Oxmiq pushes IP and design tools for India’s AI-chip teams

November 18, 2026

Oxmiq pushes IP and design tools for India’s AI-chip teams

Image: ChatGPT

Oxmiq Labs is positioning itself as an enabler for Indian AIartificial intelligenceAI stands for artificial intelligence-chip development, with a toolkit and intellectual property aimed at teams that want to design their own silicon. The company is also preparing an India initiative that Raja Koduri expects to outline in December.

The core idea is not just to produce faster chips, but to optimize the full stack around AIartificial intelligenceAI stands for artificial intelligence output. Koduri described the challenge as a memory-hierarchy problem that spans NAND1. Not ANDA logic gate that outputs 0 when all inputs are 1, and outputs 1 in all other cases2. negative-ANDNAND denotes a logic operation and is also the name of a widely used flash memory type. flash storage, DDR memory attached to the CPUcentral processing unitMain processor that executes program instructions and performs data operations., HBM1. human body modelESD test model that simulates a person discharging static electricity into a component.2. High Bandwidth MemoryHBM is a stacked DRAM architecture designed to provide very high memory bandwidth. attached to the GPUgraphics processing unitA GPU is a parallel graphics processor widely used for AI acceleration., and SRAM1. static random access memoryMemory technology or system for temporarily or permanently storing digital information.2. static random-access memorySRAM is used here as a test vehicle for embedded memory performance.. In that view, the relevant metric is how efficiently a system converts power into tokens, not just raw floating-point performance.

Oxmiq is also working with AM Intelligence Labs, the AIartificial intelligenceAI stands for artificial intelligence compute arm of AM Group. The partnership covers site selection, compute design, partner selection, rack placement, networking and storage requirements, thermal management, cooling, and software stack operations. The first AIartificial intelligenceAI stands for artificial intelligence factory in Hyderabad is described as the first such deployment in Asia. It has ordered about 9000 Nvidia Rubin GPUgraphics processing unitA GPU is a parallel graphics processor widely used for AI acceleration.s for a 30 MWmolecular weightMass of a molecule, commonly expressed as relative molecular or molar mass. facility, with Vera Rubin NVL72 rack-scale systems scheduled for the first quarter of 2027. An additional order for 20000 Rubin GPUgraphics processing unitA GPU is a parallel graphics processor widely used for AI acceleration.s has since been placed for sites in India and Malaysia.

Koduri argued that custom silicon begins to make economic sense at scales of at least 100 MWmolecular weightMass of a molecule, commonly expressed as relative molecular or molar mass.. He put the cost of 100 MWmolecular weightMass of a molecule, commonly expressed as relative molecular or molar mass. of off-the-shelf computing capacity at roughly US$5.5 billion (approximately €5.0 billion), and estimated that a 1-GW data center could cost about US$55 billion (approximately €50.1 billion) to build. In his view, a tailored system can be significantly more cost-effective because it is designed for one data center and its specific constraints rather than for many different customers.

He also said a state-of-the-art chip manufactured by TSMCTaiwan Semiconductor Manufacturing CompanyTSMC is named as a founding partner of the EPIC Center. could require US$100 million (approximately €91.1 million) to US$150 million (approximately €137 million) in R&D, while a complete custom system could cost up to about US$500 million (approximately €455 million) to develop. If a custom design roughly halved the cost of the deployed system, the savings could exceed US$2 billion (approximately €1.8 billion) for a 1-GW buildout even after development spending.

Manufacturing location, he said, is secondary to ownership of the product and the design capability behind it. Chips can be designed in India and built in Taiwan or South Korea; what matters is that India controls the resulting products. Oxmiq’s customer base in India is focused on infrastructure and silicon design, while its broader portfolio also includes companies working on new memory, silicon photonics and photonics-based interconnects.

The backdrop is a fast-expanding Indian data-center market. Installed capacity in India stood at 1.57 GW in August 2026, up from about 375 MWmolecular weightMass of a molecule, commonly expressed as relative molecular or molar mass. in 2020, and is projected to approach 8 GW by 2030. Against that growth, Koduri estimated India would need at least 6 GW of infrastructure to provide intelligence to its population, while the currently available GPUgraphics processing unitA GPU is a parallel graphics processor widely used for AI acceleration. capacity in the country remains far below that level.

Oxmiq’s pitch is therefore broader than chip design alone: it is selling the IP1. image placement2. Internet protocolInterface or communications concept for transmitting data between devices and systems.3. Intellectual PropertyIP denotes licensable semiconductor design blocks or technology elements., tools and engineering support needed to design custom silicon, systems and software around AIartificial intelligenceAI stands for artificial intelligence data-center constraints. The company’s role spans technology selection and system-level integration, with Nvidia GPUgraphics processing unitA GPU is a parallel graphics processor widely used for AI acceleration.s used initially and locally developed designs intended to follow.

Source: EE Times

← All posts