India aims to accelerate its own large-scale Artificial Intelligence capabilities, and in that context, infrastructure weighs just as much as the models. In this environment, AMD and Tata Consultancy Services (TCS) have announced the expansion of their strategic partnership to co-design and deploy in India a “rack-scale” AI architecture based on the AMD “Helios” platform, with a data center roadmap prepared for up to 200 MW.
The agreement, announced on February 16, 2026, positions HyperVault AI Data Center Limited (a TCS subsidiary) as the vehicle for design and implementation. The stated goal is to support the country’s national AI initiatives and what both companies call “sovereign AI factories”. This phrase has become common among governments and large corporations: training and inference capacity under local control, with architecture and operations tailored to data residency, compliance, and security requirements.
What is “Helios” and why does the “rack-scale” approach matter for operations
Beyond the headline, “Helios” signifies a shift in mindset for AI deployment in data centers: instead of building clusters from loose components, the entire rack (power, networking, accelerators, CPUs, cooling, cabling, and software) becomes the “building block” for design, validation, and operation.
AMD presents “Helios” as an open, rack-scale platform for training and inference, supported by its ROCm ecosystem. In earlier communications, the company described “Helios” as a “blueprint” for yotta-scale infrastructure and claimed it could deliver up to 3 exaFLOPS of AI performance in a single rack (always based on its own measurement parameters and configurations).
Practically, the value for systems and platform teams often lies in less glamorous aspects: deployment time, repeatability, failure control, and operational efficiency. The TCS partnership emphasizes this: a platform designed to improve efficiency, reduce time-to-deployment, and enable real industry impact.
The technical “stack”: MI455X, EPYC “Venice”, Thinking “Vulcano”, and ROCm
The architecture announced for India relies on AMD components end-to-end:
- AMD Instinct MI455X GPUs as the main accelerators.
- AMD EPYC “Venice” CPUs of the latest generation.
- AMD Pensando “Vulcano” NICs for scalable connectivity.
- And the ROCm ecosystem as the open software foundation.
For a technical audience, the subtext is clear: AMD aims to establish a full stack (compute + networking + software) capable of competing in large-scale deployments, especially where clients and governments seek alternatives to more closed platforms. Several Indian media outlets have interpreted this move as an effort to reinforce an alternative pathway to NVIDIA’s dominance in AI infrastructure, leveraging open design and partnerships in integration and services.
200 MW: when the “blueprint” moves beyond the lab
The figure of 200 MW is significant. In the data center market, talking about hundreds of megawatts involves:
- Electrical planning at campus scale,
- Supply agreements and energy resilience,
- Thermal and cooling engineering for high density,
- And a network strategy prepared for massive east-west traffic typical of distributed training.
AMD and TCS frame this capacity as a “blueprint” (AI-ready data center blueprint) and state they will work with hyperscalers and AI companies to accelerate data center construction in India.
At the same time, the announcement serves a political-industrial purpose: to reinforce that India will not only consume AI but can produce and operate it at scale, with infrastructure prepared for sovereign and enterprise workloads.
The role of TCS and the HyperVault precedent: from software to data center engineering
TCS is not approaching this from scratch. The company launched HyperVault in 2025 with the declared ambition to build AI-ready and secure infrastructure at gigawatt scale for hyperscalers and large firms.
Additionally, Reuters reported in 2025 about discussions related to a sovereign AI data center initiative linked to HyperVault, amid a surge in global demand for capacity driven by the AI cycle.
This background helps explain why this announcement is about more than just “hardware”: the partnership relies on TCS’s ability to execute complex data center engineering projects, operations, and services, alongside AMD’s strategy to promote open, rack-scale architectures as a foundation for massive deployments.
What Lisa Su and K. Krithivasan say: from pilot to industrialization
AMD CEO Lisa Su frames the move as a market transition: AI “is moving from pilots to large-scale deployments,” requiring “a new infrastructure blueprint.” In this context, “Helios” is presented as a long-term investment in performance, efficiency, and flexibility.
TCS CEO K. Krithivasan highlights that the agreement lays the groundwork for the first “Helios-powered” deployment in India and emphasizes the combination of AI, connectivity, sustainable energy, and advanced data center engineering. The message to the local market is clear: it’s not just about consuming computing power, but about building reference infrastructure for global enterprises and native AI companies.
Key points for the sector to watch in 2026
The announcement sets a starting point, but system administrators, architects, and developers should consider three critical questions in the coming months:
- Actual platform availability and timeline (hardware, validated racks, supply, and deployment).
- Operational maturity of ROCm in large-scale enterprise deployments (tools, observability, compatibility, support).
- Standardization of “rack-scale”: how much of this blueprint translates into repeatable designs and how much remains client-specific integration.
If the Indian market manages to turn “Helios” into an industrial foundation—beyond just an announcement—the movement could serve as a case study on how to build sovereign AI factories without relying on a single technology stack.
FAQs
What does an “rack-scale” AI architecture like Helios mean compared to a traditional cluster?
It involves treating the rack as a design and operation unit: compute, networking, power, and cooling are validated together to accelerate deployments and standardize operations.
What components are included in the announced Helios stack for India?
According to AMD and TCS: AMD Instinct MI455X GPUs, AMD EPYC “Venice” CPUs, AMD Pensando “Vulcano” NICs, and the ROCm software ecosystem.
What does a “blueprint” for AI-ready infrastructure up to 200 MW entail for a data center?
A design prepared to scale campus-wide, with high-density electrical and cooling planning, aimed at enterprise training and inference.
What role does HyperVault play within the AMD and TCS partnership?
HyperVault, a TCS subsidiary, acts as the infrastructure arm for co-design and deployment in India, aligned with TCS’s strategy to build large-scale AI data centers.
via: AMD