Cost-Cutting From Tech Giants Beginning To Slow Data Center Growth
As the largest tech companies scramble to adjust to persistent economic volatility and the sudden emergence of an artificial intelligence arms race, they are making strategic decisions that could have major implications for data center leasing and development.
Cloud providers Amazon Web Services, Microsoft and Google, along with social media juggernaut Meta, are by far the largest data center users, the center of a digital infrastructure landscape that has largely been shaped by their needs. But quarterly earnings reports and calls over the past month have revealed lingering macroeconomic headwinds are having a greater impact on these companies than their leadership — and many analysts — expected.
The hyperscalers, as the big tech firms are known in the data center space, are cutting costs as a result. How they plan to make those cuts could have widespread repercussions for data center real estate ranging from lower leasing demand to reduced tax revenue in the industry’s key markets.
At the same time, an escalating fight has emerged for control over the future of artificial intelligence, with hyperscalers pivoting massive resources toward technologies that are changing where and how data centers are built.
Following an earnings season that saw significant strategic shifts from the cloud and social media behemoths, here are the most important takeaways for data centers.
Data Center Capital Expenditure Growth Is Slowing
Hyperscalers’ exploding spending on data center leasing and development has driven the data center construction boom, but data center capex growth is slowing, at least for now.
For the first time in a long time, one of the four largest tech companies is actually cutting back its data center capex. Meta told investors last week it will spend $4B less on data centers in the coming year, an announcement that came on the heels of the social media giant canceling a planned data center campus in Denmark and indefinitely pausing projects in Texas, Alabama and Idaho.
The decision to reduce data center spending comes amid wider cost-cutting efforts at Meta. While Meta’s leadership hasn't refuted the notion that lower data center capex means a smaller data center footprint than previously planned, Chief Financial Officer Susan Li framed the spending cuts as part of a pivot toward AI-focused data centers that will be cheaper to build.
“We're shifting our data centers to a new architecture that can more efficiently support both AI and non-AI workloads,” Li said on the earnings call. “Along with the new data center architecture, we're going to optimize our approach to building data centers, so we have a new phased approach that allows us to build base plans with less initial capacity and less initial capital outlay, but then flex up future capacity quickly if needed.”
Meta may be the most dramatic case, but quarterly earnings reports from cloud giants AWS, Microsoft and Google also show the overall growth of their digital infrastructure spending plateauing. Hyperscaler data center capex grew at just 9% in 2022, according to Synergy Research Group, well below the 20% average annual growth rate over the previous two years.
Google’s leadership said its overall capex, a large portion of which is focused on data centers and other technical infrastructure, will remain flat in the coming year after growing from $23.5B to $31.5B between 2019 and 2022. That is a significant slowdown, even if the company says a larger percentage of that spending will be devoted to technical infrastructure. Microsoft’s capex is also expected to remain flat, while AWS doesn't disclose capex figures.
This shift comes amid slowing earnings growth for all the major cloud providers.
“The market growth rate has slowed down,” said John Dinsdale, managing director at Synergy Research Group. “The worsened economy has caused some enterprises to more closely review spending on cloud services.”
While Dinsdale and industry leaders emphasized their belief that this slowdown is temporary, there is evidence things will get worse before they get better. AWS’ revenue growth fell 20% from a year earlier, but Chief Financial Officer Brian Olsavsky indicated on the company's earnings call that the figure fell even further in the first month of 2023.
Temporary or not, hyperscalers’ cutbacks suggest that spending on data centers and other digital infrastructure may not be as immune to short-term macroeconomic volatility as many had previously suggested.
Cost-Cutting Decisions Could Hit Coffers Of Key Industry Hubs
Tech giants are increasing the life span of their data center servers. This seemingly mundane IT decision may well lead to some unintended side effects: lowering the tax revenues that data centers provide for local governments and potentially making life more difficult for developers.
