North American Data Center Leasing Area Surges 80% in a Single Quarter, AI Computing Demand Drives Infrastructure Construction Frenzy
According to the latest report released by Colliers, a global leader in real estate services and investment management, on August 5, 2026, the North American data center market experienced unprecedented explosive growth in the second quarter of 2026. Data shows that data center leasing area in the quarter surged nearly 80% compared to the same period last year, setting a new historical high. This astonishing growth rate is directly driven by AI computing demand represented by generative AI and large language models.
The report indicates that the total power capacity of data centers currently under construction in North America has exceeded 15 gigawatts, but the vast majority of this capacity is pre-leased by tech giants and cloud service providers before completion. This "pre-lease before construction" phenomenon highlights the current market's hunger for AI computing infrastructure. Analysts at Collar emphasized in the report that due to extended delivery cycles for power supply, transformers, and other critical equipment, as well as land scarcity, the delivery speed of data centers can no longer keep pace with demand, leading to continuously declining vacancy rates in core markets and rising rental prices.
Tech Giants' Capex Race: Betting Hundreds of Billions on the AI Future
Behind this data center construction boom is an undisguised capital expenditure race among tech giants. Based on the latest disclosed financial data, the combined capital expenditure of the four giants—Microsoft, Google parent Alphabet, Amazon, and Meta—in 2026 is expected to easily exceed $250 billion, with the vast majority flowing into AI infrastructure, including data center construction, computing chip procurement, and related power facility support.
Specifically, Microsoft's capital expenditure in the fourth quarter of its 2026 fiscal year (ending June 30) increased by over 60% year-over-year, primarily supporting the infrastructure expansion of its Azure cloud services and AI platform. Alphabet's capital expenditure also saw a significant year-over-year increase in the second quarter, with company executives explicitly stating during the earnings call that the risk of underinvesting in AI infrastructure far outweighs the risk of overinvesting. Amazon's capital expenditure also remained high, with its AWS cloud service division accelerating the deployment of next-generation AI computing clusters globally. The capital expenditure guidance from these tech giants sends a clear signal to the market: the AI computing race is far from over, and infrastructure construction will remain a core theme for years to come.
Power Bottleneck Emerges: From Site Selection to Clean Energy Solutions
However, the rapidly expanding data centers are facing a severe practical problem—where will the power come from? The power capacity of a hyperscale data center can easily reach hundreds of megawatts, equivalent to the electricity consumption of a small city. Collar's report specifically points out that power supply has become the primary bottleneck constraining data center delivery. In core markets such as Northern Virginia, Dallas, and Silicon Valley, local power companies are struggling to meet the rapid power connection demands of data centers, with grid connection waiting times for some projects stretching for years.
To solve this challenge, tech giants are increasingly turning their attention to regions with abundant, clean, and cheap electricity. Ohio, Iowa, and parts of Texas in the United States, with their rich wind and solar energy resources, are becoming hotspots for next-generation data center site selection. Meanwhile, nuclear energy, as a stable, zero-carbon baseload power source, is receiving unprecedented attention. Recently, Microsoft signed an agreement with Constellation Energy to restart a reactor at the Three Mile Island nuclear power plant in Pennsylvania specifically to power its data centers. Additionally, investment in small modular reactors (SMRs) is accelerating. Although large-scale commercial deployment is still years away, they are already seen as one of the ultimate solutions to meet the long-term energy demands of AI computing.
Chip Supply and Supply Chain Restructuring: The Game of H200 and Next-Gen Blackwell Ultra
Beyond power, the supply of computing chips also tugs at the market's nerves. Although NVIDIA's H200 chips have entered the mass delivery phase, strong demand still leaves some customers facing extended lead times. Meanwhile, market expectations for the next-generation Blackwell Ultra B300 platform have been pushed to the limit. According to supply chain sources, NVIDIA has finalized the mass production timeline for the B300, but whether the initial production ramp-up can meet the explosive demand growth remains an unknown.
This tight supply-demand situation not only solidifies NVIDIA's position as the dominant force in computing chips but also brings sustained performance growth momentum to its supply chain partners, such as TSMC, SK Hynix, and Supermicro. However, investors are also beginning to closely monitor the potential impact of competitors like AMD's MI400X on the market landscape, as well as the progress of tech giants' self-developed chips (such as Google's TPU and Amazon's Trainium). This healthy competition and strategic maneuvering are driving the entire computing chip industry forward at an accelerated pace.
Investment Implications: Full-Chain Opportunities from Hardware to Infrastructure
For US stock investors, the investment logic for AI computing power is expanding from pure chip stocks to the broader full chain of infrastructure. Beyond chip leaders like NVIDIA and AMD, data center real estate investment trusts (REITs), power facility suppliers, server manufacturers, and related software platform companies are all benefiting from this construction wave.
Collar's report provides a unique perspective for observing AI industry trends: the construction speed of data centers is essentially a hard indicator of the AI industry's prosperity. In the foreseeable future, as multimodal large models and AI agent applications further proliferate, the demand for inference computing will succeed training demand as the engine for a new round of data center expansion. This race centered around computing real estate and power energy has only just begun.
