India's AI Race: Winning Isn't About Running Faster

 Every Olympic Games begins with the same beautiful illusion.

Hundreds of athletes walk into the stadium under one flag. Yet, within minutes, they scatter into completely different arenas. Some sprint the 100 metres. Others prepare for the marathon. Some dive into the pool, while others pick up a javelin or stand before a weightlifting platform. No nation expects to dominate every event. Success comes from understanding its strengths, investing consistently, and choosing the competitions that matter most.

Yet, when the conversation turns to Artificial Intelligence, we suddenly behave as though there is only one race. Every headline seems to ask the same question:

"Can India build its own GPT?"

We speak as if the future of an AI-powered nation depends on crossing a single finish line before everyone else. The more I listen to these debates, the more I feel we are asking the wrong question. Not because foundation models are unimportant, but because AI is not one race. It is many races running simultaneously.

And perhaps India's greatest challenge today is not running too slowly. It is deciding which races will create lasting strategic advantage

When every revolution grows up

History has a habit of repeating itself because the structural shift of every technology revolution follows a familiar journey: it starts as a product and eventually matures into an ecosystem.

The internet first became visible to most people through static websites; over time, it matured into a vast ecosystem of platforms, applications, networks, and cloud infrastructure. Electricity began as a scientific breakthrough. Eventually, it became something every factory, hospital, and household simply assumed would be available. Nobody today asks who built the best electrical generator. We ask who built reliable power grids.

Today we admire the skyscrapers. Tomorrow we will depend on the roads, electricity, water, and governance that keep the city alive. AI is beginning the same transition. Models attract the headlines, but ecosystems create enduring value.

That distinction changes everything.

Every generation of AI models will eventually be surpassed. That is the nature of software. Infrastructure, however, compounds. Every new model demands more compute, better networking, faster storage, stronger orchestration, and trusted deployment. Ironically, every breakthrough in AI increases the value of infrastructure rather than reducing it.

Infrastructure is not merely following the AI revolution.

It is quietly making the revolution possible.

Five races, one ecosystem

One reason the public debate feels confusing is that we often treat models, GPUs, startups, and funding as isolated objects. In reality, they are interconnected parts of a broader system. To understand where India belongs, we must look at how these layers interact.

Discovery is one race. This is where new scientific breakthroughs happen, like the creation of the Transformer architecture. India should contribute to this frontier because scientific capability creates long-term innovation. Yet history also reminds us that research alone rarely determines economic leadership.

Foundation models represent a different race. These models—GPT, Claude, Gemini, Llama, DeepSeek, and others—are extraordinary engineering achievements. Should India possess the capability to build them? Absolutely. Not because we must always build the highest-scoring model, but because capability itself creates resilience.

Every nation eventually reaches a point where technology shifts from being a commercial choice to a strategic one. Semiconductors did. Energy did. Telecommunications did. Artificial Intelligence is beginning that journey. The question is no longer who has the smartest chatbot. It is who controls enough of the ecosystem to remain technologically independent.

Infrastructure is where India must think much bigger. As someone working closely with India's AI infrastructure ecosystem, I increasingly see sovereign compute not as an industrial aspiration but as strategic insurance. Without trusted compute, discussions about AI sovereignty eventually lead back to someone else's data centre.

That is why investments in Make in India AI infrastructure, sovereign cloud, enterprise AI platforms, and indigenous compute capabilities matter. They allow India to participate in the global AI revolution on its own terms.

Enterprise adoption is perhaps the least glamorous race—and the most valuable. Most enterprises do not wake up wishing for another general-purpose language model. They want better lending decisions, more resilient supply chains, smarter manufacturing, and faster diagnostics.

Enterprise AI is not won by the largest model. It is won by understanding workflows, integrating institutional knowledge, and embedding intelligence into everyday decisions.

This is where AI stops being impressive and starts becoming useful.

Population-scale deployment is where value is finally realised. India has already demonstrated something remarkable over the past decade through UPI, Aadhaar, and Digital Public Infrastructure. These successes were implementation achievements, not merely technology achievements.

The capability to deploy digital systems at extraordinary scale may become one of India's greatest AI advantages. History rarely rewards technology that remains inside laboratories; it rewards technology that changes how societies function.

Beyond the false choices

The debate often forces us into unnecessary binary choices.

Should India build models or applications?

Should we invest in research or infrastructure?

These are false choices.

A healthy AI ecosystem requires all of them. The question is not whether India should participate across the stack, but where India must lead, where it must collaborate, and where it simply needs strategic capability.

The current AI conversation is understandably captivated by model releases. Every few months another benchmark falls, another leaderboard changes, and another model becomes state-of-the-art.

Those moments matter.

But they rarely define entire industries.

