The IMD Wrap: Are server life extensions putting profits before performance?

The sharp-eyed readers among you will no doubt have noticed that WatersTechnology recently began expanding its coverage to include the earnings calls of relevant companies in our space.

Our aim isn’t to bring you the basic earnings numbers. Rather, we hope to pick up on other nuggets of information that the companies share in their earnings discussions that might fly below the radar of the mainstream financial press. So, for example, new product announcements, commentary on the latest trends like generative AI and how it affects their business, or insight into the impact of partnership deals and pending or completed acquisitions.

Sometimes those nuggets are hard to detect. In their recent earnings calls, both Google parent Alphabet and Microsoft spent more time talking about YouTube TV’s NFL Sunday Ticket deal and Microsoft’s Xbox gaming console than about their cloud offerings. Yet, an interesting item in both companies’ financial results was that both saw a material impact on their bottom line from reassessing the expected lifespans of the servers and equipment that underpin the companies’ cloud offerings.

Google noted that in January, it completed an assessment of the useful lifespans of its servers and network equipment, and as a result, adjusted what it deems to be the useful life of a server from four years to six years, and the estimated life of certain network equipment from five to six years.

Big deal, you might think. Well, actually, yes…yes, it is.

The result of that reassessment was a reduction in depreciation expense of $977 million in Q3 and $2.9 billion year-to-date, which translates to additional net income of $761 million and $2.3 billion for the respective periods.

Meanwhile, Microsoft’s own reassessment, completed last year, extended the life of its servers and network equipment from four years to six years, citing advances in its software that allow the company to use hardware more efficiently. The change helped offset a decline in device sales and an increase in the cost of revenue “driven by growth in Microsoft Cloud,” which translated to an increase in operating income of $804 million and net income of $653 million for the Q4 period.

Amazon also extended the estimated useful life of its equipment last year to five years for servers and six years for networking equipment, also citing software advances helping hardware operate more efficiently.

Snowflake didn’t make mention of any similar change in its results. That may mean it doesn’t feel the need to do that, or assesses its risks differently, or perhaps is already doing it and doesn’t feel the need to mention it again.

But there’s a clear reason for eking out every last penny: As uptake of the cloud by all types of customers and industries continues to rise, the clouds must necessarily get bigger to accommodate them and all their data. Additionally, the amount of equipment required to support those will inevitably also increase, as will the costs of purchasing and supporting that equipment.

For example, Snowflake spent almost $6.3 million on purchasing property and equipment in Q2 this year, up from $3.85 million in the same period last year. And while the vendor increased its revenues and client base, it saw a net loss for Q2 of about a quarter of a billion dollars.

I’m not a finance wiz, and nor do I play one on TV, so I asked someone who is: Larry Kinsella, currently CFO of Livv.ai. He has also served as CFO of Tabb Group, OneMarketData, BT Radianz and Moneyline Telerate, and also held senior finance roles at Dow Jones.

Kinsella calls depreciation “more of an art than a science,” adding that it’s routine for companies to adjust depreciation policies to make sure they properly reflect the loss in value each month of an asset, such as computing equipment.

“This review can shorten or lengthen the depreciation period, such as 30+ years ago when all computers were ‘big iron,” lasting a long time. Then laptops came about, which due to their build and their price, almost became disposable. In those cases, companies adjusted their depreciation policies to reflect the faster drop in the PCs’ value, by shortening their useful life. So, simply put, a longer useful life/depreciation period results in lower depreciation expense per month, and a shorter useful life/depreciation period results in higher depreciation expenses each month,” Kinsella says.

While this doesn’t affect a company’s EBITDA (earnings before interest, taxation, depreciation and amortization), it does affect net income and earnings per share. “So, if you lengthen the useful life of the asset, you decrease the monthly depreciation expense, which will, in turn, increase the profits of the company, increasing net income,” Kinsella says.

Steven Roe, CEO of West Highland Support Services, which was acquired last week by network operator TNS, knows about supporting hardware and software.

“Historically, hardware has been depreciated on your books over three years, which was aligned with the typical lifespan of hardware,” and in particular, failure-prone components like fans and drives,” Roe says. “Of course, hardware can be maintained longer than that, but financial services is a very stringent world.”

Amazon, like many others, used to cut off the useful lifespan of a server at three years, but began extending that in 2020. And it’s important to note that this isn’t simply an accounting tweak, as some have called it: Amazon repairs and reuses working components from older and failed equipment to ensure it remains operational and can be put back into circulation (here’s a video of exactly how Amazon does this), reducing spend on new hardware and minimizing waste.

Other bottom-line benefits, Roe notes, are saving on the cost and resources associated with large “lift-and-shift” hardware replacement projects, and how the resulting changes in supply chain can also benefit a company’s carbon footprint—though cloud providers would need to weigh that against the prospect of replacing existing hardware with drastically more energy-efficient machines, which could also yield positive results.

As you can tell from the links in this story, I’m by no means the first person to notice the financial impact. But what about other risks from those old components—especially given the ongoing need for higher-performance machines as each cloud provider and many of their clients commit increasing resources to AI-related developments?

Alex Wolcough, CEO of UK-based technology consultancy Greenbirch Group, notes that cloud providers have always offered a range of newer and older machines, often segregating them to support specific functions and available at different price points. For example, today, Intel Sapphire Rapids is the fastest (and most expensive) machine available, with Icelake offered as a balanced price/performance machine. Five years ago, Icelake was the fastest machine on offer.

Under a platform-as-a-service or software-as-a-service model, the cloud provider would decide and allocate which machines are devoted to clients and services. Under an infrastructure-as-a-service model, the client can choose their hardware. Wolcough warns, though, that “signing up for a single virtual machine on a computer that fails could be catastrophic for the end-user.”

Roe echoes this concern, noting that on a shared cloud infrastructure, the operator can move a client’s compute processes and data from a failing machine onto different virtual machines. One thing that West Highland would assess from a risk perspective, he says, is what level of service is needed and specified to ensure that a firm isn’t at a disadvantage to its peers in the event of a hardware failure in the cloud.

London-based IT consultancy CJC is another monitoring specialist. “I would say most modern and tech-savvy firms do great observability on their infrastructures, and it’s a service/product that CJC provides with our Mosaic platform,” says Steve Moreton, global head of product management at CJC.

He says cloud providers deliver excellent insight into the usage of their infrastructures, which probably informs their decisions about the suitability of extending a hardware asset’s lifespan.

“If you have good observability, you can potentially make excellent savings on keeping some legacy infrastructure,” Moreton says. However, he also warns that “there is a point where this can bite you on the backside. Most companies look to get three years out of a server, so four to five years usage on paper gives a good return on investment, unless there are other technologies that are cheaper to run or are getting to the point where the server dies or starts failing. You can quickly find yourself in a bad place with keeping legacy tech too long.”

And though he won’t name the affected company, he says CJC has already seen instances of this in 2023.

Ultimately, what do companies want from the cloud? Price, performance, elasticity—sure. But at the top of the list are reliability and security. The fact that the cloud providers conduct thoughtful assessments before extending a server’s life, rather than picking an arbitrary number, gives me comfort. And hardware should be more resilient now than in the past. But with cloud usage, overall, increasing, and data volumes ballooning—and with market events likely to affect all participants at once—let’s hope resources aren’t being stretched too thin.

Or, as West Highland’s Roe says, check your contract to ensure your providers can guarantee appropriate resources in the event of any failure.

We’d love to hear how you approach this challenge. You can get the discussion going on our LinkedIn and Twitter accounts, or write me privately here: max.bowie@infopro-digital.com