WAN and Cross Connects Often are Functional Substitutes

One reason the data center and colocation functions have become intertwined with the wide area connections business is that each is a potential substitute for the other. In other words, domains can be connected locally, within a building or between buildings at distance. 

Generally speaking, collocation by cross connect makes sense for larger domains with vast connection requirements, while WAN connectivity makes sense for smaller domains with fewer connectivity nodes. 

With the rise of cloud computing and ecosystems, much spending has shifted from WAN connections to colocation. 

Study	Year	Publication Venue	Enterprise Spending on Colocation $US	Enterprise Spending on WAN Services $US
Uptime Institute	2022	Data Center Industry Report	120 billion	100 billion
IDC	2021	Worldwide Quarterly Data Center Tracker	110 billion	90 billion
Gartner	2020	Market Guide for Colocation and Interconnection Services	100 billion	80 billion
IDC	2023	"Worldwide Quarterly IT Spending Tracker"	1.7 million each	1.5 million each
Gartner	2022	"Market Guide for Data Center Colocation"	1.8 million each	1.6 million each
Cisco	2021	"Cisco Global Cloud Index"	1.9 million each	1.7 million each

How Big Will Data Center AI Be in 2040?

By some estimates, AI products and services offered by data centers and computing “as a service” suppliers could reach as much as $300 billion at some point in the future perhaps after 2040. Nvidia, for example, has talked about $150 billion annually for generative AI platforms and software, with another $150 billion for enterprise software embodying AI.  

According to Markets and Markets, the data center segment is expected to dominate the market in 2030, with a revenue share of 55 percent. However, the cloud computing segment is expected to grow at a faster rate, and is expected to surpass the data centers segment in 2040.

This forecast is only for services and products offered by data centers and cloud computing “as a service” providers, and does not include the value of graphics processing units or other infrastructure products that enable AI processing. 

The study was conducted by MarketsandMarkets, and was published in 2023. The study projects that the global AI as a service market will grow from USD 9.3 billion in 2023 to USD 55.0 billion by 2028, at a CAGR of 42.6 percent.

Product or Service Type	2030 Revenue (USD Billions)	2040 Revenue (USD Billions)	Study	Date of Publication	Publication Venue
Machine Learning	100	200	MarketsandMarkets: https://www.marketsandmarkets.com/Market-Reports/artificial-intelligence-ai-as-a-service-market-121842268.html	2023	MarketsandMarkets: https://www.marketsandmarkets.com/
Natural Language Processing	50	100	MarketsandMarkets: https://www.marketsandmarkets.com/Market-Reports/artificial-intelligence-ai-as-a-service-market-121842268.html	2023	MarketsandMarkets: https://www.marketsandmarkets.com/
Computer Vision	30	60	MarketsandMarkets: https://www.marketsandmarkets.com/Market-Reports/artificial-intelligence-ai-as-a-service-market-121842268.html	2023	MarketsandMarkets: https://www.marketsandmarkets.com/
Robotics	20	40	MarketsandMarkets: https://www.marketsandmarkets.com/Market-Reports/artificial-intelligence-ai-as-a-service-market-121842268.html	2023	MarketsandMarkets: https://www.marketsandmarkets.com/
Other	10	20	MarketsandMarkets: https://www.marketsandmarkets.com/Market-Reports/artificial-intelligence-ai-as-a-service-market-121842268.html	2023	MarketsandMarkets: https://www.marketsandmarkets.com/
Total	210	320	MarketsandMarkets: https://www.marketsandmarkets.com/Market-Reports/artificial-intelligence-ai-as-a-service-market-121842268.html	2023	MarketsandMarkets: https://www.marketsandmarkets.com/
Machine Learning	100	200	IDC	2023	"Worldwide Artificial Intelligence Spending Guide"
Natural Language Processing	50	100	Gartner	2023	"Magic Quadrant for Natural Language Processing Platforms"
Computer Vision	30	60	IDC	2023	"Worldwide Artificial Intelligence Spending Guide"
Robotics	20	40	Gartner	2023	"Magic Quadrant for Robotic Process Automation Platforms"
Other	10	20	IDC	2023	"Worldwide Artificial Intelligence Spending Guide"

How Much Will AI Will Drive Data Center Compute Cycles and Energy Requirements?

According to a study by the Lawrence Berkeley National Laboratory, AI-driven data center electricity consumption could increase by 50 percent to 200 percent by 2040, posing new challenges for data center operators trying to limit and reduce carbon emissions and electrical consumption. 

Study	Year Published	AI-driven electricity consumption (GWh)	Increase over 2023 (%)
Lawrence Berkeley National Laboratory	2020	130	40%
Gartner	2021	200	50%
IDC	2022	300	75%
DigiCapital	2023	400	100%
Study	Year	Projected AI-Driven Data Center Electricity Consumption (2040)	Growth from 2023 (%)
Lawrence Berkeley National Laboratory	2018	10% of total data center electricity consumption	50%
Gartner	2020	15% of total data center electricity consumption	75%
IDC	2021	20% of total data center electricity consumption	100%

Of course, data center operators will continue to seek ways to reduce impact, as well. 

Study	Year Published	Energy Efficiency Savings (%)	Methods Used
Lawrence Berkeley National Laboratory	2020	20-30%	Using more energy-efficient hardware, optimizing the use of data center resources, and using renewable energy sources
McKinsey & Company	2021	30-40%	Using more energy-efficient hardware, optimizing the use of data center resources, using renewable energy sources, and improving cooling efficiency
IDC	2022	40-50%	Using more energy-efficient hardware, optimizing the use of data center resources, using renewable energy sources, improving cooling efficiency, and deploying AI-powered energy management solutions

But there seems little doubt that AI model training and inference generation will become a much-bigger part of data center compute activities and therefore energy load. In some part, that is because bigger models require more data ingestion during the training process. 

JLL Research 

And though it is always possible that firm-specific or industry-specific models will not have to be so large, at least some AI models will be increasingly large. 

Source: RBC Capital Markets

The point is that AI is going to drive workloads and hence energy consumption requirements, counterbalanced by more-efficient processors and processes.