Indian AI cloud provider Neysa has deployed a cluster of 8,192 AMD Instinct MI350X graphics processing units (GPUs) in Mumbai, expanding the country’s dedicated computing infrastructure for artificial intelligence. The cluster is operational on Neysa’s Tytan bare-metal platform and is designed to support large language model training, fine-tuning, generative AI applications and high-throughput inference workloads.
The deployment, announced on November 18, 2026, is significant because global customers have already contracted the computing capacity for AI deployments in India. The project also expands the choice of GPU hardware available to Indian businesses, as Neysa pursues a silicon-agnostic strategy that includes AMD and Nvidia accelerators. The development reflects a broader shift in India’s AI market: demand is moving beyond access to individual GPUs towards large, dedicated clusters that can reliably train and operate advanced AI models.
Key takeaways
- Neysa has deployed 8,192 AMD Instinct MI350X GPUs in Mumbai.
- The cluster is operational and was brought online in a single phase.
- It runs on Neysa’s Tytan dedicated bare-metal AI infrastructure platform.
- Global customers have already contracted the capacity for AI deployments in India.
- The cluster uses AMD EPYC processors and the AMD ROCm software ecosystem.
- Customers can access the infrastructure through bare-metal, Slurm and Kubernetes configurations.
- The deployment supports Neysa’s strategy of offering customers a choice of accelerator hardware rather than relying on a single GPU supplier.
- The expansion comes as Indian enterprises and international companies increasingly look to India for AI computing capacity.
Neysa deploys 8,192 AMD Instinct MI350X GPUs in Mumbai
The new cluster represents a major expansion of dedicated AI computing capacity available through Neysa.
The company has deployed 8,192 AMD Instinct MI350X accelerators on its Tytan platform. The infrastructure is designed for organisations that require dedicated computing resources for demanding AI workloads rather than relying entirely on shared cloud environments.
According to Neysa’s announcement, the deployment is already operational, and global customers have contracted the capacity for AI projects in India.
The cluster was brought online in a single phase in Mumbai. It combines AMD’s Instinct MI350X GPUs with AMD EPYC processors and the ROCm open software ecosystem.
The three components perform different roles:
| Component | Role in the infrastructure |
|---|---|
| AMD Instinct MI350X | Accelerates AI computing workloads |
| AMD EPYC processors | Provide central processing capabilities for the system |
| AMD ROCm | Provides the software ecosystem for developing and running GPU applications |
| Neysa Tytan | Provides dedicated infrastructure and cluster operations |
The deployment is designed to support large-scale computing requirements, including model training, fine-tuning and inference.
However, Neysa has not publicly disclosed the total investment, power capacity, detailed networking specifications or the identities of customers that have contracted the cluster.
What are AMD Instinct MI350X GPUs?
The AMD Instinct MI350X is a data-centre accelerator designed for artificial intelligence and high-performance computing.
Unlike conventional graphics cards used in gaming computers, data-centre GPUs are built to handle large amounts of parallel computation. They are used to train AI models, process large datasets and run models that generate responses to user requests.
Modern AI models require substantial computing resources because they process large quantities of numerical operations across billions of parameters.
GPUs are particularly useful for these workloads because they can perform many calculations simultaneously.
The MI350X belongs to AMD’s Instinct accelerator family, which competes in the market for advanced AI computing hardware.
The deployment is also important for AMD because it places its accelerators inside a large operational AI cluster in India, providing another hardware option for enterprises and AI developers.
Neysa’s use of AMD hardware does not mean the company is abandoning Nvidia. Instead, it is expanding the range of accelerator architectures available through its infrastructure platform.
Why 8,192 GPUs matter for AI development
The number of GPUs in a cluster matters because advanced AI workloads often require thousands of accelerators working together.
Training a large language model involves processing enormous datasets and adjusting model parameters through repeated calculations. The work can be distributed across multiple GPUs to reduce the time required.
Large clusters can also support several customers or workloads, depending on how the infrastructure is configured and allocated.
The 8,192-GPU deployment gives Neysa the ability to offer a substantial pool of dedicated computing resources for customers that need to train, fine-tune or operate AI models at scale.
There are three major workloads the cluster is designed to support.
1. Large language model training
Training is the process through which an AI model learns patterns from large datasets.
