Sify IPO represents India’s push for data center colocation, giving AI startups affordable access to India’s data center capacity, even as the Google AI hub and other hyperscalers dominate investment, the data localization rules make neutral domestic options essential for startups navigating compliance and cost pressures.

As Chennai-based Sify Infinit Spaces filed papers in October 2025 for an INR 3,700 crore initial public offering, it was a moonshot moment for Indian Data Center infrastructure. The IPO: INR 2,500 crore fresh issue plus INR 1,200 crore offer-for-sale—marking India’s first pure-play data center listing.

The Sify IPO proceeds will fund specific expansions, including about INR 465 crore for Tower B at the Chennai campus 02, roughly INR 860 crore for Towers 11 and 12 at Navi Mumbai’s Rabale facility, and approximately INR 600 crore to repay borrowings.

Meanwhile, in October 2025, Google announced a $15 billion joint venture with Adani for a Google AI hub in Visakhapatnam, forming a gigawatt-scale campus dedicated primarily to Google Cloud workloads and partner services.

These parallel moves highlight a widening gap. Consequently, hyperscalers like Google, Amazon Web Services, and Microsoft are pouring billions into proprietary data centers that prioritize internal needs.

In contrast, neutral operators such as Sify offer data center colocation that any company can rent on transparent terms. For AI startups with limited capital, this distinction determines whether they can afford cutting-edge compute or get squeezed out by single-vendor pricing and lengthy contracts.

Why colocation cuts costs and time

Sify rolled out hourly, pay-per-use data center colocation at NVIDIA DGX-Ready facilities in Chennai, Noida, and Navi Mumbai, bundling hosting, power, cooling, and infrastructure support so AI startups can deploy H100, H200, and GB200 GPUs without upfront capital outlays.

Traditional hyperscaler clouds charge hourly rates that appear low but quickly accumulate hidden fees, data transfer egress, managed-service premiums, and proprietary add-ons, pushing annual bills significantly higher than advertised GPU costs. By contrast, Sify’s model quotes an all-in hourly rate that includes power, cooling, and base services, reducing the risk of surprise charges.

Consider real numbers. An AI startup training a language model for 6 hours a day needs roughly 1 NVIDIA H100 GPU. On AWS, an on-demand H100 instance costs about $8 per hour; over 6 hours a day for a year, that totals approximately $17,520 in compute alone.

However, data transfer out of AWS costs $0.12 per gigabyte, so moving 10 terabytes of training data monthly adds another $14,400 annually, and managed service layers can tack on an additional 20–30 percent.

Total realistic AWS spend climbs toward $30,000–$40,000 per year for that single-GPU workload. In contrast, Sify’s hourly colocation rate of INR 600–INR 800 per hour translates to roughly INR 13–17 lakh ($15,500–$20,500) annually for the same six-hour daily usage, including power and bandwidth, representing a 30–50 percent saving.

Moreover, hyperscaler spot instances promise steep discounts: Google Cloud H100 spot instances can drop to $2.25 per hour but carry termination risk with only two minutes’ notice, potentially wiping out hours of training progress. Sify’s dedicated colocation guarantees uptime under service-level agreements, so AI startups avoid the gamble of interrupted runs.

The vendor lock-in burden

Hyperscale clouds create dependency through proprietary services. If an AI startup builds its stack on AWS Lambda for serverless compute, DynamoDB for databases, and SageMaker for machine learning, migrating to Hyperscaler later requires re-architecting code, re-training pipelines, and transferring massive datasets at punishing egress rates.

Industry data shows exit costs ranging from $50,000 to $500,000 and timelines of 6 to 12 months. Furthermore, once locked in, customers report renewal price hikes of 20–50 percent because providers know switching is prohibitively expensive.

However, Data center colocation avoids this trap. AI startups own or lease their servers, install standard NVIDIA GPUs, and run the software stack of their choice. Suppose they decide to switch from Sify to another colocation provider, such as CtrlS or Yotta. In that case, they relocate physical hardware, a process taking one to two weeks and costing roughly INR 2–5 lakh for transport and reconfiguration.

