AI Infrastructure and Applications
For the companies building AI infrastructure, the companies building on it and the sponsors of compute infrastructure funds.
Three layers that decide each other.
The site is real estate, energy and project finance. The compute is a services business sold on capacity and term. The applications are software companies with hard questions about data and ownership.
Most lawyers cover one layer. We cover all three, because the power arrangement determines whether the site can be financed and the compute contracts become the lender’s collateral. A lawyer who sees one layer negotiates the wrong things well.
Many clients start with one matter, like a neocloud’s first enterprise services agreement or a seed round for an agent company.
Across all three layers.
Buyer-side acquisition of a two-hundred-megawatt data-center campus at $55 million, with $47 million of seller financing and the lien subordination that kept the site financeable for a construction lender.
Binding anchor offtake term sheet for seven megawatts of AI compute capacity in a greenfield data center in Southeast Asia, on behalf of the anchor customer.
Master services agreement and statements of work for a layer-one blockchain foundation commissioning a native agentic-payments platform, originated through a partnership with a major cloud provider.
The full list for this practice is on the Selected Work page.
Infrastructure: the site, the power, the financing.
We act for developers on the acquisition of sites and the arrangements that make them run. Every document at this layer is read later by a lender, so we negotiate with the capital structure in view: who owns the generation, what the gas costs and where the liens sit.
- Site purchase agreements, including land, mineral and gas interests
- Seller financing: notes, security packages, subordination and standstill
- Gas supply and power purchase agreements
- Behind-the-meter generation joint ventures
- Colocation and hosting agreements
- Binding term sheets and exclusive options on development sites
Compute: offtake, capacity and the supply chain.
We act for providers, the operators and neoclouds selling capacity, and for anchor customers contracting for capacity at new sites. For a provider the contract book is the collateral, so commitments, payment timing, assignment and termination are drafted so the next financing can be raised against that revenue.
- GPU compute master services agreements: reserved instances, service levels, order forms
- Offtake and take-or-pay commitments; anchor-tenancy term sheets
- Marketplace, broker and resale arrangements
- Hardware procurement across servers, memory, interconnect and power services
- Export-control and denied-party provisions
- Availability and support service levels, and the credit regimes behind them
Applications: the companies building on top.
We represent the companies building with the compute, from agent platforms and robotics to decentralized AI networks, on their financings, commercial agreements and the data and IP questions that determine what they own. More of the work at this layer now sits where AI meets crypto infrastructure, in agentic payments and in platform builds, where we act for the protocol foundations that commission them.
- Formation, seed and convertible financings, board and governance work
- Master services agreements and statements of work for platform builds
- Data-architecture and intellectual-property review
- Ownership of outputs, integrations and training data
- Open-source licensing decisions
- Platform terms of service
Compute infrastructure funds.
Compute infrastructure is increasingly held in funds, and we form them for sponsors around an asset whose useful life is shorter than a conventional fund term. That work, including a compute infrastructure fund raising up to $900 million, is under Fund Formation and Private Securities.
How this work usually begins.
A term sheet for a site, often with seller financing attached, that has to be signed quickly and financed later.
A first enterprise master services agreement, or a marketplace’s paper, landing on a provider that has been selling on consumption.
A miner converting capacity to AI hosting, whose offtake, colocation and customer contracts no longer fit the entity that was built for mining.
A sponsor raising a first compute fund, with a second already in mind.
A platform build commissioned by a protocol, a foundation or a cloud partner, where the ownership terms have to be right before the code is written.
If yours starts somewhere else, it is still worth a conversation.
What the matters taught us.
Seller Financing on a Data-Center Site Will Block Your Construction Lender
Publishing nextWho Owns What the Agent Builds
In preparationWhat each piece argues is on the Writing page.
Bring us the term sheet, not the signed agreement.
Most of what goes wrong in these deals goes wrong before the definitive documents exist. The earlier the conversation, the more there is to work with.