Are the Cashflow Kings becoming Capex Kings?

When the online & software companies that dominate the market today were overtaking the largest industrial & financial companies in terms of market capitalisation, one clear point of contrast had been how much more efficient the new leaders were in their use of capital.

Until the mid-2000s, many of the most highly capitalised companies in the world required a lot of capital to run their operations. Globally, these were companies like the oil majors, global banks, General Electric, Toyota, Vodafone, and Intel. Although the most valuable companies in the world were typically achieving better operating metrics than their less-highly-valued peers, the capital-intensive nature of the industries they operated in meant that they were typically achieving returns on equity of between 10% and 20%.

If you look further back in history or further down in market cap, you would see many capital-intensive businesses that were doing much worse than this. For example, in the 1990s the New Zealand share market had been full of very capital-intensive businesses like Fletcher Challenge, Carter Holt, and Fisher & Paykel Appliances, which often struggled to achieve double-digit returns on invested capital.

By contrast, the new market leaders of the 2010s required very little capital to run their operations. By 2010, Microsoft, Apple, and Google already ranked amongst the world’s most valuable companies, yet each of them held less than US$8 billion of fixed assets. By 2015, when Amazon and Facebook (now Meta) had joined these companies in top-10 ranks of market capitalisation, Amazon had fixed assets of US$22 billion (mainly distribution centres for its retail operations) and Facebook still had fixed assets of less than US$6 billion.  

By 2015, the contrast was stark. In 2015, Exxon was still the world’s most valuable “old world” company with a market capitalisation of US$324 billion, but to achieve the earnings required to justify that valuation it held fixed assets (including developed petroleum reserves) with a depreciated balance sheet value of US$252 billion, representing 78% of market capitalisation. By contrast, at the same time, Apple was valued at US$599 billion, yet only carried fixed assets of US$22 billion and Microsoft was valued at US$440 billion with fixed assets of just US$16 billion.

By my estimates, the net capital that Apple and Microsoft were using to run their businesses in 2015 were both about zero, because some customers were paying these companies in advance and they were paying their suppliers a bit more slowly than their customers were paying them. The combined cash “float” from these two aspects of their business meant that their fixed assets were effectively internally funded by a favourable cashflow mismatch from their everyday businesses. 

The Importance of Being Earn-y

When a business is growing, a high return on capital can be extremely valuable.

To understand this, let’s first consider a capital-intensive business that is expected to grow at 10% per annum, but is only achieving a 10% return on equity. This means that if this business is generating $1 billion of profits, it would need to hold $10 billion of equity. Assuming that the capital requirements of this business would grow in proportion to growth in revenues and profits, if the company’s profits grew from $1 billion to $1.1 billion over the course of a year, we should expect that the capital required by this business would also grow from $10 billion to $11 billion. This would mean that all of the profits of the business would be absorbed in funding the increased capital requirement, and that there would therefore be no “left-over profit” to distribute to shareholders.

Now let’s consider a second much less capital-intensive business that is also growing at 10% per annum, but is achieving a 50% return on capital. This means that it would only require $2 billion of equity to generate $1 billion of profits. And (assuming that capital required grows in proportion to profits), if profits grew by 10% to $1.1 billion, this company would only need to retain 20% of profits to grow its equity proportionately from $2 billion to $2.2 billion. This business could distribute 80% of profits to shareholders and still have enough capital to fund its growth.

Clearly, a company that can grow at 10% per annum while distributing 80% of profits to shareholders is significantly more valuable than a company that is also growing at 10% per annum but is unable to distribute any profits to shareholders. This example demonstrates that a good return on capital is invaluable for growing businesses.

I think this point has sometimes been missed by investors who have chased rapidly-growing businesses in capital intensive industries. In particular, some investors have been prepared to pay particularly high valuations for Tesla, apparently linked to its growth opportunity. Unlike many companies operating in software and online services, Tesla operates in a relatively capital-intensive low-margin industry, and will therefore probably need to invest several dollars of incremental capital for every dollar of incremental profitability.  This would tend to mean that it should be less valuable than a software company with a similar growth opportunity.

