AI Is More Than GPUs, Leaders Say, as India Bets on Full-Stack Compute

News

1. Why Sovereign AI Demands More Than Just Data Centers and Chips

2. How India Could Leapfrog to AI-Native Factories

3. AI Is More Than GPUs, Leaders Say, as India Bets on Full-Stack Compute

4. How Hospitals Are Turning to AI to Win the Race Against Time

5. Blackstone Leads $600 Million Funding Round in AI Cloud Startup Neysa

6. When AI Ambitions Meet the Geopolitics of Compute

At the India AI Impact Summit 2026 in New Delhi, industry leaders argued that the conversation around AI hardware has become too narrowly focused on GPUs, even as networks, memory, energy, and edge devices increasingly determine performance and economics.

Jason Oxman, president and chief executive of the Information Technology Industry Council, said the debate ultimately comes down to confidence in the ecosystem. “It’s all down to trust. We have to trust in the ecosystem, the tech stack, if you will, in order to make sure that we’re getting the products and services that we want and they’re doing the things that we want them to do,” he said.

Oxman added that trust needs to go beyond national borders. “Trust crosses borders, data crosses borders,” he said, warning that sovereignty should not be interpreted as technological isolation. Global standards, he argued, are essential to allow companies to sell and scale products internationally. 

Executives from hardware companies said AI infrastructure should be viewed as a system, not just a single part. Bhavna Agarwal, senior vice president and managing director at Hewlett Packard Enterprise India, said the industry often focuses only on GPUs in AI talks, but that misses the bigger picture.

“When you look at AI as a journey, that’s the way it should be and not just one of the components,” she said. Comparing AI infrastructure to city planning, she said that GPUs are like tall buildings, but “you also need efficient city planning.”

Agarwal said organizations are increasingly moving toward AI native architectures rather than trying to force-fit AI into old IT systems. She added that modular design and planned scalability help control costs and prevent overprovisioning.

Sustainability is also becoming central to infrastructure decisions. Agarwal said energy efficiency must be built in from the start. “Sustainability must be there as a principle on day zero and not as an afterthought,” she said.

Chip designers pointed to bottlenecks beyond compute. Naveen, country head of Marvell Technology in India, said that while GPUs remain important, networking and memory increasingly limit performance.

“We thought that AI means compute,” he said. “But along with compute, you need the whole network.” He likened high-end processors to a sports car stuck in traffic, saying computing without sufficient interconnect and memory bandwidth reduces efficiency.

Telecom executives agreed, saying AI workloads are creating new demands for connectivity providers.

Abhishek Biswal, chief business officer for digital services at Airtel, said operators must evolve beyond being “passive carriers.” He said the next phase requires intelligent connectivity and predictable performance. “Connectivity should be intelligent,” he said, adding that networks should eventually provide more predictable latency rather than best-effort speeds.

Biswal acknowledged that AI infrastructure demands heavy capital investment but said telecom operators are accustomed to such cycles. “Heavy capex cycles are not new,” he said, adding that the key is to architect investments around sustainable unit economics.

The panel also highlighted the growing importance of edge AI, particularly in emerging markets. 

Wilson White, vice president for government affairs and public policy for Asia Pacific at Google, said many users experience AI primarily through devices.

“Most people are interacting with AI through devices,” he said, pointing to on-device models that process data locally. He said such approaches can enhance privacy and ensure functionality even when connectivity is weak.

White emphasized that trust and privacy must underpin adoption. “People must trust the systems in order to want to use them,” he said.

Becky Fraser, Vice President, Global Government Affairs, Qualcomm, said hybrid AI models that span the cloud and devices are unlocking new use cases in healthcare and agriculture in India and other developing markets. On-device processing, she said, enables “near instantaneous response” and continued functionality even with limited connectivity.

Across the discussion, speakers stressed that the AI race will not be decided by chip counts alone. Standards, software, interconnects, energy systems, and skills development will shape how widely and effectively AI is deployed.

As Oxman said, “We need to take an international approach,” adding that collaboration between governments and the private sector is essential to building trusted AI infrastructure.