microsoft – collectively spent almost $97 billion on CAPEX in 2020, up 32% from 2019’s $73.5 billion. Amazon and Microsoft’s spending again hit new highs (bonkers new highs in Amazon’s case) while Google’s spending declined for the second year in a row. Amazon spent more on CAPEX than any company (not just cloud companies) in 2020 [*].
Standard caveat: The reported numbers are the companies’ total CAPEX spend, not just cloud infrastructure, so includes land, office buildings, campus redevelopments, warehouses, panopticonvenience stores, manufacturing tooling, self-driving cars, delivery trucks, flying machines, satellite constellations, hardware that both is and is not required for quantum computing, and – the dream is admittedly on life support – a Google space elevator. The numbers include finance leases for both Amazon and Microsoft, as well as build-to-suit leases for Amazon (these are debt instruments used to finance specific CAPEX expenditures, namely servers and buildings).
In 2020, Amazon’s CAPEX was up a mere 69% to just shy of an absolutely bonkers $54 billion, Google’s spend declined over 5% to $22.3 billion and Microsoft’s increased 14% to $20.6 billion. Amazon (company slogan: “a penny of free cash flow is a terrible thing to waste on abstract accounting constructs like profits”) is accelerating its spending and may soon achieve orbital escape velocity on the back of its CAPEX trajectory, rendering Jeff Bezos’ personal space program unnecessary.
It is beyond bonkers that “a company that sells books” spends more on CAPEX than any other company [*]. And the idea that our other two “software companies” have a CAPEX spend in the same league or beyond the biggest automakers, energy companies, semiconductor manufacturers, and telcos in the world is likewise bonkers.
[*] Where “any other company” is defined as the random set I looked up and charted above. Are there other big spenders I’ve missed? I didn’t look at Chinese companies, mostly because I’m lazy, but also because often their investments are non-economic, accounting dodgy, and audits optional, even for companies listed in the US (if only domestic companies had that luxury, but Wall Street still dreams of selling a collaterized debt obligation to each and every person in China).
The three hyperclouds’ cumulative CAPEX spend since 2000 is almost $444 billion, with over $170 billion of that occurring in the last two years. Efforts to find an amusing comparable failed.
As a percentage of revenue, hypercloud CAPEX is clustered between 12 and 14% of revenue, with Amazon edging out Microsoft for the top spot this year.
Azure and Google Cloud (which includes both Google Cloud Platform and whatever they are calling Google Docs this week) have both seen growth accelerate slightly since “a bat out of Wuhan” became the new “butterfly effect”. The AWS growth rate is lower, but they grew revenue by over $10 billion in 2020, so their dollar growth alone is bigger than GCP.
Amazon #bonkers
By most standards, Amazon had an absolutely magnificent pandemic year (some might say “bonkers”). Revenue growth nearly doubled to 38% from 20% growth the previous year, bringing in over $386 billion. Profits almost doubled. And they spent the aforementioned bonkers CAPEX to support that growth while providing an essential lifeline to hundreds of millions of people in isolation amidst a global pandemic. Not bad.
Amazon’s cash out of pocket for CAPEX exploded: 2020 spending was 238% of the previous year, with over $23 billion incremental spent on hard assets. Debt-funded CAPEX declined 8% (with capital leases — probably AWS servers — down 15% while the smaller sum of built-to-suit real estate leases increased as they grew their warehouse space by 50%). Over the past couple years, Amazon’s CAPEX outlay was split roughly half cash and half debt. In 2020, the cash spend ballooned to 75%. There clearly was a push to spend as much as possible on hard assets (“Quick, buy as many wide-body airplanes as you can before the end of the year!”).
Amazon has always managed for free cash flow, as opposed to profits. Despite the numbers above, the pandemic year was an all-time low when measured against their financial priorities. Even though they stuffed an additional $23 billion into CAPEX, the company was unable to reinvest most of its cash flow into the business and (embarrassingly) reported profits of over $21 billion. One could almost speculate that Bezos’ departure as CEO is not so much his choice, but reflects the board’s disappointment in his dismal 2020 cash flow performance.
At this point, AWS accounts for a small fraction of Amazon’s CAPEX spending, and examining overall Amazon CAPEX – except as an excuse to use the word bonkers– is ever less relevant to cloud infrastructure discussions. The finance leases historically have been the best tell on AWS cloud infrastructure investments, as they reflected server purchases.
If we look at AWS revenue vs. finance leases, we can see they have decoupled, particularly in the last two years. The company probably paid cash in 2020 to convert some of its pandemic windfall into servers, and has less need to use debt to realize its CAPEX ambitions.
Following Amazon last year, Google is now moving to depreciating their servers over four years:
As noted in our earnings press release, we have adjusted the estimated useful lives of servers and certain network equipment starting in 2021. We expect these changes will favorably impact our 2021 operating results by approximately $2.1 billion for assets as of year-end 2020.
That bold bean counter move has a multi-billion impact on their accounting earnings, but as seen by their CAPEX/revenue number, Google has been on a four-year server cycle for a long time. Yet more evidence that accounting can be distracting. CAPEX obsessives care about the gross spend on silicon, glass, steel and concrete, and don’t waste time netting out accounting abstractions like depreciation.
Based on that four-year cycle, we expect Google’s next CAPEX up-cycle in 2022. The company has highlighted they are investing heavily in Google Cloud, but we don’t see it in the CAPEX numbers. Even as they expand their footprint into more geographies, the infrastructure for search and YouTube are vastly larger. Google’s commentary on Google Cloud emphasizes most of that investment is in people.
Google (why does the Alphabet conceit still exist?) also stopped breaking out CAPEX for Other Bets this year, as Loon joined Makani and Titan in the aerial wing of the Killed By Google cemetery. Google continues to shed its Googley-ness and will fail to deliver a space elevator, calling into question their entire existence and social contract compliance. I await Congressional hearings on what went wrong with the space elevator.
Microsoft
Microsoft’s CAPEX remains boring, monotonic and inscrutable, as I have not been able to tease apart their spending on Azure, Office 365 and Bing (which may be a distant number two to Google, but search still has huge infrastructure requirements to store all those copies of the entire Web). There was a brief moment of excitement right after the pandemic broke when Microsoft was short on servers due to supply chain disruptions, but that was remedied the next quarter and boringness restored. Hopefully they can do more CAPEX-wise this year to stand out in this annual commentary.