Infrastructure lasts longer than benchmarks. Ecosystems outlive products.

Adoption creates more economic value than demonstrations.

If there is one lesson India should take from this moment, it is this:

We do not have to win every race. We have to know which races compound into lasting national advantage.

The question, then, is no longer whether India can build another model.

The far more interesting question is this:

Can India build an AI ecosystem that the world chooses to build upon?

To me, that is the race worth running.

Mythos Is Not the Threat. Slow Defence Is.

A few weeks ago, the cybersecurity world got its latest scare: Mythos. The headlines made it sound like a supervillain had entered the room — an AI so powerful that it could allegedly discover weaknesses, chain them together, and turn them into real attacks much faster than humans could react.

That sounds terrifying. But the real story is more useful than the panic.

Mythos is not magic. It is more like a high-speed metal detector in a giant warehouse full of loose wires, open doors, old locks, and forgotten keys. If the warehouse is messy, the detector finds problems very quickly. If the warehouse is organized, locked down, and watched carefully, the same detector becomes far less alarming.

That is the part many people miss.

What Mythos Really Changes

For years, cyberattacks followed a familiar rhythm. A weakness was found, then tested, then weaponized, then used. Security teams often had days or even weeks to respond.

Mythos compresses that timeline. It pushes attackers closer to machine speed. Instead of needing a team of experts to manually explore every step, a model can help discover likely weak points much faster, and in some cases chain them into an attack path.

That does not mean every system is suddenly doomed. It means the slow, old rhythm of “we will patch it next month” is now much more dangerous.

A good analogy is traffic. A city can survive a few fast cars if it has traffic lights, signals, police, and lanes. But if the roads are broken and the rules are weak, one fast car can create a pile-up. Mythos is that fast car. The real question is whether the road system is ready.

Why People Are Scared

People hear “AI finds vulnerabilities” and imagine an all-powerful machine breaking into everything.

That is the wrong image.

A better image is this: Mythos is like giving a very smart, tireless junior analyst a huge checklist and letting it work at 3 a.m. without getting tired. That is powerful. But it still needs vulnerable systems, exposed pathways, and opportunities to move laterally.

So the fear is not that Mythos can do anything. The fear is that many organizations still have too many things left exposed.

Can the Same AI Help Defend?

Yes — and this is important.

If an AI can search for weaknesses, it can also help security teams search for them first. It can help red teams test systems faster. It can help defenders prioritize patching. It can help map where the biggest risks are hiding.

But there is a catch. The same tool that helps a defender can also help an attacker. So the answer is not “ban it” or “trust it blindly.” The answer is to control it, limit it, audit it, and use it inside a disciplined security process.

Think of it like a powerful kitchen knife. In a trained chef’s hand, it is a precision tool. In the wrong hand, it is dangerous. The knife itself is not the full story. The process around it matters.

The Indian Context: Why This Matters Now

India should treat this as a wake-up call, not a distant headline.

The country is moving fast on digital payments, cloud adoption, AI usage, fintech growth, and connected public infrastructure. That is a strength — but it also means the attack surface is growing quickly. Banks, NBFCs, fintech firms, IT services companies, startups, government-linked systems, and supply chains are all becoming more interconnected.

That creates two risks.

First, India has many organizations that are modern in customer experience but uneven in cyber maturity behind the scenes. A company may have a slick app and still rely on weak identity controls, delayed patching, or fragmented monitoring.

Second, India is now too important to ignore. As the digital economy expands, attackers will increasingly see Indian institutions as high-value targets, not just local targets.

In simple terms: when the digital highways get busier, criminals, spies, and opportunists all show up.

What India Needs to Do

India does not need panic. It needs discipline and response at machine speed.

1. Faster patching

The old model of “patch in a few weeks” is becoming risky, especially for public-facing systems. Critical vulnerabilities need faster triage, faster testing, and faster deployment.

2. Better identity security

A lot of modern breaches start with credentials, not movie-style hacking. Strong authentication, least privilege, and tighter access review matter more than ever.

3. Segmentation and containment

If one part of a system is breached, it should not become a freeway to everything else. Many organizations still have too much internal trust.

4. Stronger monitoring

If attacks accelerate, defenders need better detection, better alerts, and better response playbooks. Security teams should be able to spot unusual behavior quickly, not after damage is done.

5. Board-level cyber thinking

Cybersecurity can no longer sit only with the IT team. For banks, fintechs, manufacturing firms, and large enterprises, this is now a business resilience issue. Many organizations still confuse compliance with security.

6. AI governance

India also needs rules for where AI security tools can be used, who can use them, and how outputs are reviewed. The goal is to use AI defensively without creating new risks.

The Real Lesson from Mythos

The real lesson is not “AI has defeated cybersecurity.”