A large language model processes text and other information during training and adjusts its internal parameters to improve its ability to generate useful responses.
This process can require enormous computing capacity, particularly when models contain billions of parameters or are trained on extensive datasets.
A large GPU cluster allows the training workload to be distributed across multiple accelerators.
2. Fine-tuning AI models
Fine-tuning involves adapting an existing AI model for a particular purpose.
For example, a company may take a general-purpose language model and further train it using specialised information related to banking, healthcare, manufacturing or customer service.
Fine-tuning can help organisations build models that understand their industry, terminology and business requirements without necessarily training a foundation model from scratch.
Dedicated GPU infrastructure can help businesses perform this work with greater control over computing resources and deployment environments.
3. AI inference
Inference is the process of running a trained AI model to produce an output.
Whenever a user asks a chatbot a question, requests an image or interacts with an AI-powered application, the underlying model performs inference.
For businesses operating AI applications at scale, inference can become a major computing expense.
The Neysa cluster is designed to support high-throughput inference, allowing organisations to run large numbers of AI requests using dedicated infrastructure.
What is Neysa Tytan?
Neysa Tytan is the company’s dedicated bare-metal AI infrastructure offering.
Bare-metal infrastructure gives customers direct access to physical computing resources without requiring them to operate entirely through a conventional virtualised cloud environment.
This approach can be useful for organisations that need greater control over hardware configuration, networking, storage and software deployment.
For large AI workloads, customers may want to configure the infrastructure according to their specific model architecture and computing requirements.
Neysa says Tytan allows customers to choose different deployment configurations:
| Deployment option | What it does |
|---|---|
| Bare metal | Provides direct access to dedicated physical infrastructure |
| Bare metal with Slurm | Adds workload scheduling and management for computing clusters |
| Bare metal with Kubernetes | Supports container orchestration and deployment of applications |
These options address different operating requirements.
Slurm is a workload manager widely used in high-performance computing environments. It helps allocate computing resources and schedule jobs across a cluster.
Kubernetes is a container orchestration platform used to deploy, manage and scale applications.
By supporting both approaches, Neysa is targeting organisations with different infrastructure and engineering requirements.
Why Neysa is following a silicon-agnostic strategy
One of the most important parts of the announcement is Neysa’s emphasis on a silicon-agnostic infrastructure strategy.
Silicon refers to the semiconductor chips that power computing systems. In the AI infrastructure market, Nvidia has established a strong position through its GPUs and software ecosystem, while AMD is expanding its presence with its Instinct accelerator family.
A silicon-agnostic approach means that an infrastructure provider does not depend entirely on one chip manufacturer.
Instead, customers can select accelerator hardware based on performance requirements, software compatibility, availability, pricing and operating needs.
Neysa’s strategy allows customers to choose their preferred accelerator architecture, network design, storage platform and deployment model.
This flexibility could become important as AI infrastructure costs rise and organisations seek alternatives to a single hardware ecosystem.
However, offering multiple GPU platforms also creates operational challenges. Different accelerators can require different software tools, development frameworks, optimisation techniques and engineering support.
Neysa will need to maintain reliable infrastructure operations across these different environments.
AMD expands its presence in India’s AI computing market
The deployment also gives AMD a larger presence in India’s expanding AI infrastructure market.
The global AI computing industry has been heavily influenced by Nvidia’s GPUs and CUDA software platform. However, customers and cloud providers are increasingly interested in having alternative accelerator options.
AMD is attempting to compete through its Instinct GPU portfolio and ROCm software ecosystem.
The ROCm platform provides tools and libraries that developers can use to build and run GPU-accelerated applications on AMD hardware.
For AI developers, software compatibility is a critical consideration. A GPU’s computing capabilities are only part of the equation; developers also need frameworks, libraries and tools that allow them to use the hardware efficiently.
The Neysa deployment provides an opportunity to make AMD-based computing capacity available to businesses operating in India.
For AMD, partnerships with infrastructure providers can help expand access to its accelerators without requiring every enterprise to purchase and operate its own data-centre hardware.
For Neysa, the collaboration expands its ability to offer different computing architectures to customers.
Why global AI companies are looking at India for computing capacity
The deployment comes at a time when India is attracting interest from international companies looking for additional locations to run AI workloads.
