Webscale Network Operators - 4Q23 Market Review: Network/IT Capex up 4% in 2023 even as Total Capex Dips, Good Outlook for 2024-25 driven by GenAI Land Grab, Big Jump in Sector Profitability amidst Layoffs

This report reviews the growth and development of the webscale network operator (WNO) market since 2011. In the most recent 12 months (1Q23-4Q23), webscalers represented $2.37 trillion (T) in revenues (+6.2% YoY), $293 billion (B) in R&D spending (+9.4% YoY), and $192B in capex (-5.1% YoY). They had $679B of cash and short-term investments (+9.8% YoY) on the books as of December 2023, and $562B in total debt (+1.9% YoY). Webscalers employed approximately 4.093 million (M) people at the end of 2023, down from the YE2022 total of 4.194M.

VISUALS

bscale revenues accelerate YoY growth rate each quarter of 2023

Revenues for the webscale sector floundered in 2022, but the opposite happened in 2023. In 2023, global economic growth improved, the digital ad market recovered, TikTok began to face some backlash, cloud services penetration marched on, and Huawei's device business remained in the doldrums. All factors tended to benefit the revenue growth measured by our webscale tracker.

Each quarter of 2023 saw an increase in the YoY growth rate, resulting in annual 2023 revenues of $2.237 trillion, up 6.2% from 2022. Improvements were widespread. China's leading cloud services providers (Alibaba, Baidu and Tencent) all went from revenue declines in 2022 to increases in 2023; Amazon's revenues grew 11.8% in CY23 increase (2022: 9.8%) and Microsoft's rose 11.5% (2022: 10.4%). Most significant, perhaps, was Meta's big jump, growing revenues by just under 16% in 2023 after 2022, when revenues fell for the first time in history. Alphabet remained stable with revenue growth again in the 8-10% YoY range. Apple disappointed with a -0.5% YoY change as it still struggles to find growth now that 5G networks are widely deployed. Oracle performed very well for its size, growing 12.1% due to a mix of acquisition activity and success with its Oracle cloud infrastructure platform.

Revenues per employee in the webscale sector ended 2023 at $584K, from $539K in 2022; free cash flow per employee jumped even more noticeably, from $73K in 2022 to $109K in 2023. These swings are due both to rising revenues and even more quickly rising profits, and the sector's headcount growth taking a pause in the last two years. Webscale headcount ended 2023 at 4.093M, slightly up from the YE2021 figure but handily down from the 2022 total of 4.194M. Meta saw by far the biggest dip in its workforce, down 22% in 2023 to 67,300. This recent wave of layoffs was due in part to overhiring during COVID. There is a good chance that we will see additional layoffs across the sector though as big webscalers attempt to impement Generative AI in their own operations in search of labor cost savings. The same thing is already happening in the telco sector.

Tech spending: total capex dipped in 2023 but the tech portion rose 4%; R&D spend remains elevated

The main reason we cover webscalers is because we care about their technology spend. Webscalers spend heavily on data centers and related cloud infrastructure in support of both their services and operations. So, while it's important to know about revenue & profitability trends in webscale, the vendors selling into the market (our main clients) care about technology spend. That means capex and R&D. And not just capex in general, but more specifically the technology component of capex, i.e. "Network/IT/software" broadly defined.

Total capex did fall a bit in 2023, down 5.1% to $192B. This doesn't imply a negative market sentiment. Capex spend by quarter can vary significantly due to supply chain and other issues, and the webscale market is especially volatile since it's driven by just a few big players; the top 4 capture 77% of global capex, after all. Moreover, the tech portion of capex actually grew in 2023, up 4% YoY. The disparity is due to an easing of spend on transportation, logistics, fulfillment and non-tech infra categories at Amazon, Alphabet, Alibaba and others.

R&D spend within webscale amounted to 12.3% of revenues in 2023, even higher than the 12.0% recorded in 2022. The R&D intensity ratio has been creeping up in webscale for some time, as companies spend heavily to enter into new markets such as robotics, healthcare, financial services, and more. A good chunk of this R&D cash also targets the development of proprietary tech for the physical infrastructure of data centers underlying their operations: new chips and other hardware, not just smarter software.

Looking beyond the 2023 numbers, what is most important is that last year the webscale market found a new lifeforce, a new reason for being.

For the prior several years, adoption of cloud services was a primary motivator for incremental investments; they drove data center spread and design evolution at Alphabet, Amazon, Microsoft and Oracle. Short-form video content and gaming were also important drivers. This could be seen in the big related investments made by Alphabet and Meta, and Microsoft's biggest acquisition ever (of Activision, for $69B). Then in early 2023 - alongside these other trends - GenerativeAI's potential suddenly reached mass market awareness. In reality, GenAI was cooking for many years prior to this, but January 2023 was a turning point with the release of ChatGPT: it reached 100 million users by the end of the month. Other platforms were rushed to market, and any big tech company (webscaler or not) without investments in GenAI quickly scurried around to cobble something together, or invest in a third party. Amazon, for instance, invested heavily in Anthropic, as did Alphabet. Chinese webscalers each launched their own native offerings.

