The AI boom is creating a new asset class around data centers, chips, power, long-term contracts, debt and insurance. The decisive question is not only how much gets invested, but who bears the loss if expectations fail.
Published on September 12, 2026 · by Alex · Artificial intelligence · Data centers · Financing
Artificial intelligence could require several trillion dollars of data-center investment before 2030. The exact figure depends on the scenario and source, but the direction is clear: banks, funds and insurers want to participate while trying not to keep too much risk.
That contradiction may matter more than the AI boom itself. Projects are getting larger, contracts longer and the technology behind those contracts changes every year.
A large data center can require billions of dollars. Big technology companies have announced enormous infrastructure commitments, but part of that capacity needs debt and contracts that survive several technology cycles.
A loan can last for decades while Nvidia launches new chip generations and Google, Amazon and Microsoft design their own processors. Each generation can make the previous one less competitive, leaving loans backed by assets whose value ten years from now is uncertain.
| Part of the system | What to watch |
|---|---|
| Data center | Utilization, power, permits and residual value |
| Lease | Duration, customer concentration and renewal price |
| Debt | Maturities, guarantees and refinancing |
| Insurance | Outages, disasters, permits and litigation |
| Outside capital | Who absorbs the first loss |
Banks financing these projects are exploring structures that share or transfer exposure to funds, insurers, pensions and independent vehicles. That does not by itself prove another financial crisis, but it changes the question: the larger the investment, the more important it is to know who bears the loss if expectations fail.
In August 2026 Nvidia said it had signed memoranda with major financial firms to mobilize more than $500 billion of third-party capital for AI infrastructure. That is not cash Nvidia has delivered or one loan: these platforms and agreements still have to become projects, contracts and actual financing.
Nvidia's 10-Q shows the interdependencies: commitments with select AI clouds, conditional guarantees and power and data-center projects tied to future contracts. A maximum guarantee is not a realized loss.
The market is demanding higher returns as it tries to value what happens when the first data-center lease ends. Will another customer pay the same price? Will those chips remain competitive? Will computing demand be unchanged? If not, the data center remains physically there but its cash-generating capacity falls while the debt remains.
Data centers depend on electricity, networks, permits, water and equipment. An outage, disaster, permit delay, lawsuit or community opposition can stop a facility that cost billions. Insurance can limit part of the loss, but it does not remove interruption, cost-overrun or refinancing risk.
The International Energy Agency estimates data-center electricity use could nearly double between 2025 and 2030. That expansion also depends on projects finding financing and earning a sufficient return.
Wall Street is building a new asset class around AI, data centers, chips, power, long-term contracts, debt and insurance. If five years from now there is more AI capacity than needed, the data centers and the debt will still be there.
That is why it is not enough to watch how much Microsoft, Meta, Google or Amazon spend. Watch who is buying the risk they no longer want to keep. When we know where that risk ends, we will know who may pay the bill for the AI boom.
This analysis is not a prediction of a crisis. It is a framework for following financing, utilization and guarantees. Read also Nvidia turned its chips into financial assets… and that is where the risk appears.
Primary sources: Nvidia, AI infrastructure financing platforms; Nvidia, fiscal 2027 second-quarter 10-Q; International Energy Agency, energy and AI. References to several trillion dollars and risk distribution are market scenarios, not guaranteed forecasts.
Data centers, power and servers are being built with debt, contracts, guarantees and third-party capital. There is no single loan and banks do not all carry the same exposure.
It can be spread across banks, credit funds, insurers, pensions, infrastructure vehicles and operators. The answer depends on each contract, guarantee, insurance policy and project.
Lease duration, real utilization, power costs, permits, guarantees, customer concentration and who finances the asset when the first contract ends.
This article is the written version of the Saturday analysis.
Watch the video on TikTok · Follow the channel
