Data Centers and AI Infrastructure: The Coming Wave of Controversy

Data Centers and AI Infrastructure: The Coming Wave of Disputes

Industry Background: Unprecedented Capital Expenditure and Potential Contract Disputes

The global data center industry is in the midst of the largest infrastructure buildout cycle in modern history. Goldman Sachs Global Investment Research projects cumulative AI infrastructure spending of approximately $7.6 trillion between 2026 and 2031, with annual spending rising from roughly $765 billion in 2026 to approximately $1.6 trillion by 2031, encompassing computing, data centers, and power. BloombergNEF data shows more than 23 gigawatts of IT capacity under construction globally. JLL forecasts that nearly 100 gigawatts of new capacity will come online between 2026 and 2030, doubling global installed capacity to approximately 200 gigawatts, with a compound annual growth rate of about 14%.

Against a backdrop of abundant capital and intense growth pressure, power availability has become the key factor constraining buildout geography. A March 2026 survey by Vinson & Elkins of 200 senior infrastructure developers and private capital investors found that 89% of investors expressed confidence in the sustainability of current capex levels over the next one to five years, but 96% of developers expect power supply to reshape the geographic landscape of construction.

However, historical experience shows that in any industry driven by large-scale project financing, disputes tend to emerge in the operational phase, typically after the first major market dislocation—demand corrections, sudden shifts in input prices, regulatory changes, counterparty defaults, or technology transitions that leave asset economics falling short of initial financing assumptions. The long-term supply contracts being signed today often carry 10- to 20-year terms, take-or-pay obligations, and detailed performance commitments; these contracts will be tested over the coming years.

Market Impact: Long-Term Contracts, GPU Price Volatility, and the LNG Lesson

Disputes during the construction phase, while important, are already well studied. More challenging are operational-phase service level agreements (SLAs), power purchase agreements (PPAs), and long-term capacity commitments and take-or-pay arrangements between hyperscalers and operators.

The economic scale of these contracts is enormous. Two AI Factory campus leases signed by Applied Digital with the same US investment-grade hyperscaler—Delta Forge 1 and Polaris Forge 3—have been described as 15-year take-or-pay leases with a combined base-term value of approximately $15 billion, approaching $36 billion if all renewal options are exercised. Hut 8's 15-year Beacon Point lease with another US investment-grade hyperscaler, on a "triple net, take-or-pay" basis, is valued at approximately $9.8 billion over the base term. When unit economics shift, these commitments will become fertile ground for disputes.The most compelling omen comes from the GPU rental market. The hourly rental price of NVIDIA H100 fell from a peak of about $8 in 2023 to about $1.70 by the end of 2025, then rebounded to about $2.35 in early 2026—a more than fourfold swing in less than 30 months. The accelerating pace of GPU generational turnover—H100, H200, B200, B300, soon to be replaced by Vera Rubin—means that the unit economics of compute underlying long-term contracts can quickly diverge from reality.

The liquefied natural gas (LNG) industry offers a reference. LNG export facilities, like data centers, are capital-intensive, long-term assets that rely heavily on take-or-pay mechanisms. The dispute surrounding Venture Global LNG is particularly illustrative: buyers such as Shell, BP, and Edison accused it of reselling contracted cargoes on the spot market. Yet the outcomes were strikingly different: Shell lost in ICC arbitration, BP won at the liability stage, and Edison reached a commercial settlement. This pattern reminds us that even in an industry with decades of arbitration precedent, outcomes still depend on the facts, the tribunal, and the advocates.

Competitive Landscape: Who Benefits, Who Faces Pressure?

The coming wave of data center disputes will reshuffle the bargaining dynamics across the industry chain. Hyperscalers with top-tier credit ratings hold stronger negotiating positions in disputes and can shift risk onto operators and equipment vendors. Second-tier players—emerging neoclouds, frontier AI labs, and specialized GPU providers—have less stable business models and are more vulnerable to performance-related disputes.

The boundaries of responsibility among power suppliers, equipment manufacturers, and operators will be continuously tested. Long-term service agreements (LTSAs) and joint venture arrangements will add complexity. In the LNG cases, divergent arbitration outcomes among buyers show that legal uncertainty will also shape the market landscape. It is foreseeable that law firms, arbitral institutions, and technical advisors will benefit from this wave of disputes, while capital providers will begin demanding more refined contractual protections.

Implications for Companies: Prepare for a Down Market Before Signing Contracts

For companies, the most important task now is to identify risk windows in long-term contracts. Four characteristics that differ significantly from the LNG template warrant particular attention:1. Asset Lifespan and Technological Obsolescence: LNG facilities can operate steadily for 20 to 30 years, whereas data centers are exposed to rapid changes in cooling technology, power density, GPU generations, and workload mixes. AI workloads are expected to shift from training-dominated to inference-dominated around 2027, which will alter tenants' demands on the underlying assets. 2. Counterparty Profiles: LNG buyers are mostly national gas companies or regulated utilities, whereas data center tenants include not only hyperscalers with extremely strong credit but also emerging players whose business models are not yet mature. Dispute characteristics will vary by counterparty. 3. Price Review Mechanisms: Long-term take-or-pay contracts should include price adjustment or periodic review mechanisms to avoid widespread arbitration waves similar to those common in the European natural gas industry when market prices fluctuate significantly. 4. Force Majeure and Change-in-Law Clauses: Global AI regulatory policies (such as the AI Act) and data governance regulations may alter the contract enforcement environment, so contracts need to define related risks in advance.

Enterprises should re-examine termination rights, caps on damages, acceleration clauses, and dispute resolution clauses in existing contracts to ensure they have legal buffers amid unforeseen market changes.

Future Outlook: 12 Months, 24 Months, Three Years

Within 12 months, as numerous front-end projects enter the final stages of construction, the first wave of operational disputes may surface, mainly centered on SLA breaches and delivery delays.

Within 24 months, the GPU replacement cycle will intensify changes in the unit economics of computing supply, and disputes over pricing and supply volumes under take-or-pay contracts may increase. If the market experiences slowing demand or tighter financing, contract disputes will be exposed more rapidly.

Over the next three years, if AI infrastructure investment undergoes cyclical adjustments, large-scale disputes over long-term contracts may arrive, and an arbitration wave resembling the European LNG price review arbitration wave will repeat. By then, arbitration and regulatory cases will shape the design of a new generation of contract clauses, and today's first movers will bear the first round of lessons.

Conclusion

The wave of AI infrastructure construction is irreversible, but the wave of legal and commercial disputes is also brewing. For investors and corporate decision-makers, understanding this undercurrent will be key to surviving the AI industry cycle.