Don’t Make Humans Carry AI’s Cognitive Load

You had a front-row seat as mobile and data went from a luxury to something every Indian takes for granted. Is computing power about to go through the same journey with AI?

I have seen enterprise computing evolve from the mainframe-era EDP (electronic data processing) departments of my early career, through the MIS (management information system) era, when boards and CXOs ran on standard reports, into the client-server era, when back-office functions such as finance, accounts payable and HR were progressively encoded. The cloud and Web 2.0 era then pushed catalogues and customer engagement online, turning what had been back-office systems into the front office.

AI is the next layer on top of that stack. It enhances computing from neural networks and predictive analytics into what I call “situational awareness” for employees, customers and products alike. This is enriched further by generative AI, agentic chains, retrieval-augmented generation pipelines and real-time streaming data, all in service of serving customers faster, helping employees reach outcomes quicker, or making products fit more use cases.

I don’t see this growth curve stopping. Each new generation is becoming more intuitive with devices — from desktops to mobiles and, next, what I expect from wearable and ambient intelligence such as smart glasses. Underlying all of this is the same human demand for instant gratification, depth and personalisation.

The International Energy Agency (IEA) expects data centre power demand to roughly double by 2030, and companies are signing nuclear power deals to keep up. When did sustainability stop being a compliance checkbox and start becoming a boardroom conversation in your industry?

Data centres evolved from server rooms into standards-driven facilities around 2003-04, governed by frameworks such as the Uptime Institute standard and the TIA-942 telecom standard, and supported by engineering disciplines such as IEEE 1394 for resilient electrical design and ASHRAE guidelines for precise mechanical cooling.

Facilities scaled quickly. CtrlS’s first Mumbai campus, planned in 2011 as a 30-megawatt build that I expected would take a decade to fill, reached 80 per cent occupancy within four years. This was driven initially by a bank consolidating its branch infrastructure and then by the digital payments boom.

As AI-era facilities scale into the gigawatt range, sustainability has become inseparable from resiliency and the service-level agreement itself. There is a distinction between Tier III design, which has a defined “risk window” during scheduled maintenance that sits outside the service-level agreement (SLA), and CtrlS’s Tier IV, concurrently maintainable design. In Tier IV, every element — cooling, power, transformers, lines — can be serviced without any impact to uptime, so both scheduled and unscheduled downtime count against the SLA.

That design philosophy took hold once digital infrastructure moved from being a support function to something financial systems became directly dependent on.

What made India, specifically, become a favoured destination for green data centre investment?

There are several concrete comparisons. On solar power, India’s roughly 250 gigawatts of installed solar capacity dwarfs the projected 8 gigawatts of demand from Indian data centres by 2030. India’s roughly 300 days of annual sunshine — compared with about 180 days in Southeast Asia and closer to 110 in parts of Europe — also gives solar panels imported from the same manufacturing base a far better return on investment when deployed in India.

On water, look at Paris, which has spent 23 years cleaning its rivers to bathing standard, making any water draw politically fraught. In the United States, fragmented state-level grids make water use a highly localised and contentious issue.

India, by comparison, is still building out sewage treatment infrastructure in cities such as Delhi and Hyderabad, and data centre operators can step into that gap. Rather than drawing potable water, CtrlS treats grey or waste water on-site and evaporates it. In one Mumbai facility, we draw treated water from a nearby effluent plant that had been discharging into a creek. That project pays back in 14 months on the profit and loss (P&L) alone, while also helping clean the surrounding waterway.

On building design, India’s still-developing infrastructure means new data centre campuses can be built into raw hinterland alongside national grids and ring roads, rather than displacing an existing district’s traffic or economy. This allows sustainability to be designed in from the start rather than retrofitted.

CtrlS has notched several industry firsts — the world’s first Rated 4 solar-powered data centre, LEED Platinum campuses, a captive solar farm in Nagpur, and a 2030 target of 100 per cent renewable power. What was the thinking behind these bets, and which one are you proudest of?

I credit much of this to a revenue-model innovation by CtrlS’s founder. In 2008, he broke from the industry norm of billing customers by rated kVA capacity and instead split contracts into rack rental plus energy billed as a function of the facility’s Power Usage Effectiveness (PUE).

The lower we could push our PUE, the better our margins — and the greener the outcome for the customer. That aligned incentives on both sides. The model was distinctive enough that competitors, including large telcos, adopted it within a couple of years. I credit it with helping CtrlS stay resilient even as international operators such as Equinix entered the Indian market.

That incentive structure drove roughly 200 internal innovations aimed purely at reducing PUE. It also reframed facility managers’ core metric from “no customer complaints” to “how much power are you consuming as overhead?” We pushed design efficiencies at the power infrastructure level, stepping high-voltage grid power down on-site through GIS transformers rather than relying on the 13-17 per cent distribution losses typical of discom networks.

We also recognised that buying renewable power from a third party only made CtrlS a secondary carbon reducer. So the company began building its own generation: a 1-megawatt solar wrap on a Mumbai building, which was later found to violate building code but served as a proof of concept, followed by a 300-megawatt captive solar plant in Nagpur. That made CtrlS a primary carbon producer for its customers, with long-term renewable power agreements filling the gap in cities such as Chennai.

You mentioned Telangana’s power mix. How does the wider grid picture factor into hitting that 100 per cent renewable target?

Telangana is a good case study. Hydro sources, together with growing solar capacity, already put close to 59 per cent of the state’s power on renewable sources. When that is layered on top of CtrlS’s own captive generation, a facility there is already getting roughly 70 per cent clean power.

Looking ahead, I see battery energy storage systems (BESS) as the next inflection point. Our own analysis suggests that once the spread between peak and off-peak production costs crosses roughly four rupees, storing solar power generated during the day for release at night becomes a straightforward business case.

Data centre load actually peaks at night because of backups and scheduled maintenance jobs. BESS could therefore potentially allow operators to pass savings on to customers who shift compute-heavy jobs to off-peak hours.

You’ve watched every technology wave since client-server computing. What’s your honest advice to young managers on working with AI rather than worrying about it — and what can’t a co-pilot do for them?

I draw a distinction between what I call muscle memory and intelligent memory.

Repetitive, mechanical tasks — from onboarding forms and data scanning to routine customer onboarding — are things humans have always been error-prone at, however much they want the outcome to be consistent. I’m comfortable letting AI take those over entirely, much as accountants once relied on cross-checkers to catch manual entry errors.

Where I draw the line is on decisions with cross-generational or multi-year consequences — organisational design, long-term resiliency choices and so on. Today’s AI has no way of knowing what technology landscape will exist three to five years out, or which version of a system will even be running by then.

Those mid-to-long-term, macro decisions still require human intellect, curiosity, pattern recognition and risk judgment. My rule of thumb is: don’t make a human carry a predictable, present-day cognitive load that a calculator — or an AI — is already built to carry.

One myth about data centres you’d like to bust forever?

That data centre power is inherently unsustainable. I believe CtrlS and the broader Indian industry are actively disproving that perception through renewable generation, water reuse and efficiency-linked business models.

And one piece of advice for a room of roughly 26-year-old future leaders?

Engage with current technology as deeply and continuously as possible. I see this hands-on fluency, more than any single skill, as what distinguishes today’s generation of managers from the ones before it.

Sustainability has become inseparable from resiliency and the service-level agreement itself

The lower we could push our power usage effectiveness, the better our margins — and the greener the outcome for the customer