Hyperscalers

I promise I will give my take on NVIDIA’s quarter results but before that I want to focus on the magnitude of CapEx investments by the leading hyperscalers like Microsoft, Google, and Amazon. They are noteworthy, reflecting both the strategic importance of AI and the significant financial commitment required to build the necessary infrastructure. 

1)    Microsoft reported a 79% increase in CAPEX (Q1 2024), reaching a record $14 billion in a single quarter, driven by investments in data centers, GPUs, and specialized chips necessary for AI model training and deployment. Amy Hood, Microsoft’s CFO, told analysts that CapEx will keep growing in the 2025 fiscal year.

2)     Google parent Alphabet reported capital expenditures of $12 billion, up 90%, “driven overwhelmingly by investment in our technical infrastructure,” CFO Ruth Porat told analysts. The increase “reflects our confidence in the opportunities offered by AI across our business,” she said.

3)    The total CapEx of AWS for the Q1 2024 was approximately $14 billion, and the company expects this to be the lowest quarterly figure for the year, with annual CapEx projected to exceed $60 billion. This spending is primarily driven by the need for advanced infrastructure to support growing demand in AI and cloud services(The Stack)

4)    Facebook parent Meta reported $6.7 billion in quarterly capital expenditures, down from more than $7 billion a year earlier, but raised its CapEx guidance for the year to a range of $35-40 billion, up from $30-37 billion before.

Source: Bloomberg/Apollo Research

These investments are not merely expenditures on physical assets like dedicated GPUs and specialized chips to train and run AI foundation models or laying transoceanic cables;  but also on the intellectual infrastructure required for AI, including software development and advanced R&D in AI technologies. This dual focus is crucial as it underpins the ability of these firms to scale AI models from the experimental phase to widespread operational deployment.

From an economic perspective, once the investment have been made by the hyperscalers, then a transition to OpEx occurs, when these AI systems are operationalized, necessitating ongoing expenditures for maintenance, updates, and energy consumption. OpEx increases are expected as firms move from the initial phase of deploying AI infrastructure to the continuous operation of AI systems at scale.

Technical considerations play a critical role in this transition. The operational costs associated with running AI models, particularly those that are computationally intensive, such as large language models (LLMs), include significant expenditures on electricity, cooling systems, and data management. For example, as data centers become more densely populated with AI-optimized hardware, the demand for advanced cooling technologies and power distribution systems escalates, directly contributing to OpEx.

Last year, Omdia's data pointed to AI accounting for all server spending growth. Now it claims demand for AI has accelerated datacenter investment, with 2024 projected to see a 28.5 percent increase in capex spending "backed by the corporate cash reserves of major hyperscalers."

 Servers sales are set grow 74 percent to $210 billion this year – up from 2023's figure of $121 billion. However, datacenter thermal management spend is forecast to grow by 22 percent, reaching $9.4 billion. Power distribution infrastructure revenue will exceed $4 billion in revenue for the first time, and uninterruptible power supply revenue will grow 10 percent to $13 billion.

In USA, the roughly 5,300 data centers are mostly run by big tech firms like Google, Amazon, Microsoft, Meta and Apple, and consumed more than 4% of all electricity in the U.S. in 2022. It's projected to more than double to 9% by  2030, according to the Electric Power Research Institute, a research organization and nonprofit focused on energy.

Obviously, its extremely difficult to calculate the additional investments needed by the national grid operators in order to support and transfer electricity to those datacenters. A cost that will be federalized to each and every taxpayers of the countries with most datacenters.

Source Cloudscene/Statista

In USA alone, the electricity demand is growing twice as fast as previously expected due to the rapid expansion of data centers and electric vehicle adoption (US electricity load growth forecast jumps 81% led by data centers, industry: Grid Strategies | Utility Dive). At the heart of the near-term load growth is a roughly $630 billion investment in facilities that have large loads, including $481 billion for manufacturing and industrial facilities and $150 billion for data centers, according to the report from Grid Strategies, entitled “ The Era of Flat Power Demand is Over” , if it is to be believed (https://gridstrategiesllc.com/wp-content/uploads/2023/12/National-Load-Growth-Report-2023.pdf)