A Bengaluru startup thinks it can build AI inference chips that sidestep NVIDIA's ecosystem entirely. The hard part? Actually delivering them.
Sanchayan Sinha has worked at some of the chip world's most interesting addresses: Groq, the inference darling that briefly made waves with speed claims; AMD, the perpetual challenger; Lightmatter, which bets on photonics. Now he's trying something arguably harder—building an AI chip company in India, from scratch, betting that open architectures can crack a market NVIDIA has sewn up.
His startup, Turiyam AI, closed a pre-seed round from Ankur Capital late last year. The amount? Undisclosed, as these things often are at the earliest stages. What matters more, perhaps, is the timing. Turiyam is surfacing just as India's semiconductor ambitions shift from aspirational to institutional, with government backing, capital commitments topping ₹1.60 lakh crore, and a suddenly crowded field of homegrown chip ventures.
Turiyam positions itself at a tricky intersection: building what it calls a "full-stack generative AI inference platform" on RISC-V architecture—no NVIDIA CUDA in sight. The pitch is technical but pointed. Custom silicon. Middleware that routes around the dominant GPU stack. Compatibility with open-source AI models. A hybrid memory design mixing SRAM and high-bandwidth memory. Even a compiler that leans on reinforcement learning to map workloads more efficiently.
In Ankur Capital's phrasing, the goal is "world-class token-per-dollar-per-watt performance" for large language model deployments. That's venture-speak for: cheaper, faster inference at scale. Whether Turiyam can actually deliver on that—well, that's the bet.
The Pedigree and the Pitch
Sinha isn't working alone. His co-founder, Praveen Jain, previously led India operations for XCOM Labs, a wireless tech outfit. The team now sits somewhere between 11 and 50 people, according to LinkedIn—a range that suggests the company is still in buildout mode rather than full production sprint.
The technical stack Turiyam describes is ambitious, even by the standards of well-funded Silicon Valley accelerator startups. Vertically integrated. Software-first. RISC-V instruction sets as the foundation. It's the kind of roadmap that sounds plausible on a pitch deck but becomes exponentially harder when you're actually taping out silicon, validating performance, and trying to convince enterprise customers to swap out proven infrastructure.
Still, there are reasons to take it seriously. India's government is rolling out the red carpet. In mid-February, Sinha and SVP Mukul Ojha met with the Office of the Principal Scientific Adviser—no small thing—to demo inference tech and discuss, in the government's words, "a fully indigenous, data-centre-class AI GPU." Weeks later, Turiyam showed up at the India AI Impact Summit in New Delhi. The optics matter. So does the policy backdrop.
Sovereign Compute, Real Money
India's Semiconductor Mission has committed fiscal support covering up to half the cost for approved projects. In early March, officials unveiled ISM 2.0, targeting deep-tech startups and design ecosystems specifically. The message is clear: India wants its own chip stack, particularly for AI, and it's willing to subsidize the learning curve.
This isn't just nationalist posturing. There's a practical angle. Data sovereignty concerns are real. Enterprises running high-volume LLM inference—especially in regulated sectors—are increasingly wary of vendor lock-in to a single GPU architecture. If Turiyam (or anyone else) can credibly offer "CUDA-free" infrastructure with comparable performance and better economics, there's a market.
The question is execution. Chip startups are expensive, slow, and littered with cautionary tales. Even well-funded ventures stumble on yield issues, compiler bugs, or the sheer inertia of an industry that defaults to NVIDIA.
A Suddenly Crowded Space
Turiyam isn't the only Indian startup chasing this moment. Sensesemi pulled in Rs 25 crore in January for edge-AI system-on-chips. Tattvam AI closed a $1.7 million pre-seed in late February, focused on AI-assisted chip design automation—basically, using AI to build AI chips, which has a certain recursive elegance. Mindgrove Technologies raised $8 million back in December 2024 for microcontroller production.
Globally, the inference accelerator category is contested and getting more so. SiMa.ai, which runs R&D in Bengaluru, raised $70 million in April 2024 for a multimodal GenAI chip. EdgeCortix, which also has an India footprint, closed part of its Series B in August 2025, pushing total funding near $100 million.
What's striking is less the individual raises and more the cluster effect. India suddenly has a visible cohort of chip startups, backed by domestic and international capital, operating in overlapping but distinct niches. That's the shape of an emerging ecosystem—or the prelude to a shakeout, depending on how you squint at it.
The Open Question
For now, Turiyam remains pre-product in any commercial sense. Availability timelines? Undisclosed. Pilot customers? Not announced. The company recently signed a memorandum of understanding with SVKM's NMIMS in Mumbai to collaborate on talent pipelines—smart, given the chronic shortage of chip designers in India—and has moved into workspace at Bridge+ in Bengaluru, the kind of co-working hub favored by hardware startups that need more than a laptop and Wi-Fi.
The bet Turiyam is making is structural. That open standards—RISC-V instead of proprietary ARM or x86, open-source models instead of closed APIs—can carve out a wedge in the inference layer. That India's infrastructure needs, combined with government tailwinds and enterprise wariness of single-vendor dependence, create an opening.
Whether a pre-seed stage company can actually manufacture data-center-grade silicon, validate performance claims, and win over cautious IT buyers remains very much an open question. The chip graveyard is full of promising architectures that never made it past the whitepaper stage.
But if India is serious about building sovereign compute capabilities—and the policy signals suggest it is—then someone has to be first. Turiyam is making a case that it could be them. The next 18 months will tell us if that's ambition or overreach.