Pay attention to changes returns on invested capital

Of course, returns on capital are not always constant over time, and some businesses that are achieving strong returns on existing capital may not be able to maintain these strong returns if they try deploying additional capital to grow their business.

A recurring example of this is that over the years many companies that have been achieving strong returns on capital in New Zealand (often due to a strong market position in a relatively oligopolistic market) have interpreted the strong returns from NZ as evidence of their management genius, and have therefore sought to replicate their success in Australia by spending a lot of money building or buying a business on the other side of the ditch. More often than not this trans-Tasman expansion results in significantly lower returns on capital than the companies had been achieving in New Zealand.

More generally, it seems that companies that “force” growth by expanding into new areas or by spending capital more quickly than their customers demand it will typically experience a deteriorating trend in returns on invested capital, whereas companies that are stingier with capital spending (only deploying at a rate that is necessary to meet growing demand from customers) will typically maintain or even improve their returns on invested capital.

Consistent with this, academics and quantitative analysts who have looked at measures like return on equity or return on invested capital have generally found that these measures do not work nearly as well for picking stocks in isolation as they do when they are used in combination with indicators (like the change in returns on capital, or growth in fixed assets) that help to signal whether the returns that companies are achieving on new capital are likely to be as good as the returns that they have achieved in the past.

This perspective of differentiating companies that are “pushing” capital spending from those that are simply meeting the pull of increased demand helps to explain an apparent paradox in investment markets. The apparent paradox is that:

• In aggregate the share market values companies much more highly than the historic book values of the money they have deployed in building their businesses;
• This observation could be seen as logically implying that to create value, businesses could just deploy more capital in fixed assets and so-on, and wait for the share market to value this at a premium to book value;
• BUT, the empirical evidence shows that the companies that deploy new capital at the fastest rate have typically produced lower shareholder returns that companies that are being more careful with their capital spending.

This apparent paradox is easily understood when we make the distinction between companies that are “pushing” capital in the hope of generating extra profits and companies that are simply spending the capital that is required to meet growing demand for their products and services.

Are the Magnificent Seven now playing “build and they shall come”? 

As discussed earlier in this article, companies like Microsoft, Apple, Alphabet (Google), Amazon, and Meta (Facebook) had historically stood out as beacons of capital efficiency.

However, many of these companies are not as capital-efficient as they used to be. Microsoft’s balance sheet value of property, plant, & equipment has more than doubled over past 2 ½ years, and this trend seems to be accelerating, as you can see from the graph below:      

 

Source: Microsoft, Bloomberg

Over the 2 ½ year period that Microsoft’s investment in fixed assets has doubled, revenues have grown by +32.6%, so it seems that every dollar of incremental revenue is demanding a significantly higher level of fixed asset investment than was the case with Microsoft’s pre-existing revenue base.

Alphabet, Amazon, and Meta have also been building their fixed assets at a faster rate than the growth in revenues. Each of these companies have grown their revenues by between +27% and +29% over the past 2 ½ years, but they’ve each grown their investment in fixed assets by between +37% and +78%.

Why are these companies growing their fixed assets so fast? Three key contributors stand out:

1. Amazon, Microsoft, and Alphabet all have rapidly-growing cloud services businesses (AWS, Azure, and Google Cloud respectively). These are more capital-intensive than those companies’ previous income streams.
2. All of these companies are making significant investments in AI, which is not yet generating significant revenues.
3. All of these companies have changed their accounting policies recently, to write computer servers off over 6 years, rather than the previous standard of 5 years.

I will discuss each of these issues in turn:

Growth of Cloud Services

Cloud services essentially involve the cloud services provider (e.g. Amazon’s AWS or Microsoft’s Azure) renting out time on their servers, so that customers can perform computing tasks (or store data) without having to invest in their own computer servers.

At its most commoditised level (e.g. bulk data storage), cloud services is little more than a financing transaction, not too dissimilar to leasing buildings to tenants. This is an inherently capital-intensive business model.

