The AI Boom and its Impact on Data Centers

SKIP AHEAD TO

• Construction
• Power 
• Uncertain Outlook
• Efficiency Improvements
• AI technology plateau
• Energy shortfall

The recent boom in AI has been a transformational event for many sectors of the economy. Huge amounts of capital are being invested in AI chips and computational infrastructure, data centers are rapidly sprouting across the country, energy costs are skyrocketing, and the labor market is being reshaped by the adoption of AI technology. This impact is clearly visible in commercial real estate in the rapid changes in the industrial real estate sector, driven by frantic data center leasing and construction. In this blog, we investigate how the AI revolution is reshaping commercial real estate and the major unknowns that will determine whether today’s boom develops into a sustainable long-term growth or a short-lived bubble.

Construction

Data centers are being built in the US at an unprecedented pace. Within the last year, plans for dozens of $100+ million and a handful of billion-dollar data centers have been announced. In Texas, KKR/ECP plans to build a $4 billion complex, and Vantage plans to build a 25 billion data center. In Phoenix, first announced in 2021, Meta plans to spend over a billion dollars on its Mesa, AZ data center complex. AWS has invested over $10 billion to build out its Ohio data center infrastructure. AWS has also made a multi-billion-dollar investment in Atlanta area data centers.  

These deals indicate a staggering pace of data center buildout.

Power 

Data centers consume a massive amount of power. The Crypto industry is estimated to consume between 100-200 TWh a year, roughly equivalent to a small EU country like Greece or Norway. The AI industry has already gone well past the 200 TWh range, and by 2026, it is projected to consume 1,000 TWh a year, an amount equivalent to Japan. This energy consumption is increasingly a major constraint on data center construction – they need to be built in an area with reliable energy supplies or ideally a surplus. Even so, the sheer size of the AI demand shock to the American power market is driving up prices and distorting the entire economic reality.  AI is breaking the traditional linkage between power production and economic growth. AI power consumption may even damage the rest of the economy as higher power prices raise operating costs for local businesses and depress energy-intensive economic activity like manufacturing and construction. These changes increasingly shape where data centers can be built, and how the presence of massive AI data centers has impacts that spill over into non-AI-related components of the CRE sector.

Uncertain Outlook

It can be tempting to assume that trends in AI technology will continue indefinitely. Demand for computation has grown exponentially since the advent of large language models. Projecting these trends forward implies a need for a staggering buildout of data centers in the coming years. However, there are major potential headwinds that may crimp demand for computation in the coming years:

Efficiency Improvements

Eight months ago, the advent of the Chinese AI model DeepSeek led to the largest single-day market cap decline in history, erasing nearly $600 billion from NVIDIA’s market cap. This decline was prompted by DeepSeek’s efficiency – it used clever design to achieve the same performance as OpenAI with far less computational horsepower. If AI researchers are able to push this progress further, our assumption that AI models will gobble up more and more computation on the path to AGI is misguided. If the link between AI progress and more and more computation is broken, we will no longer need to build more and more data centers.

AI technology plateau

The history of AI has been punctuated by alternating periods of rapid progress and plateaus – “AI winters.”  There are some hints that another AI winter may be coming.  The much-ballyhooed release of OpenAI’s GPT5 was a complete dud. The rate at which AI is improving on specific cognitive tasks is slowing down. Although there is a way to go for AI adoption in the workforce, many tasks we expected to be automated with this generation of AI may remain out of reach. If it increasingly becomes clear that throwing more and more computation at the problem is yielding diminishing returns, companies will pull back their AI spending and pop the bubble of AI data center construction.

Energy shortfall

The computation needed to power AI models requires a vast amount of energy. As AI models get bigger and bigger, the energy they need to train is becoming a significant component of total energy consumption. At some point, this lack of energy will become a hard constraint – data centers can’t be built if there is no energy to power them, and local communities and utilities may choose to shut out AI data centers if their power consumption causes dramatic price rises for locals. This scenario would kneecap the AI revolution and drive computation to available markets within the US or even offshore.

These factors are certain to manifest in some form in the coming years. The question is the degree to which they manifest. We lay out three scenarios to capture this wide range of outcomes:

• The AI revolution continues: so far, the LLM revolution has been driven by increases in computational power. The smartest minds in AI and tech are betting this will continue.  

Impact: Current data center buildout trends continue at their torrid pace. Increasingly, markets with limited power or land will be left behind, with construction booming in regions where there is excess.  Across the board, energy prices will increase and spill over into non-AI sectors, affecting their real estate markets and increasingly distorting the economy around AI.  Look for data center prices to continue to boom and for new energy-rich markets, such as Georgia and Ohio, to become major parts of the data center CRE sector. Nationwide, office starting rents of AI tenants are already on an upward trend having increased 9.2% from $76.03 per square foot in Q3 2024 to $78.68 per square foot in Q3 2025, when weighted by transaction size, according to CompStak data. That could be a sign of increased data center demand on the horizon. Additionally, the average weighted starting rent of AI office tenants over the last 12 months is 12.2% higher compared to the average weighted current rent being paid–$68.71 per square foot vs. $61.25 per square foot. Another possible indication that higher build out activity is already underway is the higher ratio of work value to total deal value, which CompStak data shows has increased over that same period, from 8.4% to 15.5%.

• Soft landing: There are many reasons to believe spending on AI computation has moved too far, too fast. It’s entirely possible that AI will continue to reshape our world and hoover up computational resources, but at a reduced rate driven by some combination of efficiency gains and more sustainable spending.  

Impact: While this scenario will still support a build-out of data center infrastructure, current projects will probably be sufficient to address industry needs, and overall demand for new construction will moderate going forward. This scenario should support stable prices in the data center CRE sector, and although energy costs will remain elevated, they are unlikely to rise catastrophically.

• AI winter: AI is, on some level, undeniably a bubble. Investment has exploded at an unsustainable pace, and most decision makers have adopted the rosiest possible projections for AI progress. At some point, the music stops as we realize our assumptions about AI are wildly out of touch with reality.  In the San Francisco and Bay Area office markets, the term length for AI tenants reached a most recent peak in the first quarter of this year at a weighted average of 91.3 months and has since declined to 86.1 months by the third quarter, according to CompStak data. Meanwhile, all other office tenants in that market reached a most recent peak in the second quarter of this year at 92.7 months and has slipped to 83.7 months as of the third quarter.

Impact: In this scenario, demand will collapse, some data center projects may be abandoned or scaled back before completion, and the once unthinkable prospect of excess computational capacity becomes a reality. In this scenario, rent drops and vacancies will become a reality, and growth in new markets will potentially be stopped before they reach critical mass. Energy prices may fall as capacity is built out for demand that never arrives.

Indicators to look for include:

• Server/GPU shipments (Nvidia/AMD earnings)
• Power Purchase Agreements (PPA) strike prices (rising bid prices signal tight supply)
• Utility queue filings (falling requested MW could indicate dampening)
• Vacancy/absorption reports (e.g., CBRE vacancy creeping up)

The current moment in the AI revolution is a paradox – it could go many different ways, each of which will invariably have a major impact on the fabric of the economy and the CRE landscape.  The coming months will be crucial in shaping the direction AI takes the economy. Follow CompStak to be the first to hear how the AI revolution biomes apparent in CRE data. If you need more detailed, granular insights on AI and Data Center lease and sale comps, join CompStak today.