Businesses and research institutions in India are recalibrating their AI strategies in the face of soaring GPU costs and access barriers, giving rise to a shift towards more affordable CPU-based machine learning solutions. With the artificial intelligence market expanding at a projected CAGR of 25–35% according to a recent BCG-NASSCOM report , the pressure to democratise AI access and reduce dependence on imported hardware has never been higher.
The collaborative effort from Indian Institute of Technology Madras (IIT Madras), Ziroh Labs and IITM Pravartak Technology Foundation to establish an AI Centre of Excellence exemplifies this shift, as it looks to deliver practical solutions using CPUs already widely available in data centres and edge devices. This approach is positioned as a counterpoint to the dominance of ‘Big AI’ models that typically require expensive GPUs. As highlighted in the press release, Ziroh Labs’ Kompact AI platform claims to deliver at least three times the performance of existing CPU-based AI alternatives, altering the hardware calculus for AI development in emerging markets.
CPUs Re-Enter the Race for AI Accessibility
No ads. No tracking.
We don’t run ads or share your data. If you value independent content and real privacy, support us by sharing.
Read More
Market research and technical analysis reinforce the strategic timing of IIT Madras’ and Ziroh Labs’ initiative. While GPUs have become the preferred choice for high-volume AI computations—powering deep learning across industries—they remain costly and in short supply. According to IndiaAI coverage , computation on global AI models can cost $2.50–$3 per hour, while new Indian AI projects target far lower costs. The Indian government has acknowledged the structural need for affordable compute resources, launching subsidised GPU access schemes and even aiming to develop indigenous GPU technology over the next three to five years. Until these efforts mature, optimising AI workloads for CPUs presents a pragmatic alternative.
Data from PIDATA Centers suggests over 80% of enterprise AI projects rely on both CPUs and GPUs or specialised accelerators. While GPUs excel at parallel tasks, CPUs remain essential for certain algorithms and are more cost-efficient, particularly in edge computing and scenarios with limited infrastructure. The introduction of platforms such as Kompact AI, capable of running foundational models like DeepSeek and Llama directly on CPUs with significant performance gains, holds potential to level the AI playing field for enterprise users and public sector applications alike.
AI Democratization Aligns with India’s National Push
The move towards locally-optimised, task-specific AI models is consistent with the Indian government’s push for technological self-sufficiency and broad access. The AI Centre of Excellence initiative echoes recommendations from the Carnegie Endowment and the Atlantic Council , both advocating public-private collaboration for AI R&D tailored to India’s specific use cases. IIT Madras itself has a track record in developing solutions for rural healthcare and education—a tradition acknowledged in its current efforts to design smaller, domain-specific AI models for grassroots deployment.
Professor V. Kamakoti, Director of IIT Madras, reinforced the social and economic rationale, stating in the press release: ‘This effort by Ziroh lab and IITM Pravartak is a nature-inspired one wherein they provide a platform that uses custom and trained domain-specific models to provide accurate inferences on affordable conventional compute machines. This effort is certainly a major step in arresting the possible AI divide between one who can afford the modern hyper scalar systems and one who cannot.’
Competitive Responses in the Indian AI Sector
India’s focus on affordable, accessible AI is consistent with broader competitive trends across Asia and emerging markets, where cost remains a decisive factor in AI adoption. According to the Indian Council for Research on International Economic Relations , India’s AI start-up ecosystem is rapidly growing but still faces challenges in scaling to unicorn status, particularly when competing against well-funded US and Chinese firms that dominate the GPU market. As a result, indigenous approaches using available CPU infrastructure are seen as a pathway to both autonomy and inclusive growth.
Meanwhile, the National Association of Software and Service Companies (NASSCOM) reported more than 1,600 AI-focused start-ups in 2023, with government-backed AI Centres acting as incubators for sector-specific models in fields such as agriculture, healthcare and logistics. Multiple studies cited in industry reports indicate that cloud compute, open data and the proliferation of AI on mobile devices are accelerating new business models beyond traditional tech hubs.
Looking Ahead: Structural Shifts and Broader Implications
With GPU supply remaining volatile in the short term, CPU-optimised AI models will likely gain traction among organisations seeking to deploy AI at scale without prohibitive infrastructure costs. Enterprises and public sector bodies stand to benefit from platforms such as Kompact AI, which promise high performance without the regulatory and operational complexities of using internet-dependent models. According to IBM analysis , while GPUs remain crucial for large-scale deep learning, increased efficiency of CPUs for particular applications and optimised algorithms is reshaping the competitive equation.
IIT Madras and its partners are positioning themselves at the intersection of market need, policy direction and technical progress. As the Indian AI sector matures, developments that improve accessibility and practical utility will drive adoption in sectors far removed from traditional technology hubs. The next phase of growth will depend on how effectively AI can be decentralised and tuned for local relevance, setting benchmarks for emerging markets worldwide.