However, fortunately for the cloud services providers, most on-premise computing resources are only used to their full potential a small proportion of the time. As Marvin the robot complained in Hitchhiker’s Guide to The Galaxy “Here I am, brain the size of a planet, and they ask me to pick up a piece of paper”. People who have a highly spec’d computer sitting on their desks but spend most of their time reading emails and .pdf documents are effectively asking Marvin to pick up pieces of paper for them. If their reason for having the highly spec’d computer is that they occasionally need to perform a much more computationally difficult task, then a more economically sensible solution for them is to have a lower spec’d computer sitting on their desktops which will occasionally call on the greater computing resources of AWS or Azure or Google Cloud when needed. This way, the cloud services providers (AWS, Azure, or Google cloud) can effectively “rent” short periods of time on the same computer server to hundreds of different customers and achieve revenues that are multiples of what could be achieved from a financing lease to a single customer.

(By way of clarification - there would currently be very few examples where a normal office worker’s desktop PC automatically uses cloud service providers to perform computationally difficult tasks, but the principle still applies. Rather, the cloud service providers are providing a platform that is more often used by IT professionals and data geeks to run specific applications or processes which might otherwise be run on their own servers. Many of these processes generate short bursts of computationally-intensive activity separated by large periods of relative inactivity.).

Looking at the financial reporting of Microsoft and Amazon (parent companies of Azure and AWS respectively) over the years, it seems that a rough rule of thumb has been that they need about $1 of investment in fixed assets for every $1 of revenues from cloud services. As these businesses generate pre-tax operating margins in the order of 35%, this produces a healthy return on invested capital of roughly 35% before tax.

35% pre-tax returns on invested capital are a fantastic outcome for shareholders, and accordingly, no-one would disagree that the cloud services companies should continue investing to meet demand in this area.  However, it does represent a dilution of returns compared to what Microsoft is used to achieving from its core business of software subscriptions and what Alphabet achieves from online search, as both of these activities require very little capital and generate very high margins. The returns on these core businesses of Microsoft and Alphabet are close to infinite, as the net capital employed in these businesses is close to zero.

Amazon’s traditional retail business also achieves high returns on capital, even though it generates low margins and requires significant investment in distribution centres and delivery infrastructure. The reason that Amazon can achieved high returns on invested capital from online retailing is that it gets paid by customers faster than it pays its suppliers, and the “float” it gets from holding onto money for the period between when customers pay it and when it pays its suppliers is more than sufficient to fund all of its investment in fixed assets.

Overall then, growth of cloud services is leading to some dilution in the returns that the key protagonists (Amazon, Microsoft, & Alphabet) are achieving on their invested capital, but this is because the returns that they achieve on their existing businesses are so good, and not because there is anything shabby with the returns they are achieve from cloud services.

Investing in AI

The recent acceleration in capital spending at Microsoft, Amazon, Meta, and Alphabet (in the interests of brevity, let’s call them “MAMA” for the remainder of this article) is largely due to the investments these companies are making in AI. MAMA are building and buying AI servers, and building data centres to house these AI servers. They are also investing in the training of AI models, although in most cases this investment in AI training is not being capitalised on the balance sheet (and if it is, it would be classified as “Intangibles” rather than “Property, Plant, & Equipment”).

For the cloud service providers, a significant element of this AI investment is building AI resources that they can hire out to customers looking to train their own AI models. However, all of the MAMA companies are also investing in adding AI capability to their other businesses.

As Tim Chesterfield pointed out in last week’s “Margin Call”, there is not yet a lot of visibility about how demand for AI capability and services will contribute to revenues and profits in the future. Hence, there is arguably a large “build it and they will come” aspect to current investment in AI capability.

The MAMA companies do not deny that there is uncertainty about future demand for AI, but they seem to be taking the attitude that the risks of not investing in AI are greater than the risks of investing in AI, and therefore seem committed to continue to invest in AI servers for at least the next 18 months. When you look at the long list of companies that lost relevance over the years due to a failure to align their businesses to emerging technology trends, this attitude probably makes sense. For each of the MAMA companies, the capital investment that they will make in AI capability over the next 12 months is unlikely to be more than 4% of market cap.

