The Hidden Cost Of Compute

India is in the middle of a full-blown data centre buildout, with global tech giants stepping in, backed by government support across policy, land, and infrastructure. From global hyperscalers to domestic infrastructure players, billions of dollars are flowing into server farms, cloud regions, and AI-ready capacity across the country. 

To put it into perspective, the market size reached a valuation of ₹9.33 Lakh Cr in 2025 and is projected to more than double to ₹20.53 Lakh Cr by 2030.

The scale of this push is already visible on the ground. Google has begun constructing a $15 Bn AI data centre hub in Visakhapatnam to deliver large-scale computing power, enhance global connectivity, and support India’s fast-growing digital and AI-driven economy.

Microsoft has also committed $3 Bn to expand its cloud and AI infrastructure in India, while AWS continues to scale its local footprint alongside large capacity additions from players like AdaniConneX, STT GDC, and CtrlS across Mumbai, Chennai, and Hyderabad.

But with all this momentum, more pressing questions emerge. 

Does India have enough natural resources to sustain this boom? Who are the true beneficiaries of India’s data centre surge? Is this infrastructure building domestic capability or enabling global extraction? What’s the real cost of sustaining this infra? Finally, does India control the value it is helping create? Let’s try to answer some of these questions in this edition of The AI Shift. 

Numbers Look Good… But Look Closer

India’s data centre story looks compelling – lower land and labour costs, improving infrastructure, and a rapidly digitising economy. All these factors position India as a natural contender in the global race to host cloud regions and AI workloads. 

But beneath this momentum lies a more complex reality. The same factors that make India cost-effective also mask the true price of building and sustaining this infrastructure at scale, particularly when it comes to resource use and long-term environmental impact.

According to Amit Chaurasia, founder and CEO of cloud services company Dataneers, the cost advantage is precisely what is driving India’s emergence as a data centre hub. 

“The availability of technical expertise at lower costs, combined with state-level incentives, has made building and operating data centres in India significantly more cost-effective compared to Western markets,” Chaurasia said.

This essentially means that companies investing in AI infrastructure in India stand to save millions in capital, while the burden of operating within those cost constraints ultimately falls on India’s resources and workforce.

Adding weight to this is a recent study by the Council on Energy, Environment and Water (CEEW), which states that data centres consumed about 0.5% of India’s electricity in 2025 and 150 Bn litres of water annually. Both of these are predicted to more than double by 2030.

A single 100 MW hyperscale facility can consume up to 2 Mn litres of water per day, depending on cooling technology and climatic conditions. The choice between water-based cooling and energy-intensive air cooling becomes a trade-off between two scarce resources.

Power tells a similar story. Energy costs account for 60-70% of a data centre’s operational expenses. In India’s case, this often translates into preferential power arrangements, open-access procurement, or state-backed incentives that prioritise reliability and cost efficiency for data centre operators.

In practice, this means the choices being made today around location, cooling, and power sourcing will shape resource-use patterns for decades to come.

Who Is The Real Beneficiary?

The demand side of India’s data centre boom reveals a different reality. According to the founder and CEO of Utho Cloud, a cloud infrastructure provider, Manoj Dhanda, roughly 65-70% of capacity today is driven by global hyperscalers. “AWS, Azure, Google Cloud, they come in early, pre-book at enormous scale, and anchor entire facilities.”

Hyperscalers provide the guaranteed demand that makes large-scale infrastructure viable. Without them, facilities simply do not get financed.

Indian AI startups and enterprises, despite numbering over 3,000, occupy a very different position in this ecosystem. “Their share of infrastructure consumption is still in the low single digits. Most are running on hyperscaler credits, not paying for dedicated compute,” Dhanda said.

Training even a mid-scale AI model on high-end GPUs like H100 can cost between $80,000 and $120,000 per month. For most Indian startups, that is not a sustainable cost structure.

What emerges is a layered market. At the top are global enterprises and hyperscalers consuming large-scale infrastructure. At the bottom are startups operating on subsidised access to that infrastructure, without complete ownership or control.

However, rising GPU costs, data localisation mandates, and enterprise digitisation are expected to push domestic demand higher.

Dhanda estimates that Indian companies could account for 15-20% of data centre demand over the next four to five years.

“India has no shortage of AI ambition, but the demand is not uniform. A significant share of current GPU consumption is still tied to global workloads,” said Karan Kirpalani, the chief product officer at Neysa. 

In the near term, he added, India’s data centre capacity will be absorbed through a hybrid model – a mix of global hyperscaler demand, enterprise AI adoption, and public sector use cases – before domestic, production-grade workloads begin to dominate.

Concluding the “who is the real beneficiary” debate, A S Rajgopal, the MD and CEO of NxtGen Cloud Technologies, said, “The data centre boom is very real, and it is necessary. But today, it is largely serving global technology companies and their customers. India is getting jobs, tax revenues, and infrastructure credibility. What it is not yet getting is ownership of the value chain.”

Who Owns The Value In India’s Data Boom?

India’s data centre boom is unfolding at scale. From large data centre campuses coming up around big cities to frequent announcements of new cloud regions, capacity is expanding faster than ever. 

