GPU-Based Financing in the Global Data Center Market: A New Standard for Large-Scale Investment Structures

The AI industry is expanding at extraordinary speed. With each new generation of AI models, the demand for computational power — and thus for high-performance infrastructure — increases significantly. Data centers form the backbone of this development. Goldman Sachs estimates that approximately USD 736 billion will be invested in AI infrastructure by the end of 2026, while Morgan Stanley forecasts that cumulative investments may reach as much as USD 2.9 trillion by 2028. Expectations for the economic potential of the AI market are accordingly high.

1. Off-Balance-Sheet Infrastructure Financing

The scale of required capital expenditures is such that even the largest technology companies cannot — or strategically do not wish to — fund them solely through their own balance sheets. Debt financing is therefore becoming increasingly important. In order to preserve balance sheet flexibility, companies are turning to off-balance-sheet structures.

In this context, a special purpose vehicle (SPV) is typically established together with a project developer or financial investor, which undertakes the planning and construction of the data center (D-SPV). Where structured so that the sponsoring AI company neither exercises control over the management nor holds decisive voting rights, the debt associated with the infrastructure financing is not recognized on the corporate balance sheet.

Instead, the AI company commits to the long-term usage of the data center under lease-oriented arrangements. The resulting future lease receivables of the D-SPV are frequently aggregated, securitized and placed with institutional investors as asset-backed securities to refinance the initial investment.

2. New Trend: Off-Balance-Sheet GPU Financing

Data center financing no longer stops at the physical shell. The core assets are the GPUs (graphics processing units) and the high performing chips in them that enable modern AI performance. These chips are capital-intensive, subject to rapid technological cycles and thus present unique financing challenges that traditional structures cannot sufficiently address.

As a result, GPUs themselves are increasingly becoming integral elements of structured finance solutions. In some structures, GPU inventories serve as direct collateral for loans; in others, they are embedded within more sophisticated instruments such as securitizations or blockchain-based asset representations.

Of particular relevance are GPU leasing structures via SPVs. The SPV acquires the GPUs and leases them back to the AI company (G-SPV). Title remains with the G-SPV, while the AI company retains full access to computing capacity without tying up capital in depreciating hardware. Upon a payment default, the G-SPV can liquidate or redeploy the hardware. Such structures are highly scalable and provide access to proprietary compute resources even for smaller enterprises and startups that lack upfront capital.

A recent example is NVIDIA’s investment of approximately USD 2 billion in Elon Musk’s xAI, executed through a dedicated G-SPV which purchases NVIDIA GPUs and leases them to xAI under multi-year terms. This structure grants xAI access to more than USD 20 billion in infrastructure without balance sheet debt or equity dilution. At the same time, hardware-backed cashflows enhance investor confidence, while the hardware supplier effectively finances demand for its own products — accelerating growth in capital-intensive AI markets.

Traditional secured lending remains a parallel pathway: companies pledge existing GPU assets to generate short-term liquidity. Collateral may be physically segregated or remain in use under title transfer or pledge arrangements.

That said, the use of GPUs as collateral involves specific risks. Rapid technological advancements may lead to accelerated depreciation, resulting in under-collateralization. In enforcement scenarios, simultaneous collateral liquidations may trigger a market-moving oversupply, further undermining the wider collateral market for such assets.

3. Tokenized GPUs and On-Chain Financing

A particularly innovative model leverages blockchain technology to digitally represent physical hardware. GPUs are transferred to a G-SPV, securely stored and tokenized — each token representing a legally documented title to a specific GPU unit. These tokens can serve as collateral in digital lending structures. To obtain financing, the G-SPV enters into a contractual custody arrangement with a data center that physically stores the GPUs. Based on this agreement, a digital title token is created and reflects the title to the underlying hardware. The G-SPV then uses the token as collateral within the lending protocol and receives a loan in stablecoins, typically repaid in monthly installments. Because the title is segregated from the AI company, the asset base is insolvency-remote, enhancing asset protection for investors and unlocking additional liquidity.

Advanced structures combine tokenization with contractual title verification, physical audits and theft-and-loss protections. Token holders obtain direct and enforceable title to the underlying hardware, enabling near-automated collateral enforcement and secondary market liquidity. In the event of default, the token can be liquidated through the protocol, allowing the secured party to recover the underlying asset with minimal friction.

