Neysa plans 9,000-GPU AI cloud deployment at STT GDC Noida

By Rebecca Stone

Data-centre operator STT GDC plans to host up to 9,000 GPUs at its Noida 2 facility in the first phase of a multi-year working relationship with India-based AI cloud company Neysa, it said on Wednesday.

The deployment is expected to go live from the first quarter of next year. It combines Neysa’s GPU infrastructure with its Velocis AI cloud platform and is designed for high-density AI training and inference, the announcement said.

What exactly is arriving at Noida 2?

Neysa will use the Noida 2 facility for the deployment, while STT GDC provides the data-centre setting and infrastructure expertise. The announcement does not identify the GPU supplier, chip generation, rack configuration or power envelope. Nor does it say how many of the maximum 9,000 GPUs will be present when service begins.

“Up to 9,000” is therefore a capacity ceiling, not an installed-unit count or a guaranteed customer allocation. For comparison, later phases carry no disclosed GPU count, so the figure cannot serve as the partnership’s eventual total.

The planned service targets enterprises and startups that want to develop, train and deploy AI applications closer to their users and data. STT GDC also says the site will support production AI workloads in India with “predictable latency and data residency.” No latency target or residency compliance test is provided, and the release names no regulatory regime or certification.

What does STT GDC claim?

Bimal Khandelwal, CEO of STT GDC India, tied the project to the country’s compute requirements:

“India’s AI ambitions will depend not only on access to compute, but on the digital infrastructure required to support increasingly complex and compute-intensive workloads at scale,” he said.

He then framed the commercial arrangement as broader than colocation:

“By bringing together Neysa’s specialised AI capabilities and STT GDC’s infrastructure expertise, this deployment will help enterprises move from experimentation to real-world applications. It also represents the kind of ecosystem collaboration needed to accelerate AI adoption and strengthen India’s position in the global digital economy.”

The statement links compute access with enterprise adoption, but it provides no evidence that phase one has secured customers or production workloads. The concrete commitment disclosed is infrastructure with a 9,000-GPU upper limit; adoption, utilization and revenue remain unquantified.

What is Neysa promising?

Sharad Sanghi, Neysa’s CEO, placed reliability and planned capacity at the centre of the pitch:

“Our customers need more than access to GPUs. They need capacity they can plan around, reliable operations and room to scale as their workloads evolve,” Sanghi said.

“That is why we plan the compute, platform and data centre infrastructure together. Our planned deployment at STT GDC’s Noida data centre is one part of that approach, bringing GPU infrastructure and Velocis, our AI cloud platform, together to support AI workloads in production.”

Planning compute, platform and facility capacity together could reduce coordination friction for customers, although the announcement does not disclose service-level targets, reserved-capacity terms or expansion pricing. Sanghi refers to the “Noida data centre,” rather than Noida 2; the release offers no separate site scope.

When will it be operational?

The only timetable is a first-quarter window next year. The announcement gives no exact commissioning date, construction milestone or customer-onboarding schedule. It also provides no absolute year alongside “next year,” which limits the precision of that timetable.

Several figures that would let outsiders test the plan are also absent:

• GPU supplier, accelerator model and generation.
• Power capacity, cooling design, rack count and network topology.
• Throughput, latency and availability benchmarks for training and inference.
• Contract value, capital commitment, power-purchase terms or revenue-sharing arrangements.
• Named customers, workload volumes, certifications and service-level commitments.

Those omissions do not negate the announced deployment, but they narrow what outsiders can verify. A multi-year working relationship may indicate continuing commercial intent; without minimum spend, hardware orders or customer commitments, the release does not disclose the investment behind phase one.

The next checkpoint is whether the companies can bring phase-one hardware and platform integration into commercial operation in the first quarter next year and produce evidence for their reliability, latency and residency claims. Any wider rollout remains an intention, not a disclosed build plan.