Modern artificial intelligence has an electricity problem. Training and running frontier models now consumes power at data-centre scale — gigawatts, in the words of one founder — and every efficiency gain in the chip underneath translates directly into cheaper, more sustainable AI. In Hyderabad, a startup that spun out of the Indian Institute of Technology Hyderabad is attacking the problem from the silicon up: not with a better data centre or a smarter scheduler, but with AI chips designed in India, aimed at squeezing more intelligence out of every watt.
Two chips before the company existed
semXai’s unusual origin story is one of its strongest credentials. Co-founder Dinesh YVS told The New Indian Express in October 2026 that the team deliberately waited to incorporate until the hardest part of the semiconductor value chain was already de-risked: we decided not to start the company until we had figured out the ecosystem and completed a couple of chip fabrications.
Two chip iterations were designed, fabricated, packaged and brought up before the company formally spun out of IIT Hyderabad in 2025.
The chip broke cover at the T-CHIP Semicon Constitution Summit in Hyderabad in September 2025, where Dinesh told ANI it was designed entirely with Indian technology and talent. Days later, the chip appeared as an exhibit at India’s first Semiconductor Innovation Museum — The Hindu listed it among the opening exhibits as an indigenous AI chip. On the summit sidelines, Dinesh described it as India’s first indigenously designed AI chip — his characterisation, and one that should be read as the founder’s, since other Indian teams (several of them featured on this site) are also building AI silicon.
The stack: TensorTune, WattWise, AetherCore AI
What distinguishes semXai’s approach, according to the company, is that it treats the chip and the software as one co-designed problem. The company’s official site describes a two-phase stack. Phase one is the software layer: TensorTune, which restructures AI models for efficient execution before deployment, and WattWise, a workload-aware power and thermal manager that adapts execution to stay within power limits. Phase two is the silicon layer: AetherCore AI, with fine-grained control over frequencies and power domains and memory-aware execution for what the company calls absolute minimal energy per operation.
The pitch, in Dinesh’s words: right from Python till Silicon, we are reimagining the entire AI hardware and software co-design.
Rather than trying to beat NVIDIA, AMD or Intel at their own generic-compute game, semXai positions its chips as AI-native silicon that complements generic CPUs and GPUs — aimed at the wave of AI the founder expects around 2027–28, when robots, autonomous vehicles and drones run increasingly on their own.
The grants and the talent engine
The funding story so far is grants, not venture capital. Dinesh says the team raised nearly $4 million in grants while still inside IIT Hyderabad, including support from India’s Defence Ministry, the Ministry of Electronics and IT, Qualcomm and AMD — a roster that reads as both money and validation. The company’s site displays the same institutional logos alongside its research footprint: 20+ conference papers, 20+ journals and 12 patents filed.
More Read
The team itself is the university pipeline in action. Dinesh — a master’s graduate who re-enrolled for a PhD to build the technology — says 15 to 20 students work on different problem statements under him at any given time, from compilers to AI design to fabrication, with master’s and undergraduate theses feeding directly into the company. He co-founded the company with a professor at IIT Hyderabad, whose name the company has not publicly disclosed.
The company behind the chips
semXai’s bet is as much economic as technical. Dinesh frames it as a sovereignty argument: India has demand for AI in every sector but, in his words, does not really have sovereign AI chips — the country runs its AI on generic imported compute. Why can’t we be India’s Nvidia or Qualcomm, right from the roots and our talent from this place?
he asked ANI. The long-term vision is to make AI hardware cheap enough that startups and educational institutions can afford it, keeping the money — and the capability — at home.
The caveats that matter
Three calibrations are essential. First, the headline claims — India’s first indigenously designed AI chip, the grant roster, the research output — come from the founder and the company, reported by mainstream press but not independently verified; no public specification sheet, customer, or production timeline has been disclosed. Second, semXai is early-stage and pre-commercial: the two fabricated iterations are proof of capability, not products shipping at volume. Third, the company’s public communication is thin — there is no published roadmap, no named customers, and no independent benchmark data for the efficiency claims yet.
Why this one is worth watching
Chips are the hardest thing a startup can attempt in India, and most attempts die in the gap between design and fabrication. semXai crossed that gap twice before it even had a company — with a student team, grant money, and a hardware-software co-design philosophy that starts at the Python layer instead of ending there. The next test is the one every chip startup faces: turning working silicon into a product somebody buys. If an IIT Hyderabad spinout can ship energy-efficient AI silicon designed end to end in India, the sovereignty argument stops being a slogan — and the electricity bill for India’s AI future gets a little smaller.
