As much of the global semiconductor industry remains fixated on AI training accelerators and hyperscale GPUs, US-based AI processor company Blaize is making a different wager in India: large-scale inference embedded in sovereign public infrastructure.
In an interview with DIGITIMES Asia, Blaize co-founder and CEO Dinakar Munagala framed India not as an opportunistic market entry, but as a structural pillar in the company's strategy.
"India is strategically important to Blaize as a sovereign AI growth market, a deep engineering base, and a long-term platform for infrastructure-scale AI deployment."
The language is deliberate. Rather than describing India primarily as a revenue expansion territory, Munagala positioned it as both an engineering hub and a proving ground for what he calls "infrastructure scale AI deployment."
Hyderabad as execution base
Blaize already operates an India entity headquartered in Hyderabad, with engineering, sales, and business development teams on the ground. That presence was reinforced at the World Economic Forum in Davos, where the company signed an MoU with the Government of Telangana to support applied AI initiatives under the state's AI program, now operating as Aikam.
The Telangana engagement is structured around building an R&D center focused not on speculative research, but on validating hybrid edge and data center AI architectures in real-world settings.
Through the agreement, Blaize intends to establish "an R&D Center focused on advanced AI computing to support applied AI initiatives and infrastructure modernization programs."
The emphasis is on applied deployment: "The work will center on applied AI research, pilot program design, validation of hybrid edge and data center architectures, and deployment scale integration for real-world use cases such as public safety, digital services, and infrastructure monitoring."
Rather than committing to a headline hiring number, Munagala indicated that team expansion will track program momentum. The model suggests a deployment-led scaling strategy rather than a speculative talent land-grab.
US$56 million public safety validation
The more commercially tangible element of Blaize's India push is a US$56 million smart public safety initiative executed in partnership with Yotta, an Indian sovereign AI cloud and digital infrastructure provider.
Initial shipments are underway, with phased expansion through 2026. Crucially, the applications run directly on Blaize's integrated systems.
"The public safety applications in this deployment run directly on Blaize integrated AI systems, which combine our Graph Streaming Processor silicon and our Blaize software stack into a unified inference platform."
This is not a software overlay on third-party silicon. The silicon-and-software stack is positioned as the core inference layer enabling multiple public safety workloads at scale.
From a commercial standpoint, Munagala described the deployment as strategically important because it validates Blaize's architecture inside a national digital infrastructure framework. In other words, it is both a revenue and a reference case.
More importantly for semiconductor observers, the program is not framed as a pilot.
"The current program is structured around a scalable infrastructure architecture rather than a single site deployment."
The rollout began earlier this year, with shipments already in progress. The architecture, Munagala said, "is designed to be repeatable."
That repeatability is central. Public safety modernization projects across states often share similar structural constraints: energy consumption, latency, compliance requirements, and high endpoint density. If the Blaize-Yotta model proves durable, it could serve as a template for additional state-level deployments.
Edge as production phase of AI
Technically, Blaize's India thesis rests on one argument: inference must move closer to where data is generated.
"At India scale, public safety systems operate across vast geographic footprints with high endpoint density and diverse operating conditions. Centralizing all AI workloads in distant data centers creates structural constraints in bandwidth, latency, reliability, and cost."
Edge AI, Munagala argues, addresses those constraints by enabling inference at or near the point of capture. In practice, this means deploying processing at aggregation points closer to cameras and sensors while coordinating with centralized cloud infrastructure.
He describes inference in operational terms:
"Edge inference represents the production phase of AI. After models are trained, value is realized when inference operates directly in the physical world, where cameras, sensors, and devices generate continuous data streams."
In India's operating environment that spans high-temperature zones, remote locations, and infrastructure-limited areas, power efficiency becomes decisive.
"A power-efficient edge AI platform can operate reliably in high temperature zones, remote locations, unmanned installations, and other physically constrained sites where traditional data center infrastructure is not practical."
Beyond the fab narrative
India's semiconductor policy discourse over the past two years has largely revolved around fabrication incentives and domestic manufacturing ambitions. Blaize's strategy underscores a parallel development: embedding AI silicon within sovereign infrastructure programs before local fabs become operational.
Munagala's long-term ambition is explicit. "Over the next three to five years, we would like Blaize to be recognized as a foundational contributor to India's AI infrastructure transformation."
For the semiconductor industry, the significance lies less in the MoU headline and more in what it signals: India is emerging as a deployment-scale inference market, where alternative AI processor vendors can validate power-efficient architectures outside the hyperscale training ecosystem.
If repeat deployments follow, India could become not just a customer, but a proving ground for the next phase of AI silicon monetization.
Article edited by Jack Wu