Billions of dollars are being committed across land acquisition, fibre networks, and energy tie-ups, with both global tech giants and domestic infrastructure players racing to secure an early foothold. State governments, too, are competing to attract these investments through policy incentives, faster clearances, and dedicated data centre parks.

India’s data centre boom is undeniably building capacity. It is accelerating digital infrastructure, attracting capital, and positioning the country as a critical node in the global AI and cloud ecosystem. 

However, much of this infrastructure is being shaped by global demand, financed by foreign capital, and anchored by hyperscalers whose primary value chains sit outside India. The money made from AI workloads running on this infrastructure doesn’t necessarily benefit India directly.

The resource question is also pressing. Power and water consumption are rising alongside capacity, and current policy frameworks are still evolving to account for the long-term trade-offs.

Despite the rapid pace, India is still in the early stages of building its AI infrastructure – which means the direction it takes from here is still open to choice. Will it continue to be a low-cost place where global companies build and run data centres, or try to fetch more value by building its own demand, investing in its own compute capacity, and setting policies that ensure more benefits stay within the country?

Top Stories From India & Around The World

• Palo Alto Networks To Acquire Portkey: The US-based cybersecurity giant plans to acquire the homegrown AI infrastructure startup to strengthen its AI security stack. The deal, expected to close by FY26, will integrate Portkey’s AI gateway into Prisma AIRS, enabling centralised control, monitoring, and security for AI agents at scale.
• Fractal Overhauls Operations: The listed AI company has restructured its business around enterprise AI, appointing three new AI chiefs. The company will now operate across three pillars spanning AI transformation, foundations, and workforce, as it doubles down on agentic AI deployment.
• Maharashtra’s New AI Policy: The state has cleared a new AI policy targeting ₹10,000 Cr in investments and 1.5 Lakh jobs by 2031. The plan includes AI innovation cities, a compute infrastructure with 2,000 GPUs and startup funding support. Meanwhile, Goa has released a draft AI policy for public consultation.
• SoftBank’s $100 Bn AI Infra Bet: The Japanese tech giant is building a new venture, Roze AI, focused on automating data centre construction using robotics. The company is already considering a potential $100 Bn IPO as it looks to capitalise on the growing demand for AI infrastructure globally.
• Stripe Expands Agentic Payments: The fintech major has introduced agent wallet capabilities via Link, allowing AI agents to make payments with user approvals. It also added support for Pix, stablecoins, and previewed UPI acceptance, alongside new analytics tools to track conversion and payment performance.
• Google Upgrades Gemini: Users can now create and download files like PDFs, Word docs, Excel sheets, and Google Workspace formats directly from prompts within Gemini, streamlining the workflow from ideation to ready-to-use documents without switching apps.

The Weekly Buzz: How People Seek Life Advice On AI

The role of AI assistants is quietly expanding beyond productivity into deeply personal territory. New research from Anthropic reveals that users are increasingly turning to its chatbot, Claude, for guidance on real-life decisions, from careers to relationships.

Analysing over 1 Mn conversations, the company found that around 6% involved users seeking personal advice, with the majority concentrated in health, careers, relationships, and finance. This signals a shift where AI is not just answering questions but actively shaping decision-making.

However, the study also highlights a key risk: “sycophancy,” where AI overly agrees with users instead of offering balanced perspectives. While this occurred in about 9% of conversations overall, it rose significantly in relationship-related queries, where emotional context and one-sided narratives dominate.

To address this, Anthropic used these insights to retrain newer models like Claude Opus 4.7 and Mythos Preview, reducing sycophantic responses and improving the model’s ability to challenge users constructively.

The findings point to a larger shift in AI’s role. As people increasingly rely on AI for high-stakes personal decisions, the question is no longer just how intelligent these systems are, but how responsibly they guide.

Startup In The Spotlight: EarthSync

As organisations accelerate decarbonisation strategies, renewable energy decisions are becoming more complex. Enterprises must evaluate tariffs, regulatory frameworks, storage technologies, transmission costs, and emissions targets simultaneously, making deployment both data-heavy and time-sensitive.

Founded in 2024 by Mehul Kumar and Rajat Singh, EarthSync is building an AI-powered platform to simulate, optimise, and execute renewable energy investments. Its platform integrates technical, financial, and policy analysis across the clean energy lifecycle, using standardised templates that account for variables like time-of-day tariffs, taxation norms, and grid constraints.

At its core is a physics-informed machine learning engine that helps enterprises assess risk scenarios, and evaluate sustainability outcomes across solar, wind, and battery storage deployments.

Beyond modelling, EarthSync focuses on execution. A collaboration layer centralises inputs from project teams, advisors, and procurement partners, reducing coordination gaps. CXO-ready dashboards compare multiple deployment scenarios across 50+ metrics, enabling faster decision-making and board-level approvals.

Targeting industrial energy consumers and advisory firms, EarthSync is positioning itself as a decision intelligence layer for decarbonisation. As renewable portfolios grow more distributed, such platforms could bridge the gap between sustainability ambition and on-ground execution, tapping into a global market projected to reach $4.7 Bn by 2033.

[Edited by: Shishir Parasher]
[Creatives by: Varshita Srivastava]

The post The Hidden Cost Of Compute appeared first on Inc42 Media.