AI isn’t just for large enterprises: Xebia's Anand Sahay on why MSMEs could be AI’s biggest winners

India's drive for sovereign AI is gaining traction with projects like the IndiaAI Mission and funding for local computing power. However, the significant potential goes past GPUs and data centers. With businesses wanting greater command over their data, models, and AI decision-making, the discussion is moving towards developing AI ecosystems for India that are localized, speak multiple languages, and are tailored to specific industries. Simultaneously, there are unanswered questions regarding whether the advantages of AI will extend to the nation's extensive network of MSMEs and manufacturers, rather than solely benefiting large corporations.

During a conversation with ET Digital, Xebia's Global CEO, Anand Sahay, covers India's sovereign AI prospects, the increasing need for companies to manage their AI infrastructure, the development of AI that understands context and industry, and how smaller firms can leverage AI for better results despite limited resources. Edited excerpts.

The Economic Times (ET): India is rapidly expanding sovereign AI infrastructure through the India AI Mission. Why is this moment strategically important for enterprises?
Anand Sahay (AS): India’s sovereign AI push is strategically significant because it lowers one of the biggest barriers to AI adoption, which access to compute infrastructure. AI experimentation, model development, and enterprise-scale deployments require substantial compute power, and sovereign GPU initiatives can meaningfully accelerate AI adoption across industries by making that infrastructure more accessible.


However, the opportunity extends far beyond infrastructure alone. India has a unique advantage in building domain-specific and multilingual AI systems that are aligned to local business realities, regional languages, and industry-specific requirements. This creates the foundation for India to evolve not just as an AI consumer market, but as one of the world’s largest applied AI economies.

For enterprises, this is an opportunity to move faster on AI adoption while simultaneously building systems that are more contextual, localized, and aligned with emerging governance and regulatory expectations.

ET: The conversation around sovereign AI often focuses on data localization. Is sovereignty becoming a broader enterprise issue?
AS: Absolutely. Sovereignty today is no longer limited to the question of where data resides. Enterprises are increasingly evaluating whether they control the broader AI stack, including infrastructure, models, governance frameworks, and operational outcomes.

Regional cloud infrastructure addressed part of the challenge around data residency, but organizations are now asking much deeper questions around ownership and control. They want to understand who controls the models, where inference happens, how enterprise knowledge is protected, and how much operational autonomy they truly possess within AI environments.

This becomes especially critical for regulated sectors such as banking, financial services, healthcare, telecom, and public sector environments, where governance, risk management, and data sensitivity are central to enterprise operations.

The broader industry conversation is now shifting from “Where is the data stored?” to “Who controls the intelligence layer?”

ET: How does India’s sovereign AI push create opportunities for multilingual and domain-specific AI systems?
AS: India’s diversity itself creates the opportunity. Enterprise AI systems in India cannot rely solely on generalized English-language models if they want meaningful adoption across industries, regions, and user groups. There is growing demand for AI systems trained on regional languages, industry-specific workflows, enterprise knowledge, and localized operating environments.

This is where sovereign compute becomes strategically important. It enables organizations to fine-tune or build models that are aligned to Indian business realities and domain-specific operational requirements.

As enterprises increasingly move toward agentic AI systems, contextual intelligence becomes even more critical. AI systems must understand enterprise processes, regulatory frameworks, operational nuances, and industry-specific workflows — not just generic internet-scale information. The ability to build localized and contextual AI systems will become a major differentiator for enterprises operating in India’s diverse business environment.

ET: What should enterprises prioritise as they prepare for sovereign AI adoption?
AS: The first priority should be clarity around business outcomes. Organizations need to begin by identifying which workloads genuinely require sovereign deployment, which workloads can continue leveraging public cloud AI services, and where governance or regulatory requirements demand tighter operational control.

The second priority is data readiness. Many enterprises possess large volumes of data, but not necessarily structured, trusted, or usable knowledge that AI systems can effectively leverage. As AI systems become more autonomous and context-aware, the quality and governance of enterprise data become foundational to success.

Finally, organizations need flexible and modular architectures. AI innovation cycles are evolving rapidly, and enterprises should avoid overly rigid infrastructure strategies. Instead, they should focus on building adaptable environments that can evolve alongside changing technologies, models, and business requirements.

