Image: Optofidelity

The AI race is becoming too expensive to look like normal tech spending. The next phase is being built with GPUs, networking, storage, CPUs, data centers, power, cooling, land, construction contracts and long-term energy deals, which is why the numbers now look closer to global infrastructure finance than a typical Silicon Valley product cycle.

SemiAnalysis estimates that cumulative AI capital spending from 2024 to 2029 could reach about $11.1 trillion.

The same report projects about $7.1 trillion of AI-related debt outstanding by 2029, which means the industry may increasingly depend on lenders and infrastructure investors to keep the compute race moving.

However, it does not mean “$7 trillion of the $11.1 trillion will be debt” in a simple one-to-one way. It means AI infrastructure could become a massive debt-financed market, where borrowing is tied to GPUs, data centers, customer contracts and future compute revenue.

AI infrastructure spending is moving into trillions

SemiAnalysis’ estimate is aggressive, but other major forecasts point in the same direction. Goldman Sachs estimates about $7.6 trillion of global AI infrastructure investment from 2026 to 2031 across compute, data centers and power.

McKinsey estimates that data centers will require $6.7 trillion worldwide by 2030, with $5.2 trillion tied to AI workloads alone. I ssumed up everything in the below table:

Estimate	Thesis (basically)	Source
$11.1 trillion	Projected cumulative AI CapEx from 2024 to 2029	SemiAnalysis
$7.1 trillion	Projected AI debt outstanding by 2029	SemiAnalysis
$7.6 trillion	AI infrastructure investment from 2026 to 2031	Goldman Sachs
$5.2 trillion	AI data center investment need by 2030	McKinsey

Each forecast counts the AI buildout differently, so the totals are not directly interchangeable.

Some include chips and data centers, while others also factor in power generation, transmission, cooling, land, networking, storage and replacement cycles for older AI hardware.

AI debt is becoming its own infrastructure market

AI clusters require enormous upfront spending, but the money comes back slowly through cloud contracts, GPU rentals, API usage and enterprise subscriptions. That timing gap is pulling banks, private credit firms and infrastructure investors into the market.

A company building a large GPU cluster has to secure several pieces before a lender feels comfortable: chips, power, data center capacity, cooling, networking and customers willing to pay for the compute. The hard part is that each side wants confidence from the other side first.

A lender wants long-term customer contracts before financing the project, while customers want proof that the GPUs, power and data center capacity will actually be available before they commit.

At the same time, Data center operators also need to see both funding and demand before reserving scarce capacity.

You can call this an AI infrastructure financing problem. The market needs capital, customers and data centers to line up together, and delays in one part can hold back the entire project.

AI compute is starting to resemble an asset class, where lenders underwrite future GPU rental income in the same way they might look at aircraft leases, telecom towers or energy infrastructure. The hardware is expensive, the contracts are long, and the value depends on how much demand remains when the project finally comes online.

Nvidia is helping make GPU debt easier to finance

Nvidia’s role is expanding beyond selling GPUs. Data Center Dynamics reported that Nvidia has acted as a financial backstop for some neocloud customers in exchange for a share of cloud revenue.

The structure can make lenders more comfortable because it reduces the fear that expensive GPU capacity will sit idle. If a neocloud cannot rent out enough compute to customers, Nvidia can help support unused capacity under the backstop arrangement. That makes it easier for smaller AI cloud companies to finance large Nvidia-based clusters, while giving Nvidia a deeper role in the economics of the GPU rental market.

That arrangement also shows how closely the AI supply chain is becoming tied together. Nvidia sells the hardware, helps customers finance the cluster, and can participate in the revenue generated by the same GPUs.

The model works if utilization stays high and customers keep signing long-term compute deals. It becomes riskier if GPU rental prices fall, enterprise AI spending slows, or newer chips make older clusters less attractive sooner than expected.

Note: Links to online stores in articles can be so-called affiliate links. If you buy through this link, MIXED receives a commission from the provider. For you the price does not change.