The real lesson is that slow defence is dying.

If attackers can move in hours, then organizations that still operate in weeks and months are exposed. That is true in the U.S., Europe, and especially in fast-growing digital economies like India.

So the story is not about fear. It is about readiness.

Mythos is a mirror. It shows whether an organization has strong cyber hygiene or just cyber hope.

And that is a useful shock.

Mythos is not the end of cybersecurity.  It is the end of slow cybersecurity.

AI First or Be Left Behind: The Real Question Is Not Adoption, It’s Consumption

There are moments in technology where the shift is not in what we use, but in how deeply we use it. Those moments are uncomfortable, because they force a different kind of question—not “Are we adopting this?” but “Are we using enough of it to stay relevant?”

I was reminded of this while revisiting two very different but connected ideas.

Andrew Grove
once articulated a simple but unsettling principle: the leaders who survive are not the most confident ones, but the ones who sense a shift early and act before the evidence becomes obvious. That instinct—almost paranoia—is what forces change before it is too late.

Decades later, in a completely different context, Jensen Huang made a statement that at first sounds exaggerated, almost provocative. He suggested that if a highly paid engineer is not consuming a significant amount of AI tokens, it should raise concern. The comparison he draws is not subtle: working without AI in today’s world is similar to a chip designer refusing to use modern design tools.

Strip away the rhetoric, and what remains is a powerful signal.

The shift in AI is not about access.
It is about consumption intensity.

The Shift: From Having AI to Using AI at Scale

Most organizations today are already “using AI.”

Teams generate content. Developers use copilots. Analysts experiment with models. Presentations look sharper, code is written faster, emails sound more refined.

On the surface, adoption looks healthy.

But this is where a dangerous illusion begins.

Because the real shift is not:

Do you have AI?

It is:

How much AI are you actually consuming to amplify your work?

In earlier technology cycles, tools improved productivity linearly. In AI, usage directly correlates with capability. The more you use, the more you learn, the more you refine, and the more leverage you create.

Which brings us to a slightly uncomfortable but necessary idea:

Under-using AI is not efficiency. It is underperformance.

Why Token Consumption Is the New Metric

Token consumption may sound like a technical or billing concept, but it is actually a proxy for something much deeper:

• How much intelligence are you leveraging?

• How much experimentation are you doing?

• How much iteration are you allowing?

• How deeply is AI embedded in your workflow?

An engineer who uses AI occasionally is improving efficiency.
An engineer who uses AI extensively is redefining how work is done.

That difference compounds quickly.

But there is a second-order effect that is even more important.

Where does the value of that consumption go?

If all token consumption flows to external platforms, then while productivity improves internally, economic value accumulates externally. Over time, this creates a dependency loop—where capability increases, but control does not.

This is where the conversation shifts from productivity to strategy.

The Strategic Layer: Why Sovereign AI Matters

The idea of sovereign cloud and in-house models is often misunderstood as a defensive or political stance. In reality, it is an economic and architectural necessity in an AI-driven world.

When organizations build or participate in sovereign AI ecosystems:

• Token economics can be partially internalized

• Data remains within controlled boundaries

• Models evolve with contextual relevance

• Costs become predictable and optimizable

• Most importantly, learning stays within the organization

Because every AI interaction is not just usage—it is training, refinement, and feedback.

If that loop sits outside, the intelligence compounds elsewhere.

If it sits inside, the advantage compounds internally.

This is the real reason why sovereign infrastructure is not optional in the long term.

What an AI First Strategy Actually Means

“AI First” has become a popular phrase, but in most organizations, it is still interpreted too narrowly. It is seen as tool adoption rather than operating model transformation.

A serious AI First strategy must be anchored in three core questions, and these must be asked in every department, not just technology teams:

1. How can we do our job better?

Not just faster, but with higher quality, deeper insight, and fewer errors. AI should improve decision-making, not just output generation.

2. How can we do it faster?

Speed matters, but only after clarity. AI should compress cycles—design to execution, query to insight, issue to resolution.

3. How can we serve our customers better?

This is the ultimate test. Personalization, responsiveness, and anticipation of needs should all improve measurably.

But beyond these, I would add two more dimensions that are often ignored:

4. How do we increase our “AI leverage per employee”?

This is where token consumption becomes meaningful. Every role should be re-evaluated not by output alone, but by how much AI amplification it is using.

5. How do we ensure the learning loop stays internal?

Using AI is not enough. Capturing the patterns, prompts, workflows, and insights generated through that usage is what creates long-term advantage.

The Behavioural Shift Organizations Must Address

There are two opposing behaviours emerging in the AI transition, both of which need correction.