The conversation should not begin with infrastructure procurement alone. It should begin with workload strategy, governance clarity, and business value realization.

ET: Xebia recently expanded its AI engineering presence in Hyderabad. How does that connect to India’s sovereign AI opportunity?
AS: The Hyderabad expansion reflects the growing enterprise demand for AI-native engineering capabilities and scalable AI transformation expertise. As organizations move from experimentation toward enterprise-wide AI adoption, they require stronger capabilities around AI engineering, cloud-native architectures, governance frameworks, and domain-specific AI implementation.

India is entering a phase where enterprises are not just consuming AI tools but actively building AI-enabled operating models. That shift requires deep engineering expertise, modernization capabilities, and the ability to operationalize AI at scale across enterprise environments.

The expansion strengthens Xebia’s ability to support enterprises as they navigate this transition toward sovereign, modular, and enterprise-scale AI systems while also helping organizations build more future-ready and context-aware AI ecosystems.

ET: Much of the AI conversation is dominated by large enterprises with deep pockets and dedicated technology teams. Are small and mid-sized businesses at risk of being left behind in the AI transition, or are you seeing a different reality on the ground?
AS: We see a very different reality emerging on the ground. Large enterprises may dominate the AI narrative because they have the scale, budgets, and visibility to invest aggressively, but small and mid-sized businesses are often far more pragmatic and outcome-oriented in their adoption approach. They are not approaching AI as a large transformation exercise or a technology showcase. Instead, they are evaluating it through a much sharper operational lens -how to improve productivity, reduce downtime, optimise costs, and become more competitive in a rapidly evolving market.

What is important is that AI is no longer confined to organizations with massive infrastructure or dedicated research teams. The ecosystem has matured significantly over the last few years. Cloud-native AI platforms, modular architectures, smaller domain-specific models, and managed AI services have made adoption far more accessible and commercially viable for mid-sized organizations. Businesses today can start with focused use cases, prove measurable outcomes, and scale gradually without making disproportionate upfront investments.

In many ways, the real divide is not between large enterprises and SMBs. The real divide is between organizations that are building AI-ready operating models and trusted data foundations versus those that are still treating AI as an isolated experimental layer on top of fragmented systems. SMBs that move with focus, operational clarity, and agility can often adopt AI faster and more effectively than larger organizations constrained by legacy complexity.

ET: How do the AI needs of a 100-person manufacturing company differ from those of a large multinational enterprise? Are SMBs looking for fundamentally different outcomes from AI, or simply more accessible ways to achieve them?
AS: The desired outcomes are often similar, but the operational realities are fundamentally different. A large multinational typically approaches AI from the perspective of scale, governance, global integration, and enterprise-wide transformation. Their focus is on building standardized AI ecosystems that can operate across multiple geographies, business units, and compliance environments.

A 100-person manufacturing company, however, operates much closer to day-to-day business outcomes. Their priorities are immediate, operational, and ROI-driven, like reducing machine downtime, improving quality control, optimizing inventory planning, lowering energy consumption, and enabling faster decision-making on the shop floor. They are not looking for overly sophisticated AI stacks or large transformation programs. They are looking for accessible, modular, and measurable solutions that can integrate quickly into existing operational environments.

This is where the AI conversation needs to evolve. Smaller manufacturers often do not require massive foundational models or highly centralized AI systems. In many scenarios, specialized AI deployments trained around specific workflows or machine environments deliver far greater business value. For example, a lightweight domain-trained model running closer to the edge can enable real-time operational intelligence without requiring large-scale infrastructure investments.

The future of manufacturing AI will not be defined purely by scale. It will be defined by precision, contextual intelligence, adaptability, and the ability to embed AI directly into operational ecosystems where decisions need to happen in real time.

ET: Many AI solutions appear to be designed for organizations with vast data resources and complex IT infrastructure. What are the biggest barriers preventing smaller businesses from moving beyond experimentation and adopting AI at scale?
AS: One of the biggest misconceptions in the market today is that AI adoption is primarily a technology challenge. In reality, the larger challenge is organisational readiness. Many businesses already possess significant amounts of data, but very little of it is contextualised, structured, governed, or maintained in a way that AI systems can effectively utilise.