If it turns out that the cloud services companies have over-estimated growth in demand for AI training, they should be able to hit “pause” on new AI investment, and wait for demand to catch up with the capacity they have built. Hence, it is unlikely that their entire investment in AI capability will be wasted.

However, obsolescence does mean the would be a real cost to investing in AI capability head of the curve, as the rapid growth in the power of GPUs means that by 2026, the economic value of AI servers built using 2023-era technology will likely be significantly lower than their original cost.         

In summary, there is a risk that the MAMA companies could destroy a bit of value with their current rush to invest in AI servers, but the potential value destruction should be limited to no more than 4% of market cap, assuming that they will be willing to make an objective judgement around investment in this area when the underlying demand for AI training and AI services becomes clearer (probably some time in 2025).

Slowing down accounting recognition of depreciation

Over the past 3 years, each of the MAMA companies have changed their accounting policies, such that they now write-off computer servers in a straight line over 6 years rather than 5-years (which was their previous policy).

This accounting policy reflects the physical reality that they continue to use most of their computer servers for more than 6 years. However, it ignores the economic reality that any computer server purchased today incorporating Nvidia’s A100 chips (launched in 2022) will become significantly less valuable in 2025 when Nvidia’s significantly-more-powerful Blackwell chips are available, and the economic value of these servers will depreciate even further in subsequent years as Nvidia and other companies bring even more powerful chips to the market.

Regardless of its merits, this change in accounting policy is contributing in a modest way to the growth in the balance sheet value of the fixed assets held by the MAMA companies. I would estimate that it has contributed roughly 2% to the growth in their fixed assets over the past 12 months.

It is also worth pointing out that this change in accounting policy is artificially boosting the reported profits of the MAMA companies. They’re recognising depreciation on servers purchased over the last 3 years at a slower rate, because they now take a view that servers should be depreciated over 6 years, but they’re not recognising any depreciation on servers bought between 5 and 6 years ago, because they’ve already written these servers off their books.   

Should we be worried?

For a quantitative investor, the trend of rapidly growing fixed assets at the MAMA companies would be a cause for concern. Historically, companies that have grown fixed assets faster than revenues have generally not performed well for investors.

If the trend continues, then today’s market leaders will increasing look more like the “old economy” market leaders of 15 years ago, whose market values have failed to keep up with the magnificent seven.

Two of the three reasons we identified for the growth in fixed assets of these companies could be cause for concern. Firstly, the investment in AI capability is somewhat speculative (as no one really knows how much future demand there will be for this capability) and secondly, the changes in depreciation policy are arguably a form of accounting trickery that fails to acknowledge how rapidly the value of a computer server will decline in the first couple of years after it is purchased.

But the largest driver in the growth of fixed assets for Amazon, Microsoft, and Alphabet has been growth in the demand for cloud services. While the cloud services business requires more capital than some of the other areas that these companies have been operating in, it nonetheless generates extremely strong returns on invested capital, and we should not be worried if these companies continue to deploy capital in this area in rough proportion to the growth in cloud services revenues, even if this slightly dilutes the strong aggregate returns that these companies are achieving on their invested capital.

For now, we continue to hold investments in the “MAMA” companies, but we are aware that the growing capital intensity of these businesses could be a negative sign, so we will continue to monitor this issue.

Nicholas Bagnall is chief investment officer of Te Ahumairangi Investment Management

Disclaimer: This article is for informational purposes only and is not, nor should be construed as, investment advice for any person. The writer is a director and shareholder of Te Ahumairangi Investment Management Limited, and an investor in the Te Ahumairangi Global Equity Fund. Te Ahumairangi manages client portfolios (including the Te Ahumairangi Global Equity Fund) that invest in global equity markets. These portfolios include investments in Microsoft, Alphabet, Amazon, Meta, and Exxon Mobil, which were mentioned in this article.