TL;DR · 30-second read
The Short Version
A company called OneQode has agreed to rent space in a large computer facility in Chennai, India, to run the chips that power artificial intelligence. It is OneQode’s first big AI project in India.
The eye-catching part is a figure of more than $6 billion in future sales. That figure assumes the project grows to about eight times its signed size. Only the first slice is signed; the rest is still being negotiated.
Why it matters: AI’s growth runs on deals like this, and the signed part is much smaller than the headline number.
OneQode, a Singapore-datelined operator of AI computing and network infrastructure, has signed an agreement with AdaniConneX for 14 megawatts (MW) of data center capacity in Siruseri SIPCOT, Chennai. OneQode announced the deal on October 27, 2026, in a release distributed on PR Newswire. It is OneQode’s first major AI infrastructure deployment in India.
The two companies say they are continuing discussions on expansion of up to 114 MW. OneQode says that larger footprint could generate more than US$6 billion in total contract value from GPU-as-a-service, which means renting AI processors to customers as a cloud service. No timeline, pricing or customers were disclosed.
Executive Summary
The signed portion of this deal is 14 MW of data center capacity in Chennai. That is the power, space and cooling OneQode needs to house AI servers. OneQode says the site will support AI training (building models), inference (running them to answer requests), enterprise AI and sovereign AI, meaning workloads that must stay under a particular country’s jurisdiction. The Chennai site joins OneQode’s Thailand deployment as part of a wider Asia-Pacific build-out and will connect to the company’s own global carrier network.
The headline economics belong to a different, larger number. The US$6 billion figure is attached to a footprint of up to 114 MW, roughly eight times the signed capacity. That expansion is described as a discussion that depends on growing demand. About 100 MW of the 114 MW envelope, close to nine-tenths of it, is not yet contracted.
For industry readers, the deal is a useful marker of GPU-cloud operators looking to India for capacity and for a location between Europe and Asia. For anyone weighing the US$6 billion figure, the distinction between signed and discussed capacity is the central fact.
Signed Versus Discussed: Where the $6 Billion Sits
The release puts two numbers side by side. The first is 14 MW of capacity, which OneQode describes as secured. The second is 114 MW, which the companies are discussing. OneQode’s Chief Growth Officer, Johnny Blackfield, drew that line directly: “We have secured our initial 14 MW deployment and are continuing discussions with AdaniConneX on additional capacity opportunities of up to 114 MW in India as demand grows.” The more-than-US$6 billion revenue figure is tied to the 114 MW expansion, not to the 14 MW agreement. OneQode has not put a revenue figure on the capacity it actually signed.
The figure is expressed as total contract value (TCV). TCV adds up every dollar a customer contract would pay over its full life, so it says nothing about how much arrives in any single year. Spread evenly, US$6 billion across 114 MW works out to roughly US$53 million of contract value per megawatt. That is simple arithmetic, not a company figure. Without a contract length, a utilization assumption or a pricing basis, it cannot be turned into annual revenue or compared with other operators’ disclosures.
None of this makes the figure wrong. It does make the figure conditional. To reach it, OneQode would need roughly 100 MW of additional capacity agreed, built and filled with paying customers. The parties most directly affected are investors and prospective customers trying to judge OneQode’s scale. For them, 14 MW is the contracted fact and 114 MW is the stated ambition.
A Capacity Lease Is One of Three Commitments
The GPU-as-a-service model stacks three separate commitments. First, the operator leases data center capacity. That is what this agreement covers: AdaniConneX supplies the building, power and cooling. Second, the operator buys or finances the AI hardware. Third, the operator signs customers who rent that compute. Revenue only flows once all three line up.
OneQode has addressed the hardware layer in outline. The deployment builds on its May 2026 collaboration with AMD to deploy AMD Instinct GPUs, and OneQode also plans to use AMD’s Helios rack-scale solution. Rack-scale systems are pre-integrated racks of processors, networking and cooling designed to work as one large machine. OneQode has not said how many GPUs will go into Chennai or how they will be paid for. AI accelerators are typically the most expensive part of such a deployment, so that is a material question. No customers have been named.
The risk is distributed accordingly. For AdaniConneX, a signed 14 MW tenant is booked demand, with potential for more. For OneQode, the larger exposure sits in the hardware and demand layers, which this agreement does not by itself resolve.
Why Chennai: The Latency and Location Argument
OneQode’s case for India rests on geography and cost. The company describes India as sitting between its EMEA (Europe, Middle East and Africa) and Asia-Pacific markets. It argues that as inference becomes more latency-sensitive, a location that can serve both regions has value. Latency is the delay between a request and a response, and it matters for real-time AI applications. OneQode pairs that with its own carrier network and a footprint of more than 30 data centers on five continents, which the Chennai site will connect into.
The argument is coherent, and owning the network between sites is a genuine differentiator for a compute provider. The release does not quantify it, however. It gives no latency figures for Chennai to European or Southeast Asian destinations and no operating-cost comparison to support the “competitive operating costs” Blackfield cites. These are reasonable claims that remain to be substantiated with numbers.
What the Path to 114 MW Depends On
Moving from 14 MW to 114 MW is a physical problem as much as a commercial one. Each additional megawatt of AI capacity needs electricity delivered to the site, data halls built and fitted out, and cooling sized for dense GPU racks. The companies have published no schedule for the initial 14 MW coming online, no phasing for the expansion, and no detail on how the campus would supply power at the larger scale.
The release frames expansion as following demand: discussions continue “as demand grows.” That is a prudent sequencing for an operator that does not want to lease capacity it cannot fill. It also means the eightfold path is an option on future demand rather than a build plan. Readers should treat the 114 MW figure, and the US$6 billion attached to it, accordingly until either company discloses contracted expansion.
Background
OneQode is one of a growing group of specialist GPU-cloud operators, sometimes called neoclouds. These companies rent AI computing capacity to model developers and enterprises, in competition with the large public clouds. OneQode emphasizes its own carrier-grade network and a footprint of more than 30 data centers on five continents. Earlier in 2026 it announced a collaboration with AMD to deploy AMD Instinct GPUs and plans for AMD’s Helios rack-scale systems. It has also announced a sovereign AI factory alliance with Hitachi Vantara and Cylix, and it has an AI deployment in Thailand.
Operators like OneQode generally do not build their own data centers in every market. They lease power, space and cooling from colocation providers such as AdaniConneX, fill that capacity with AI hardware, and sell the compute as a service. That makes the capacity agreement the first step in a deployment rather than the whole of it. Source: OneQode Signs 14 MW AI Capacity Agreement with AdaniConneX, with a Path to 114 MW — OneQode’s October 27, 2026 announcement of its Chennai data center capacity agreement with AdaniConneX.Sources