As enterprises move from traditional analytics toward more autonomous and agentic AI systems, the quality of enterprise data becomes foundational. AI is only as effective as the operational context it can access. If organizations have fragmented systems, siloed workflows, inconsistent documentation, or weak governance structures, scaling AI becomes extremely difficult regardless of how advanced the models are.

The second barrier is economic clarity. SMBs operate with far tighter investment discipline than large enterprises. They need a clear line of sight between AI investments and measurable business value. If organizations cannot directly connect AI deployments to productivity gains, operational efficiency, cost optimization, or revenue growth, adoption naturally slows down.

There is also a significant capability gap in the broader ecosystem. Many AI solutions are still designed assuming the presence of mature engineering teams, sophisticated cloud environments, and advanced infrastructure capabilities. That is not the operating reality for most mid-sized businesses. The industry needs to move toward more modular, domain-aware, and implementation-friendly AI ecosystems that simplify adoption instead of increasing complexity.

ET: India's manufacturing sector is increasingly embracing digital transformation. Which AI use cases are delivering the most immediate and measurable value for small and medium manufacturers — productivity, quality control, supply chain management, predictive maintenance, or something else?
AS: For small and medium manufacturers, the AI use cases delivering the most immediate and measurable value are the ones directly connected to operational efficiency and production continuity. Predictive maintenance is one of the strongest examples because the business impact is immediate, measurable, and financially visible. When AI systems can identify anomalies in machine behavior before failures occur, organizations can significantly reduce downtime, improve asset utilization, optimize maintenance cycles, and avoid costly production disruptions. For manufacturers operating on tight margins, that operational continuity creates direct business value.

Quality control is another high-impact area where AI adoption is accelerating rapidly. Computer vision systems and edge-based intelligence are helping manufacturers detect defects in real time, reduce wastage, improve consistency, and enhance production accuracy across assembly lines. Traditionally, many of these processes relied heavily on manual inspection, which limited scalability and precision. AI is fundamentally changing that equation by enabling faster and more intelligent quality assurance workflows.

We are also seeing increasing adoption around supply chain visibility, inventory forecasting, energy optimization, and real-time operational analytics. However, one of the most important shifts happening now is the movement toward edge AI and specialized smaller language models. Manufacturers increasingly want intelligence to sit closer to machines, devices, and operational systems rather than relying entirely on centralized cloud environments. As AI becomes more embedded into industrial operations, the organizations that succeed will be the ones that can combine localized intelligence, operational agility, and real-time decision-making into their production ecosystems.

ET: There is a perception that the current AI wave is being built primarily for large corporations. Do you believe the industry has done enough to democratize AI for smaller businesses, and what needs to change to make advanced AI accessible to the broader business ecosystem?
AS: The industry has made meaningful progress in democratizing access to AI, but accessibility and availability are two very different things. Technology availability has improved significantly. Today, businesses have access to cloud AI platforms, open-source models, modular deployment frameworks, and increasingly affordable compute infrastructure. Government-led initiatives around sovereign compute and AI infrastructure are also accelerating access, particularly in markets like India.

However, true democratization requires more than access to tools. It requires simplification. Many AI solutions are still designed around the assumption that organizations already possess mature data environments, advanced engineering capabilities, and highly evolved governance structures. That assumption excludes a very large part of the business ecosystem.

The future of AI adoption will depend on how effectively the industry can reduce complexity for businesses that are still early in their digital maturity journey. Organizations need AI systems that are modular, easier to deploy, domain-specific, and aligned to measurable business outcomes rather than large transformation narratives.

Equally important is the question of data readiness. Many organizations today have enormous volumes of data, but not necessarily the right data architecture or contextual intelligence required for AI-driven decision-making. Businesses that focus first on building trusted data foundations, operational governance, and adaptable architectures will be far better positioned to scale AI successfully over the next decade.

Ultimately, AI will not become mainstream because the models become larger. It will become mainstream when the ecosystem becomes simpler, more outcome-oriented, and accessible to businesses of every size.