Chief Economist: Four Links That Could Take Down the AI Boom

KEY POINTS  

• The AI and data center boom depends on a complex financing chain linking hyperscalers, chipmakers and asset-backed lending, creating risk if any link weakens.

• Falling token prices, slower hyperscaler revenue growth or weaker demand for debt could undermine the financing needed to sustain data center investment.

• Construction capacity depends on materials, labor and state-level policy; contractors and building product manufacturers should watch upstream risks that could stall projects.

As anyone who has been following our monthly data center reporting knows, construction spending on data centers since 2021 has been increasing at an annual rate of over 100% with no signs of stopping.

The largest US hyperscalers anticipate investing even more in the AI and data center build out in 2027 than ever before while multi-year forecasts of the data center buildout continue to push higher with each new research article publication.

Despite the prosaic outlook, those who have experienced past booms and bust know that all good things eventually come to an end, yet those who know where to look — and what to look for — are usually best able to reduce the impact. So what should we be looking at?

The financing of today’s AI and data center investment boom is unusually complex, with hyperscalers borrowing from chipmakers while chipmakers seek asset-backed financing secured by the very chips they produce.

This creates a circular chain of credit that lacks the durability of more traditional financing.

Any links that break in this “chain” could result in successive links quickly unraveling. 

Four Links That Could Take Down the AI Boom

Link #1: Token Prices

Tokens, like currency, represent a store of value which can be purchased from the AI model builders, also called hyperscalers, and then exchanged for computing effort. The usage-weighted price (the average token price based on actual usage, not simply listed prices) has oscillated dramatically in the last 18-months ranging from a high of over $2.00 per million tokens in mid-2026 to a recent daily October low of just $0.94.

Pricing is depressed by the fact that the budget priced tokens charged for Open LLM models cost a little less than $0.50 per million tokens.

As the capabilities of all models improve quickly and dramatically, including Open LLM models, the benefits of buying premium “proprietary” models could be severely curtailed by capable models that cost 80% less.

The cost pressure of firms to substitute premium tokens for more economical tokens would result in a significant impact on hyperscalers’ revenues. 

The usage-weighted price (the average token price based on actual usage, not simply listed prices) has oscillated dramatically in the last 18-months ranging from a high of over $2.00 per million tokens in mid-2026 to a recent daily October low of just $0.94. Image: ConstructConnect

Link #2: Hyperscaler Revenues

As occurred during the dot-com boom, when companies at the leading edge of a boom are unprofitable, financiers monitor revenues, and arguably more importantly revenue growth, when making lending decisions.

This makes even a small negative deviation from expected revenue growth a significant problem for the firms that are dependent on future debt to continue the scaling up of their business.

Link #3: Debt Sales

Investment banking experts generally believe that the total cost to build and equip all of the data centers needed for an AI-centered future will cost many trillions of dollars.

This makes access to the debt markets essential. Any loss of confidence by the debt markets could see the demand for hyperscaler debt quickly sour, making financing prohibitively expensive, if available at all.

Link #4: Physical Build Out

The final link in the chain is the physical build out. Assuming all of the above links hold, the physical build out would not be guaranteed. The physical limits of what the construction industry is able and capable of constructing remains limited by real assets, primarily materials and labor.

More recently, some states have become far less welcoming to new data center construction through legislation and regulation.

The fact that the construction industry directly benefits from the data center buildout primarily at the end of this chain manifests why it’s important for the industry to monitor the stress, or “fractures”, on the preceding links.

While contractors and building product manufacturers may be most acutely aware of their immediate risks to investing in new capacity to build tomorrow’s data centers, they would be wise to also monitor for upstream tremors that could quickly stall out today’s data center buildout.

Stay Connected 

Stay connected with ConstructConnect News for construction industry news and construction market analysis to stay ahead of what’s building next.

About ConstructConnect

At ConstructConnect, our software solutions provide the information construction professionals need to start every project on a solid foundation. For more than 100 years, our insights and market intelligence have empowered commercial firms, manufacturers, trade contractors, and architects to make data-driven decisions and maximize productivity.

ConstructConnect is a business unit of Roper Technologies (Nasdaq: ROP), part of the Nasdaq 100, S&P 500, and Fortune 1000.

For more information, visit constructconnect.com