Both Google and Meta will be refreshing servers and other equipment in their data centers less frequently, the companies announced on earnings calls this month. Google increased the working life of its servers from four years to six years, while Meta extended its refresh cycle from 4.5 years to five. This continues a trend, as both Microsoft and AWS made similar announcements earlier this year.
While there are a number of reasons for hyperscalers to try to squeeze extra life out of their servers, cost reduction is the primary driver. Google expects its longer refresh cycle to save it $3.4B next year, and Meta estimates its savings at $1.5B.
While the cloud and social media giants save money, municipal budgets in major data center markets could take a hit as a result.
One of the most attractive elements of data centers for local governments is the significant tax revenue they generate, often stemming from taxes on the value of equipment inside the data center.
Because the value of that equipment depreciates annually, part of what makes certain markets like Loudoun County so appealing to data center operators and tenants is the rapid depreciation schedules for data center gear that they offer. In Loudoun, servers are taxed at 50% the first year, with the rate decreasing by 10% each year after.
When the largest data center users extend the life of their data center equipment, it means fewer new servers taxed anywhere near their full value. And that means a potentially significant revenue hit.
In Loudoun County, where data center revenue makes up a significant percentage of the county’s tax base, Economic Development Executive Director Buddy Rizer said the county expects to see tax revenue drop as a result of these longer server life spans.
“It’s not insignificant, but it’s also not insurmountable when we actually have the opportunity to plan for it,” Rizer told Bisnow, adding that data center operators generally inform the county about upcoming changes to refresh cycles well before they are announced. “The fact that the industry has given us a heads up about it has really made it a much easier issue for us to deal with.”
Elsewhere in Northern Virginia, lower revenue from data centers could make things more difficult for developers and their proponents in areas where new data center builds are facing a rising tide of local opposition. In local papers and government hearings, advocates for contested projects like Prince William County’s PW Digital Gateway have made improved schools and other beneficiaries of data center tax revenue the centerpiece of their pitches.
Declining returns in public coffers may undercut this argument and provide a talking point for the opposition.
Tech Giants Dive Headfirst Into AI Infrastructure War
An AI arms race is underway.
In the three months since the success of Microsoft-linked ChatGPT accelerated AI’s push into the mainstream, Microsoft’s Big Tech competitors have been in an all-out scramble to accelerate development of their own AI products and services.
None of these companies is new to AI — in fact, leaders at Google and Meta have suggested their companies did much of the legwork that made products like ChatGPT possible. But the significant focus on AI in quarterly earnings calls revealed the extent to which hyperscalers view this as an inflection point in determining who the leaders in this sector will be in the coming years.
Google leadership reportedly issued a “code red” following ChatGPT’s release and has quickly pumped resources into fast-tracking a range of AI-integrated technologies. The company announced this week that it is releasing a search-oriented AI chatbot called Bard, as well as its $300M equity investment in AI startup Anthropic. Anthropic’s specific products have been held close to the vest, but the company is known for creating a chatbot that passed an exam at George Mason University.
Meta has also frantically shifted its attention toward AI products, as evidenced by its redesigned data center architecture. While the social media giant has largely focused on AI as a tool for optimizing its interconnected social media and advertising products, the company has begun redirecting resources toward generative AI technologies like ChatGPT.
“One of my goals for Meta is to build on our research to become a leader in generative AI,” CEO Mark Zuckerberg said on an earnings call this month.
So what does this mean for data centers?
As Bisnow reported last month, widespread adoption of AI will fundamentally change where and how data centers are built. But the staggering speed at which competition for AI leadership has exploded — and with the world’s largest companies throwing billions of dollars at these technologies to not fall behind — this digital infrastructure transformation may happen much more quickly than expected.
It is a reality that Microsoft CEO Satya Nadella addressed on the company’s quarterly earnings call.
“Azure’s core infrastructure is being transformed,” Nadella said, speaking about the company’s cloud platform. “The core of Azure, or what is considered cloud computing, is fundamentally changing in its nature along with how compute, storage and network come together.”