.jpg)
I. Predictions and Impacts of the Data Center Construction Boom Main Predictions
The year 2025 will be the "Year of Data Centers," as we transition from the hype cycle to an industry-driven construction cycle.
AI will catalyze energy transformation, including solar construction, battery innovation, and the revival of nuclear energy.
Some hyperscale companies may not be flexible enough to cope with the rapidly changing requirements of data centers, and new industrial AI players will emerge.
There will be many reports in the next six months about delays in data center construction due to liquid cooling, cluster size, and power access issues.
The industrial capacity required to build new AI data centers will stimulate the economy and create jobs, including in the steel, energy, trucking, and construction industries.
When new data centers come online, the cost of training and inference provided by AWS, Azure, and GCP will decrease, benefiting startups.
Current Status of Data Center Construction
Amazon: In the first half of 2024, AWS announced a $50 billion new data center project, including 216 new buildings, with a commitment of $100-150 billion over the next 15 years, with project plans in multiple locations.
Microsoft: With an existing 5GW energy capacity, data center construction doubled in 2024, with project investments in multiple locations.
Google: The smallest among the three major cloud providers, it is building data centers in multiple locations and also faces the challenge of expanding TPU clusters.
Meta: Although it does not have a cloud business, it is expanding data center capacity to support internal AI projects such as Llama, accumulating a large number of GPUs, and has several new data centers under construction.
II. Challenges Faced
Energy Issues
There are a large number of energy projects that need to be connected to the power grid, requiring more electricity to support new data centers, mainly increasing power generation capacity in the form of solar and wind energy, and also creatively utilizing existing energy resources. Power limitations are particularly prominent in some major data center markets, leading to much new development in "secondary" markets.
Technical Issues
The next generation of Nvidia chips requires liquid cooling, supply chain shortages, and diesel generators need to wait for two years. Cluster sizes are unprecedented, and models may need to be trained on distributed clusters across multiple data centers, with lithium-ion batteries being key to new data center construction, and new methods are also being considered to reduce costs and increase capacity.
III. Market Competition Landscape
Hyperscale Competition
The operational rigor of hyperscale companies in building data centers will be tested, with varying results, winners and losers, and new industrial AI players will have the opportunity to fill operational gaps.
Existing Market Participants
Existing market participants such as Equinix, Digital Realty, and CyrusOne face a "demand shock," either to improve their own benefits or to lose market share to new entrants.
IV. Economic Impact and Financing Methods
Economic Impact
The AI industrial phase will stimulate the economy, with manufacturers of industrial supply chain components, energy companies, nuclear reactor operators, and others benefiting, and in the short term, job opportunities in construction and industrial labor may outnumber researchers in Silicon Valley.
Financing Methods
Large technology companies mainly deploy capital from outside their balance sheets, and financial companies also participate in providing leverage. Many private equity firms provide early-stage capital for construction and GPU procurement in exchange for Microsoft's IOUs and reasonable returns.
There will eventually be many AI factories, but it is unknown whether demand will fill them. At least the cost of training and inference should continue to decrease, benefiting startups.