AI computing infrastructure requires more than GPUs. It also depends on data-centre facilities, electricity, cooling systems, networking, storage, software and skilled engineering teams.
As AI adoption expands, companies need access to infrastructure that can support model training and inference reliably.
India offers a large enterprise market and an expanding technology services ecosystem. International businesses can also use computing infrastructure located in the country to support their regional or India-based operations.
Neysa says global customers have already contracted the capacity of its new cluster for AI deployments in India.
This is an important commercial detail because it indicates that the deployment is not simply an infrastructure expansion waiting for demand. The company says customers have already committed to using the capacity.
However, the company has not disclosed customer names, contract values, utilisation rates or the revenue contribution expected from this particular cluster.
Those details will be important in evaluating the financial performance of the deployment.
Neysa’s wider expansion plans
The 8,192-GPU deployment is part of a broader expansion strategy.
In an October 2026 interview with The Economic Times, Neysa co-founder and CEO Sharad Sanghi said the company had around 12,500 live GPUs and expected to increase that figure beyond 20,000 before March 2027.
He also discussed plans to add more Nvidia and AMD computing capacity, including future-generation accelerator systems.
The company has been exploring the possibility of building its own data centres. Sanghi told the publication that Neysa was looking at land and considering developing its own facilities, although the process could take approximately two years because of construction and approval requirements.
For the near term, the company expects to continue using rented data-centre space.
Neysa’s expansion strategy reflects the capital-intensive nature of AI infrastructure.
Buying GPUs is only one part of the investment. Companies must also secure data-centre capacity, power, cooling, networking equipment and the operational expertise needed to run large clusters.
The ability to secure customers through long-term contracts can help infrastructure providers plan their investments and reduce the risk of deploying expensive hardware without sufficient demand.
How the deployment fits into India’s AI infrastructure race
India’s AI ecosystem is expanding beyond software startups and application development.
The country is increasingly focused on the infrastructure required to build, train and operate AI models.
This includes GPU clusters, cloud platforms, data centres, networking infrastructure and domestic AI computing capacity.
The growth of these facilities could help Indian businesses access computing resources without having to build their own large-scale infrastructure.
It could also support research institutions, startups and international companies that want to run AI workloads in India.
Neysa’s deployment is one part of this wider shift.
The availability of AMD accelerators alongside Nvidia-based infrastructure could also encourage greater competition among hardware providers and cloud companies.
For enterprises, more choice could create opportunities to compare computing costs, performance and software compatibility.
However, the long-term competitiveness of India’s AI infrastructure market will depend on several factors beyond the number of GPUs installed.
Reliable power supply, data-centre efficiency, network connectivity, hardware utilisation, financing costs and customer demand will all influence whether these investments generate sustainable returns.
The bigger picture
The deployment of 8,192 AMD GPUs in Mumbai highlights how the AI infrastructure market is evolving.
The first phase of the AI boom was largely associated with foundation models, chatbots and generative AI applications. The next phase requires the computing infrastructure to train, fine-tune and operate these systems at commercial scale.
That is creating opportunities for companies that can provide dedicated GPU clusters, data-centre capacity and specialised AI cloud services.
Neysa is positioning itself as one of these infrastructure providers. Its partnership with AMD expands the hardware choices available to customers while strengthening its presence in India’s AI computing market.
The deployment also demonstrates that India is becoming a location for large-scale AI workloads serving both domestic and international customers.
However, the number of GPUs alone does not determine the success of an AI infrastructure business. Customer utilisation, long-term contracts, operating costs, power availability and the ability to deliver reliable performance will determine whether these investments become profitable.
Looking ahead
Neysa’s next challenge will be to scale its infrastructure while maintaining high utilisation and securing long-term customer commitments. The company’s stated target of expanding its live GPU base beyond 20,000 before March 2027, along with its plans to introduce additional AMD and Nvidia accelerators, will be important milestones to watch. Its ability to secure data-centre capacity and manage the cost of large hardware deployments will also influence the pace of expansion.
For AMD, the partnership offers another route to increase the adoption of its Instinct GPUs in India’s AI infrastructure market. For Indian enterprises and global AI developers, the deployment adds another source of dedicated computing capacity. The broader question is whether India can build a competitive AI infrastructure ecosystem that combines advanced chips, reliable data centres, software capabilities and commercially sustainable demand.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.



