India is building its AI factories. Siemens sees a once-in-a-generation growth story

Walk through Thane on the outskirts of Mumbai today and you can watch the future of Indian computing being poured in concrete and strung on high-voltage lines. Data centre campuses are rising. Utilities are pulling in new connections. And the people supplying the electrical backbone, the cooling and the software that runs it all are barely keeping pace with demand.

“It is happening, and it will increase,” says Robert Demann, Executive Vice President of Siemens India and head of Smart Infrastructure for South Asia. “Look at Mumbai, for example. Across the Thane area, you can see massive investments in datacenters, with utilities bringing in high-voltage connections.”

For CIOs weighing where to place their next workloads, and for technology leaders watching where the world’s next wave of AI capacity will land, Demann’s message is both a market signal and a design brief: India’s growth ambitions are real, the build-out is underway, and the winners will be those who treat the data centre as a factory rather than a building.

A growth engine with a tailwind
India’s macro picture explains the confidence. In its October 2026 World Economic Outlook, the IMF raised India’s growth forecast to 7.0 per cent for both 2026 and 2027 on a fiscal-year basis, up 0.6 percentage points from its July estimate, and described the country as one of the fastest-growing major economies, citing resilient domestic demand and robust services activity.

That momentum is flowing straight into digital infrastructure. JLL’s mid-year report puts India’s data centre capacity at about 1.6 GW in June 2026 and sees it reaching roughly 6 GW by 2029, a build-out that would need around US$110 billion in capital. CBRE expects India to close 2026 near 2 GW and counts cumulative investment commitments of about US$173 billion since 2021, including US$38 billion in the first half of this year alone.

Policy is accelerating the curve. The Union Budget 2026-27 announced a tax holiday until 2047 for foreign companies that provide global cloud services using data centres in India, a long-dated signal of certainty that industry leaders have called a strategic declaration rather than a mere incentive.

Demann sees the same story from the ground. “A lot is happening across the whole of India,” he says. “You see airports coming up in Tier 2 and 3 cities, a lot of investments are happening in ports and metros, mainline trains. Those are all topics where we play a key role.” Then comes the line that sums up Siemens’ read of the market: “There’s hardly any area where our businesses are not finding opportunities.”

Why India’s demand is structural
Demann points to fundamentals that go well beyond any single technology cycle. Energy consumption per household in India is roughly a tenth of that in Western countries – “so even if you put all the energy efficiency measures in place, with growing wealth this is going to go up.” Climate-driven cooling needs are rising. Electric vehicle production is scaling, and e-mobility schemes are being tendered. “The main motivator being two-fold: it is more efficient than traditional vehicles and they can save up to 30%. Also, it is actually helping clean the air in the city.”

On the digital side, the demographics are hard to argue with. “India has one of the largest bases of mobile users in the world. We have overtaken China,” Demann notes, and the base is far larger than the US or Europe, where users number around 300 million. Layer on UPI, e-commerce and public digital platforms such as DigiYatra, and the case for data centres builds itself. “The average age of Indian mobile phone users means that people are more willing to try it out.”

There is also an untapped prize. “Around 20% of the world’s data is produced in India, but very little of it is stored here,” he says. “Building this infrastructure could change that. India could then use its data to create knowledge, and turn that knowledge into decisions.”

From megawatts to gigawatts: The data centre reinvented
Today, most of what is being built in India is still the traditional data centre, and Demann is candid that AI factories are early days. But he describes the direction of travel in vivid terms. “If you look at data center today, typically you are talking megawatts – 10 to 40 or, if it’s a big campus, somewhere below 100. And that is shifting towards gigawatts.”

The reason is chip physics. As processing power rises, so does energy draw, and air cooling runs out of road. “That requires people to slowly start looking at liquid cooling instead of air cooling because physically it is nearly impossible to manage with air cooling.” Density changes the electrical story too. “If you look at the energy consumption in a rack, that’s going to consume as much as a 30-storey building. Now imagine one data center has a building full of such racks. If not properly planned, it can actually destabilize the whole power supply for a city.”

Several shifts follow, each one an opportunity for suppliers who can deliver them:

• Power architecture changes. “With higher power density, we will also see a shift from AC to DC power,” Demann says, and the electrification layer “requires different products, which are being developed.”

• Cooling changes. Siemens, which optimises air-based HVAC today, expects the market to move to liquid cooling, “so this is the second opportunity.”

• Reliability changes. In the past a failed system fell back to a failover concept. “But now with the Giga AI factories coming up, you can’t afford the grid to go down.”

• Operations change. “Previously you may have had half an hour or an hour to react to some effects… You have to be able to react in seconds and humans cannot do it.”

The new scorecard: from PUE to Watts per Token
Perhaps Demann’s most striking prediction is about how data centres will be measured. Today’s headline efficiency metric is power usage effectiveness. Tomorrow’s, he argues, is the energy cost of intelligence itself.

“We will be looking at Watts per Token – it takes milliseconds to create a token. And that’s a totally different KPI.” And it unlocks a different conversation with the business: “PUE measures how efficiently a datacenter operates; it is one of the key KPIs. Tokens, however, represent the conversion of information into knowledge and decisions. So, from my perspective, the conversation shifts from measuring cost and efficiency to measuring the value and ultimately the revenue a datacenter can generate. That is a fundamental change.”

He frames the stakes using a line from Siemens AG CEO Roland Busch: “Try to recall what the world looked like before electricity.” Siemens, Demann notes, “comes from” the second industrial revolution, and Busch believes AI will bring a comparably fundamental change.