The first is the ridicule of AI usage. Work that is visibly AI-assisted is often dismissed as less valuable. This mindset is dangerous. In a world where AI is foundational, the ability to use it effectively is a sign of capability, not weakness.

The second is the illusion of intelligence. Individuals presenting AI-generated output as their own thinking creates a false sense of competence. Organizations must encourage usage, but also enforce understanding.

AI should augment thinking, not replace it.

The Real Risk: Comfort

The biggest risk in this transition is not lack of technology.

It is comfort.

Organizations may feel they are progressing because they are “using AI.” But if that usage is shallow, sporadic, or constrained, they may be falling behind without realizing it.

This is exactly the kind of situation Grove warned about.

The danger is not visible decline.
The danger is invisible stagnation.

The Question That Matters

If you were to step back and look at your organization objectively, ask a simple question:

Are we consuming AI at a level that changes how we operate, or are we just using it enough to feel modern?

Because those two paths lead to very different outcomes.

Closing Thought

AI is not just a tool. It is a new layer of leverage.

And in this layer, advantage will not come from access alone, but from depth of usage and control of the underlying stack.

Those who recognize this early will not just improve productivity.
They will redefine where value is created.

The rest may continue to operate efficiently—
but within someone else’s system.

India AI: Beyond the Applause, Where Do We Really Stand

The applause was deafening.

Screens glowed with GPU counts. Words like Sovereign AI, AI Factories, and National Scale drifted confidently across the panels. There was a palpable pride in the room—and rightly so. But as I sat there listening, I found myself thinking less about what was being celebrated and more about what remains unresolved.

Applause is easy. Execution is not.

Somewhere between the ambition of the stage and the reality of the shop floor lies the question that actually matters: Beyond the standing ovation, where do we actually stand?

Momentum is Real. Direction is Unclear.

Let’s be clear: India is not "late" to AI. We possess what most nations envy: Digital Public Infrastructure (DPI) at a population scale, immense talent, and clear political intent.

We are building GPU clusters. We are talking about indigenous compute. As a staunch believer in the Make in India hardware ecosystem, I see this as more than just industry—it is strategic insulation. Compute sovereignty isn't a luxury; it’s the floor. Building infrastructure locally isn't symbolism; it’s the only way to control our technological destiny.

Are We Building Engines—or Just Renting Them?

Much of the current excitement circles around hardware:

Who has the GPUs?

How many Petaflops?

Whose cluster is bigger?

These are the right metrics for a start, but the wrong metrics for leadership. True sovereignty isn't just hosting someone else’s model on a local rack. It’s about owning the entire stack: orchestration, energy resilience, and optimization. The Insight: Make in India cannot stop at the rack. It must extend to the software stacks and security layers that make the hardware hum.

From Model Size to Market Maturity

At summits, the gravity always pulls toward model size. Bigger. Faster. More languages. But the question I keep asking is: How many Indian enterprises are truly AI-ready?

Not piloting. Not "exploring." Deploying.

How many banks have AI fully baked into credit risk workflows?

How many factories rely on predictive maintenance every single shift?

How many government departments have operational AI reducing citizen friction?

AI only becomes meaningful when it "disappears" into the process. We are world-class at Proof of Concepts (PoCs). We must now become world-class at Production.

The Real Advantage We Don’t Fully Use

India’s greatest asset isn’t a specific model; it’s our data footprint. UPI transactions, multilingual nuances, and diverse behavioral data are our "oil."

But scale without structure is just noise. Data governance, curation, and standardization are less glamorous than unboxing a new GPU, but they determine whether our AI is intelligent by design or accidental by data. Sovereign AI: A Word We Must Use Carefully

Sovereignty isn't isolation. It isn’t replacing global brands with domestic slogans. It is owning enough of the stack—Compute, Data, Security, and Deployment—to avoid strategic dependency. Resilience is built in the lab and the factory, not declared at a press conference.

Talent: Scale vs. Depth

We produce extraordinary engineers, but frontier AI demands more than engineering volume. It requires "Patient Capital"—investors willing to fund long-horizon research and deep model experimentation.

The question for our ecosystem is simple: Are we building the operators of AI, or the originators? Both are needed, but only one writes the global rules.

So, Where Do We Stand?

India is not behind, but we are not yet ahead. We are at a pivot.

The next decade will not reward announcements or headline numbers. It will reward:

Depth over Display

Deployment over Demos

Ecosystems over Isolated Wins

The applause at the summit was deserved. But applause is a moment. The real work is quiet, technical, and relentless. And it starts now.

A Personal Reflection

Walking out of the summit, I felt two things at once: Optimism, because India is finally serious, and Responsibility, because seriousness must translate into execution.

The question is no longer whether India will participate. The question is whether we will build foundations strong enough to shape the rules—not just follow them.