AI: Main Driving Force behind Growth of Data Centers

If you think AWS, Google, and Microsoft are already big now, then wait until 2029. The latest forecast from Synergy Research Group shows that although hyperscale data centers accounted for only about 20% of global data center capacity in 2017, this number will reach nearly 60% by 2029.

-By 2029, hyperscale enterprises will account for nearly 60% of data center capacity, and by 2026, they will account for half of global data center capital expenditures.

-Artificial intelligence is the main driving force behind the growth of data centers.

-Analysts say that the expectation of artificial intelligence workloads shifting towards inference should not affect data center growth forecasts.

Data center capacity refers to the megawatt level computing power used to run workloads. In 2017, local data centers accounted for around 60% of data center capacity, but by 2023, this number has dropped to 37%. Synergy believes that by 2029, local data centers will only account for 20% of global capacity.

Meanwhile, the capacity share provided by Colocation companies is expected to slightly decrease (slightly below 20% in 2017). Synergy points out that this is not because the capacity availability provided by hosting companies is shrinking, but because while their capacity is expected to remain unchanged, the capacity of hyperscale enterprises will nearly triple by 2029.

Ultra large scale cloud computing companies are planning to increase capital expenditures to establish more computing capabilities to meet the demand generated by technologies such as artificial intelligence.

In fact, a recent forecast by Dell’Oro Group suggests that global data center capital expenditures are expected to grow at a compound annual growth rate (CAGR) of 24% by 2028, with Amazon, Google, Meta, and Microsoft accounting for half of global data center capital expenditures as early as 2026.

Baron Fung, Senior Research Director at Dell’Oro, stated that artificial intelligence (AI) is the primary driving force behind capital expenditures in data centers. He said that without it, the compound annual growth rate of the field would drop to around 10% during the forecast period. Fung also stated that the biggest capital expenditure contributor to artificial intelligence infrastructure is servers (with accelerators such as GPUs), which may also include specialized networks, storage, and facilities such as new data centers.

Synergy Chief Analyst John Dinsdale pointed out that the growth of cloud infrastructure services, software-as-a-service, online gaming, e-commerce, and video services is also driving data center capacity demand. He said that for now, artificial intelligence is just icing on the cake, but ultimately it will integrate into all of these things.

Will artificial intelligence at the edge destroy everything?

We recently discussed what happens in certain data centers (known as AI factories) when artificial intelligence processing shifts towards edge inference. When asked if this expected shift would weaken their predictions, Fung and Dinsdale said no.

Fung explained, “Edge inference is a supplement to large training artificial intelligence clusters, rather than a cannibalization. Over time, training models will grow larger. We see these AI clusters grow from tens of thousands of interconnected GPUs or accelerators to possibly one million GPU accelerators. At the edge, there will be more and more devices querying trained models. Therefore, a large amount of computation is still needed to process these requests.

Dinsdale added that although inference emphasizes performance and latency more than pure computing power, the idea that data centers need to fundamentally outperform consumers is misleading.

As the workload of inference increases, data centers need to be closer to customers, and therefore closer to major metropolitan areas with a large number of businesses and consumers. However, in this case, ‘edge’ data centers can still be huge facilities, or they can be CDN type nodes in smaller local areas, presence points, or hosting facilities.