“It’s called an AI factory for a reason”
The framing matters for CIOs, because it changes what good looks like. “There’s a reason why it’s called AI factory and that is because it would be also operated like a factory,” Demann says. “You can do simulation, and like a factory you can switch to produce different type of products, it can ensure that you get repeatable output and it’s built for scale.”

This is where Siemens’ heritage becomes an advantage. The company’s strength in physics-based simulation and digital twins, honed in automotive and manufacturing, transfers directly. “Today, when you develop a car, you first design it in a computer, simulate it and then build a model. The idea is to shorten the whole development life cycle. This is going to be replicated with data centres.”

It is already tangible. At a recent Siemens Innovation Day, the company’s Corporate Technology team demonstrated digital twins of components inside a coolant distribution unit, which can be combined into a twin of the complete unit. “You can then simulate workloads coming in and anticipate the cooling capacity required… while also taking account of the weather on a given day.” Weather matters in India.

On a scorching summer afternoon the grid can supply less, while on a winter day somewhere like Kashmir “you can go full load.” As Demann puts it, “that kind of simulation is unique to us.”
Siemens’ bet on India’s talent underpins this. At its Innovation Day in March, the company said it has more than 700 digital references deployed across India with 25 ecosystem partners, and over 10,000 engineers, software architects and domain experts working on industrial AI, digital twins, automation software, electrification and cybersecurity, much of it from global innovation centres in Bengaluru and Pune.

Autonomy, with guardrails
Demann is careful to separate hype from deployment. “Today, many people talk about smart cities, smart buildings and so on. But in most cases, what is actually happening is the detection, collection and display of information. Not much is done with that information.” The next step, semi-autonomous operation, is here: systems that detect issues and predict failures weeks ahead. Full autonomy, where AI takes the decision for straightforward situations, will arrive more slowly, and deliberately.

“This is where industrial AI is fundamentally different from general-purpose AI. In an industrial environment, you need to know with a very high degree of certainty what is going to happen, because if something goes wrong, it can affect safety, operations or data security.”

The results from early deployments are compelling. “In one European industrial facility, AI helped reduce unplanned maintenance events by 70%. Think about what that means: fewer outages and, as a result, more revenue. In another case, autonomous agents helped reduce energy consumption in hospitals by 40%.” Closer to home, at Siemens’ own Kalwa campus, a system now optimises a chiller by using weather data, past occupancy and seat-booking information to predict demand, with no custom programming for a hot day. “The savings may not always be dramatic, but they are tangible. These are straightforward business cases with a short-term impact.”

Autonomy also answers a human problem. It preserves the know-how of retiring experts and helps compensate for skills shortages. “Autonomous systems can help compensate for a shortage of expertise and make the human operator more effective,” Demann says. And demand is surfacing in India, in hospitals, agriculture and manufacturing. “Is it happening at huge scale yet? No. But people are looking at it, and positive early experiences are helping adoption.”

The grid has to become intelligent too
None of this works without power, and Demann argues India is moving on it. “There are two key inputs for an AI factory: energy and data. Without them, a datacenter cannot do anything.” The country is investing heavily in generation, transmission and distribution, but he adds a point that CIOs should note: physical build-out alone will not be enough. “AI-capable grid software can optimize existing grid capacity and support some of this growth.”

He sees the data centre itself changing roles. “Today, a datacenter is largely an energy consumer. In future, it will become a prosumer: it will generate and consume energy, and it may also return residual heat to a wider network for heating or cooling purposes.” Hyperscalers are already exploring their own generation, and their strong renewable commitments add variability that the grid must absorb. The implication is bold: “We will need not only autonomous datacenters, but also an autonomous grid, one that can stabilize supply across different energy sources.”

What CIOs should take from this
Demann’s closing themes read like a checklist for any technology leader planning AI infrastructure in India or anywhere:

Orchestrate, don’t optimise in isolation. Asked whether power, cooling or skills is the biggest constraint, Demann refuses to pick one. “It is really about orchestrating all these different components… no single company can solve all these problems. You need a connection between policy and a group of companies with expertise in different areas.”

Expect standards to move. “AI factories are not going to appear overnight… The standards are evolving too; they are not yet fixed.” His advice is to plan generation and transmission capacity first and let the rest develop in parallel.

Design for security from the first component. “You cannot design cybersecurity only after a solution has been assembled; it has to begin with the individual components, because the risk increases as the system becomes more connected.” He also stresses collaboration among corporations and governments.
Invest in change management. Technology may be ready before people are.

Demann recalls his first ride in an autonomous car, hands hovering above the wheel because he did not fully trust it. “Change management will limit the speed of adoption. It is similar to the early days of ERP.” Leaders will need to answer the questions staff will ask: will it work, what happens if it does not, and what does it mean for my job.

The road ahead
Within six months, Demann expects more activity aimed at AI factories, including a shift to liquid cooling, greater use of simulation, and early examples of digital twins and autonomous operations, though not yet at scale. “It is also a matter of gaining experience and trust,” he says.

That patience sits comfortably alongside the scale of the opportunity. A country with a young, mobile-first population, a policy framework built for long-term investment, a growing economy and a fast-expanding digital footprint is exactly where the next generation of AI factories will rise. For Siemens, the move from measuring cost per rack to measuring value per token is not a threat to its business. It is the next chapter of it.

As Demann puts it, the industry is only starting to grasp the change. The shift from PUE to tokens, he says, “is going to be so significant that many people still cannot imagine what it will look like.” In India, they will not have to imagine for long.