India’s push for sovereign AI is picking up pace, but the compute layer remains tied to imported GPUs and global supply chains. Dell, HPE, Acer and a set of local players are now racing to localise assembly. Owning AI is no longer just about models. It comes down to who controls the machines and where they run.
Building GPUs in India, however, is a long-term play. In the meantime, some companies are taking a more practical middle path.
Altos Computing, a subsidiary of the Acer Group, has launched a new portfolio of Made-in-India AI servers, stepping into a market that is shifting from software-led narratives to control over compute. The announcement came at a launch event in New Delhi, where the company unveiled the Altos BrainSphere R370 F7 AI Series server, designed for high-performance workloads such as model training and inference.
The system supports dense NVIDIA and Intel GPU configurations, along with scalable memory and improved thermal efficiency. It targets enterprises, research labs, government bodies and hyperscale data centres that are tackling limited access to compute.
The event brought together Harish Kohli, President and Managing Director of Acer India, Jackie Lee, President of Altos Computing, Sudhir Goel, Chief Business Officer at Acer India, and Priya Krishnamurthy, Director at Altos Computing. Representatives from NVIDIA, Intel, MeitY, and other industry stakeholders were also present.
In an announcement, NVIDIA’s Vishal Dhupar frames the partnership, saying, “AI innovation depends on powerful accelerated computing infrastructure.” He adds that the collaboration combines NVIDIA’s accelerated computing with Altos’ servers to help organisations train and deploy models efficiently.
But the idea of Made-in-India here is still incomplete.
The 50% Rule Continues
In an interaction with AIM, Kohli clarifies that 51% of the server components by value are sourced locally, including components like the motherboard, memory, and storage. The rest, especially the components that matter most for AI, still come from outside.
This points to a deeper limitation in the entire Make-in-India push for AI infrastructure. The chassis and the assembly can be local, but control over compute still sits elsewhere.
Kohli acknowledges this directly. India’s hardware ecosystem is “still evolving”, with early progress in memory and storage, but continued dependence on global supply chains for processors and accelerators.
3,000 Servers: The First Signal of Demand
Kohli exclusively told AIM that the company is building capacity for 5,000 servers in India and expects to ship over 3,000 units in the next year. That number may not be large in global terms, but it is significant in the Indian context.
There are early customers as well. Altos has already run pilot deployments with educational institutions and private organisations, using sample units to test real-world workloads.
The more notable shift Kohli points to is architecture. India’s AI adoption is beginning to move away from a cloud-first model.
He linked this to data sovereignty and privacy concerns. Organisations are becoming more cautious about where their data sits and how it is processed. Sensitive workloads, especially in sectors like research, engineering, and public systems, are pushing companies to reconsider complete reliance on external cloud environments.
What is coming next is a hybrid model.
Critical workloads stay on-premises. Less sensitive applications stay on the cloud.
This is not a rejection of cloud. It is a rebalancing. And it aligns directly with companies like Altos. On-prem infrastructure only makes sense when enterprises are willing to own and operate compute.
Cloud looks attractive at the beginning because of low upfront investment. But as workloads scale, Kohli points out, continuous AI usage drives up cloud GPU expenses. Over time, enterprises begin to see on-prem infrastructure as more cost-effective.
This is not a new argument globally. It is now starting to become more visible in India.
The challenge, as Kohli notes, is capability. Running on-prem AI systems requires expertise that many organisations is not ready for. This is why the hybrid model is becoming the default.
It is About Supply Chain Control
What Altos is building is not full-stack independence. It is partial localisation with global dependencies. That distinction matters because the true bottleneck in AI is not servers. It is GPUs.
If the critical hardware is still globally dependent, why manufacture locally at all?
Kohli’s answer is largely operational. Local manufacturing reduces delays and improves availability, cutting down deployment timelines at a time when compute shortages are slowing AI adoption.
He points to a practical issue. Fully imported systems are tied to global supply chains that are already under strain. Local assembly allows faster integration and delivery.
In that sense, the Made-in-India argument becomes operational rather than ideological. It is about getting machines into racks faster, not building every component entirely from scratch.
And in a market where enterprises are waiting months for GPU-backed systems, that difference can decide deals.
Sovereign AI is the underlying narrative
Altos is aligning itself closely with India’s sovereign AI ambitions.
Kohli describes the servers as enabling organisations to “build and manage their AI workloads within India”, across both private cloud and on-prem setups. The company is also working with cloud providers and software partners to offer a full stack, not just hardware.
This is important.
Selling servers alone does not solve the AI problem. Enterprises need software layers, orchestration, and deployment capabilities. Altos is trying to place itself within that broader ecosystem.
The AI Wall at the event reinforced this approach. Use cases ranged from predictive analytics to intelligent video systems and manufacturing automation.
Altos is entering a crowded market. Dell, HPE, and Supermicro already have a strong hold on enterprise AI infrastructure globally.
Pricing Will Follow Scale, Not Policy
There is no immediate cost advantage. Kohli is clear that imported systems still benefit from scale. Larger global volumes make them cheaper today.
The bet is on what happens over time.
He draws a parallel with Acer’s earlier push in local manufacturing, where prices initially lagged. As volumes increased and supply chains matured, locally made products reached parity and, in some cases, became more competitive.
Altos is betting on a similar curve. As adoption grows and more components are sourced locally, pricing will improve. Until then, the value proposition is not cheaper hardware, but faster access and a more resilient supply.
Kohli’s positioning is straightforward. Altos wants to be among the top three to four players in this space, with local manufacturing and India-specific solutions as its edge.
The ecosystem will take time to come together. Dependencies will not disappear quickly. Progress will be gradual.
That is the more grounded version of India’s AI infrastructure story.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.