However, regulatory limitations remain: unlike in certain U.S. jurisdictions, tokenized proof of title are generally not recognized as legally valid title under current European member states' property laws. As a result, these models still face significant legal uncertainty and are not yet widely practicable in the European Union.

4. GPUs as Strategic Financing Components

Conventional equity and loan financing alone are increasingly insufficient to address the escalating capital needs associated with AI infrastructure. At the same time, operators seek to maintain balance sheet agility and avoid excessive leverage.

SPVs, GPU leasing structures, and the tokenization of physical compute resources create alternative pathways for capital deployment. These models allow corporations to transfer title to assets, mitigate balance sheet burdens and offer investors secured, asset-backed cashflows.

The direction of travel is clear: GPUs are evolving from mere compute tools into strategically significant financing components — forming a new asset class at the heart of global AI infrastructure.

Country Reports – Relevant Legal Considerations in GPU Financing

In essence, the answer to the relevant legal questions depends on where the seller of the GPUs is located, where the G-SPV is established, and where the GPUs themselves are situated.

Some relevant legal issues are listed below on a country-specific basis.

Germany:

Title in the GPUs: Under German conflict of laws rules, proof of title is generally determined under the laws of the jurisdiction where the GPUs are located, i.e., German law if the GPUs are physically situated in Germany. German property law does not recognize tokens representing assets as proof of title.

German insolvency law: Cross-border applicability is generally only triggered if the insolvent party has its "centre of main interest"(COMI) in Germany. In such cases, the title to the GPUs, as mentioned above, becomes critical. Under German insolvency proceedings, it is essential for the G-SPV that its title is recognized in the event of insolvency of the GPU seller.

Regulatory considerations if the G-SPV is located in Germany:

German Banking Act: Permission must be granted by the competent authority, if it determines that the G-SPV conducts financial services (Finanzdienstleitungen), such as financial leasing.

• GPU lease cashflows securitized:
  • Securitisation Regulation (EU 2017/2402): Applies to asset-backed securities, governing transparency, risk retention, documentation, and disclosure requirements.
  • Prospectus Regulation (EU 2017/1129): Governs the obligation to produce a prospectus for public offerings of securitized GPU cashflows when classified as securities (Wertpapiere).
  • German Investment Code: Relevant if funds or structured vehicles are involved, imposing oversight of the competent authority and reporting obligations.
  • German Securities Trading Act: Rules on insider information, market abuse, and transparency in trading GPU-related financial instruments.

The Netherlands:

Title in the GPUs: Under Dutch conflict of laws rules, property law matters, including questions of valid title and whether delivery has occurred, are determined by the law of the GPU location. Dutch civil law does not provide for the recognition of a token as a suitable proof of title. 

Dutch collateral over tokenised GPU’s: Using tokens as collateral in the Netherlands is legally possible, but subject to substantial uncertainties and dependent on the classification of the token, the nature of the underlying right, and the applicable regulatory framework. Current legislation does not provide an explicit property-law basis for tokens as absolute proprietary rights, which means that their protection and enforceability are more limited than for traditional property rights. Further development of Dutch property law and the implementation of European regulation (such as MiCA) are expected to provide greater clarity in the future.

Dutch insolvency law: Pursuant to the EU Insolvency Regulation (EU 2015/848), the law applicable to insolvency proceedings is determined by the debtor's "centre of main interests" (COMI). Of particular relevance to financing of GPUs through an SPV structure is the concept of true sale. Achieving a true sale of the GPUs ensures that the SPV's ownership remains unaffected by any potential insolvency of the seller

Regulatory considerations if the G-SPV is located in the Netherlands:

GPU lease cashflows securitized:

• Securitisation Regulation (EU 2017/2402): Applies to asset-backed securities, governing transparency, risk retention, documentation, and disclosure requirements.
• Prospectus Regulation (EU 2017/1129): Governs the obligation to produce a prospectus for public offerings of securitized GPU cashflows when classified as securities (aandelen).
• Dutch Financial Supervision Act (Wet op het financieel toezicht): Relevant if funds or structured vehicles are involved, imposing oversight of the competent authority,reporting obligations, insider information, market abuse, and transparency