AMD to roll out Helios GPU platform in India in H2 2026

An AMD official said global customer deployments of its Helios GPU-based high-performance computing platform are expected to start in the second half of 2026, including in India.

According to AMD senior director for data center GPU product marketing Mahesh Balasubramanian, one Helios rack links 72 MI455X accelerators and is rated at up to 2.9 exaflops of FP4 compute per rack.

The comments come as demand for AI hardware rises and Nvidia holds more than 80% of the GPU market, while Nvidia said its Vera Rubin POD can reach 3.6 exaflops.

AMD has also partnered with Tata Consultancy Services to co-develop a Helios-based rack-scale AI data center blueprint for India that supports up to 200MW of capacity.

🔗 Source: The Economic Times

🧠 Food for thought

Implications, context, and why it matters.

### The rack is the new server, and memory is the new battlefield

The word “platform” misses the scale of the shift. AMD positions Helios for rack-scale, liquid-cooled, fully integrated systems instead of individual PCIe (Peripheral Component Interconnect Express) cards 1.
One rack can pull more than 120 kilowatts, which turns AI buildouts into a civil engineering problem 1.
AMD’s challenge to Nvidia leans on memory capacity plus open standards, not raw speed alone.
The MI455X GPU includes 432 GB of High Bandwidth Memory (HBM4). That is close to 50% above Nvidia’s 288 GB, which helps keep larger AI models inside a single node 1.
Deployment is expected in the second half of 2026, though the window is contested. One report says Nvidia could speed up Rubin while AMD might slip, which could push volume ramp into 2027, and AMD has disputed those claims publicly 2.

### Sovereign AI initiatives are reshaping the global tech supply chain

The Tata Consultancy Services (TCS) partnership sets out a plan for “sovereign AI factories,” tied to a wider push for domestic AI capability 3.
AMD’s standards-based approach offers a route outside Nvidia’s proprietary stack, which may suit governments planning national-scale infrastructure over the long term 1.
The agreement also signals a rising role for hyperscale infrastructure integrators.
TCS’s HyperVault subsidiary has a combined commitment of up to Rs 18,000 crore (about $2.1 billion) from TCS plus TPG (a global private equity firm). It targets gigawatt-scale capacity, and it has held talks with multiple large players, including Nvidia and Google 4.
Even with an AMD partnership, large AI data center buyers may still spread purchases across vendors to reduce supply risk and strengthen pricing leverage 1.