There is surely some unrealistic hype being floated about the potential of GenAI to solve all the world's problems - cure diseases, find solutions to global conflict, invent new forms of transportation, etc. This happens every time markets get excited about a new technology. There is always a 'tech leader' willing to make obnoxiously grandiose statements, and always a receptive audience to echo some of the nonsense. That said, GenAI has real potential to develop new markets over the next few years, and it is a legitimate reason to accelerate data center investments. The exact shape and size and location of such investments are not yet clear, and that uncertainty can slow down investment. But GenAI is not going away. We suspect the quest to monetize GenAI will drive a land grab for more capable data centers and supporting supercomputer clusters for several years to come.

US accounts for over 50% of webscale capex

This report series traditionally breaks out revenues by region for each webscaler. Towards the end of 2023, we added our first regional breakout of capex, focused on the US. Our analysis finds that the US has amounted to between 50-60% of global webscale capex for most of the last decade. This percentage increased in the last two years, ending 2023 at just over 60%. The US will continue to be the largest single country market, by far, for the foreseeable future. Most of the key GenAI innovators are based in the US and rely heavily on US Internet infrastructure, and data center capex will follow this pattern. However, the US ratio may return below 50% within a couple of years as webscalers expand their footprints in other regions. Spending pickups by China-based cloud providers will be one driver of this moderation.

Webscale market compared to telecom

A decade ago, the webscale sector did not exist. Big tech companies were just beginning to build their own data centers to optimize their cost structure, operational efficiency, and time to market. But webscale capex was a rounding error in the overall market for network infrastructure. That's not the case anymore. Webscale capex surpassed $200B for the first time in 2022. Annualized webscale capex has since fallen below $200B, but that is a short-term blip.

Telco capex is still higher, and will remain so for the next few years. But, webscale capex is far more concentrated, as it is dominated by a few big spenders, and it is focused on a smaller range of product types and vendors. Some aspects of webscale capex are more leading edge; innovations in the data center often impact other types of networks (e.g. high-speed optics for telco backbone networks). As such, the market will continue to be important for lots of vendors - and not just chip suppliers like NVIDIA, Intel and AMD.

Spending outlook

While capex dipped a bit in 2023, the outlook is strong as key players pursue GenAI opportunities even before that market is well developed or defined. Here is a summary of the spending outlook for key webscalers:

• Amazon (27.4% of global webscale capex): CY23 capex of $48.4B was down $10.2B YoY, mainly due to lower fulfillment and transport spend. In 2024, company expects capex to increase YoY due mainly to infra spend at AWS, due partly to adding capacity in AWS for regional expansions, and "additional investments in generative AI and large language models"
• Microsoft (18.3%): 4Q23 capex was lower than expected due to a third party capacity issue, but capex will "increase materially" on a sequential basis in 1Q24, driven by cloud and AI infrastructure. Company notes, "You started to see the acceleration in our capital expense starting almost a year ago...looking forward, you'll tend to see ... accelerating capital expense to continue to be able to add capacity in the coming quarters, given what we see in terms of pipeline." Also, is considering a $100B supercomputer data center campus needing up to 5GW of power, in concert with OpenAI, to develop AI models. Could open in 2028, coined "Stargate".
• Alphabet (16.8%): reported capex in 4Q23 was $11B, driven by servers and then data centers. The 4Q stepup in reflects company's outlook "for the extraordinary applications of AI to deliver for users, advertisers, developers, cloud enterprise customers and governments globally and the long-term growth opportunities that offers. In 2024, we expect investment in CapEx will be notably larger than in 2023." Drivers include "the extraordinary applications of AI within Google DeepMind, Google Services, Google Cloud, it's across the board for users, for advertisers, developers, cloud enterprise customers, governments. And it's really the long-term opportunity that offers."
• Meta (FB) (14.2%): expects CY24 capex of $30 to $37 billion, a $2B increase above prior range. Notes that it underbuilt the infra needed to support the growth of Reels in 2023 - is building to support both Reels and another similar sized service "so we wouldn't be in that situation again." Capex growth to be "driven by investments in servers, including both AI and non-AI hardware, and data centers as we ramp up construction on sites with our previously announced new data center architecture." Not providing guidance for 2025 and beyond but says it expects that its "ambitious long-term AI research and product development efforts will require growing infrastructure investments beyond this year."
• Apple (5.0%): as usual does not provide capex guidance, just says it will "never underinvest in the business..."
• Oracle (3.6%): 4Q23 capex was $1.1B. Expected capex for 12 months ended May 2024 is $8B, meaning "second half CapEx will be considerably higher as we bring online more capacity..."
• Chinese cloud providers: no concrete guidance but early signs in 2024 point to increased spend as all key players are investing in GenAI platforms and attempting to accelerate cloud revenue growth, including overseas. Alibaba, for one, appears to be pushing a price war. Domestically, the telcos are strong in cloud services in China, and also provide carrier-neutral site services, as do GDS/ChinData/Vianet etc. Huawei is also a strong cloud player in China. So, the mainstream webscalers face more competition but also more options for building out their footprint.

Forecast update

Our Dec 2023 forecast called for $202B in 2023 webscale capex; that proved a bit too high. The official targets for 2024 and 2025 are $203B and $218B, respectively. We see no reason to modify these targets, but note that there is now significant uncertainty. They could be too low, but, some of the optimistic projections issued by webscalers will change as they face resource constraints or pursue more asset light strategies, or be crowded out by companies not currently classed as webscalers e.g. OpenAI. One certainty is that this is an exciting time to be selling into data center infra markets.

COVERAGE

Top 8 WNOs

Other WNOs