In AI infrastructure the binding constraint is no longer chips or capital, it is megawatts on the meter, and the utility interconnection queue can stall a new campus for years. That is the lens for a deal struck this week in North Dakota: an offtake agreement covering the planned Nakota AI Data Campus, a roughly 1 GW build designed around independent, behind-the-meter natural gas generation rather than a grid tie. The counterparties are Nidar Infrastructure Limited, parent of India's Yotta, on the demand side, and the private operator Tachyon9, whose public vehicle and investable detail is Nixxy, Inc. (NASDAQ: NIXX); per the companies the offtake reflects the terms contemplated under their earlier MOU, contracting about $156 million in annual recurring revenue on the initial 100 MW phase with a right of first offer toward up to $1.5 billion a year at full buildout.
Why power, not compute, is the binding constraint
To read this deal correctly you have to understand where the bottleneck in AI infrastructure actually lives right now. It is not GPUs and it is not capital. It is megawatts on the meter. A large training or inference cluster draws power on a scale that most utility grids cannot interconnect quickly, and the queue to get a new load approved can run multiple years. That interconnection wait, not chip supply, is what is gating new campus construction.
Behind-the-meter generation is the mechanism that routes around that queue. Instead of requesting a grid interconnection and waiting for the utility to build out transmission, the operator generates electricity on-site and consumes it on-site. The Nakota design does this with dedicated natural gas turbines drawing on the Williston Basin, one of North America's larger gas-producing regions. The companies frame fuel access as the project's core strategic asset, and in this sector that framing is defensible: a campus that controls its own fuel and generation is decoupled from the interconnection backlog that constrains its grid-tied peers.
What the offtake agreement technically changes
An offtake agreement is a demand-side commitment. It is the counterpart to the supply-side build. For a capital-intensive infrastructure project, a contracted offtake is what converts a development plan into something financeable, because lenders underwrite against contracted revenue rather than speculative occupancy.
Per the companies, the agreement provides for approximately $156 million in annual recurring revenue from the initial 100 MW phase. It also grants Nidar and Yotta a right of first offer on the remaining campus capacity, which management describes as a pathway toward nearly 1 GW of development and up to $1.5 billion in annual revenue potential at full buildout. Those two figures are internally consistent: 1 GW is roughly ten times the 100 MW first phase, and ten times $156 million is about $1.5 billion. The phase figure is contracted under this agreement. The full-buildout figure is a potential ceiling, attributed to management and dependent on financing and on the right of first offer being exercised.
Shahal Khan, Chairman and Chief Executive Officer of Tachyon9, said the structure "provides revenue visibility, supports project financing, and significantly de-risks the path" toward the platform. Read in domain terms, that is a claim about bankability, not about realized cash flow.
The campus specification and the asset contribution
The Nakota AI Data Campus is planned across approximately 620 acres in Williams County, North Dakota, with a targeted buildout of up to 1 GW. As described by the companies, it is engineered as Tier III-capable, liquid-cooled, with integrated carbon capture and sequestration, multiple diverse fiber entrances, and support for current and next-generation NVIDIA GPU architectures. Liquid cooling and a CCS tie-in are the relevant tells: both are what you specify when you are designing for high-density GPU racks and for a gas-fired power island rather than retrofitting a legacy facility.
Within the NIXX transaction, Tachyon9 is the primary asset and revenue contributor, supplying approximately $64 million in equipment, land-option rights, and a signed letter of intent for the entire 1 GW development.
Who the customer is
The credibility of the demand side rests on Yotta. Nidar, backed by the Hiranandani Group, is the parent and credit-support entity, with Darshan Hiranandani as Chairman of Nidar and co-founder of Yotta, and Sunil Gupta as CEO of Yotta. By the companies' account, Yotta holds an estimated 60 to 70 percent of India's deployed GPU capacity, runs three operational campuses in Navi Mumbai, Gujarat, and Greater Noida with a fourth planned in Telangana, and operates Shakti Cloud, India's sovereign AI platform built with NVIDIA. In February 2026 it announced a US$2 billion-plus investment to deploy 20,736 liquid-cooled NVIDIA Blackwell Ultra GPUs at Greater Noida, a four-year NVIDIA DGX Cloud engagement valued at over US$1 billion, and an allocation of more than 10,000 GPUs to India's IndiaAI Mission.
tldr
The investable detail trades through NIXX, but the substance is Tachyon9's energy-first build. The offtake contracts the 100 MW phase at roughly $156 million in annual recurring revenue and opens a right of first offer toward nearly 1 GW and up to $1.5 billion in potential annual revenue. The campus is planned, not built, and the larger numbers are conditional on financing and buildout. What makes the design coherent in this sector is the behind-the-meter power, which is the constraint everyone else is queued behind.