Carrier-neutral facilities allow startups to connect to multiple internet service providers simultaneously, enabling them to negotiate competitive bandwidth rates and sidestep single-vendor pricing power.

Data localization and compliance

India’s regulatory environment increasingly demands data sovereignty. The Reserve Bank of India requires all payment system data to reside in India, and the Digital Personal Data Protection Act of 2023 tightens rules on cross-border transfers of personal information.

Using colocation in Chennai, Mumbai, or Noida ensures servers remain physically in India, simplifying audits and eliminating the risk of inadvertent foreign access.

Hyperscaler compliance is conditional. AWS and Google Cloud both operate India regions—AWS’s ap-south-1—but users must explicitly configure data residency, enable encryption with locally held keys, and sign specific contracts.

Default settings often replicate data globally for disaster recovery, which can trigger violations. Additionally, the US CLOUD Act allows American law enforcement to compel disclosure of data held by US companies, even if stored abroad, raising sovereignty concerns for Indian fintechs and health platforms.

Consequently, regulated industries, banking, insurance, healthcare, increasingly prefer domestic colocation where physical control and transparent audit trails meet statutory requirements without complex cloud configurations.

The Google factor and market balance

Google’s $15 billion Google AI hub in Visakhapatnam will span roughly one gigawatt of capacity, making it the company’s largest investment outside the United States and positioning Andhra Pradesh as a global compute node.

The facility, developed jointly with Adani Infrastructure, will integrate renewable energy sources and advanced cooling, targeting commissioning by 2030. However, this Google AI hub is purpose-built for Google Cloud services, Google DeepMind workloads, and select enterprise partners who negotiate directly with Google.

That structure leaves thousands of smaller Indian AI startups competing for scraps or paying premium cloud rates. Suppose India’s data center capacity consolidates into a handful of hyperscaler campuses.

In that case, Google’s gigawatt hub, AWS Mumbai expansions, Microsoft’s investments—then pricing power shifts away from startups, and neutral alternatives become scarce. Therefore, the Sify IPO matters, as it channels INR 2,500 crore of fresh capital into open-access facilities that any company can rent on standard terms, counterbalancing the Google AI hub’s private model.

India’s data center capacity stood at roughly 1.2–1.7 gigawatts in early 2025 and is projected to reach 8–9 gigawatts by 2032, with colocation operators such as Sify, CtrlS, and Yotta accounting for a large share of the multi-tenant supply.

Yet hyperscale builds are growing faster, 21.5 percent compound annual growth rate versus colocation’s steadier pace, so without continued investment in neutral facilities, the market risks tipping toward single-vendor dominance.

Real startup math

An AI health-tech startup in Bengaluru with INR 5 crore Series A funding wants to train a diagnostic model for six months. Building its own one-megawatt data center would cost INR 2–3 crore in construction, INR 1.5–2 crore in servers and GPUs, and INR 50–100 lakh for power and networking, totaling INR 5–6.5 crore, exceeding the entire funding round before hiring a single engineer. That option is off the table.

Renting from Sify’s hourly colocation costs roughly INR 1 lakh per month for moderate GPU usage, or ₹6 lakh over six months, leaving INR 4.94 crore for salaries, marketing, and R&D. Using AWS with reserved instances might cost INR 10–15 lakh over the same period but locks the startup into a one-year contract and builds dependency on AWS-specific tools.

If the startup later wants to switch providers, migration costs balloon to INR 20–50 lakh in engineering time and data transfer fees.

Thus, data center colocation delivers immediate flexibility at a predictable cost, a critical advantage when 9 out of 10 startups fail and founders need the freedom to pivot without sunk infrastructure.

Why five operators aren’t enough

Today, NTT Global, Sify, ST Telemedia, Airtel Nxtra, and CtrlS control roughly 88 percent of India’s existing colocation capacity. Meanwhile, hyperscalers are adding dedicated gigawatt-scale sites that won’t appear in shared inventories.

If neutral capacity stagnates while proprietary builds accelerate, AI startups face a shrinking pool of affordable options, and the handful of colocation providers gain pricing leverage. The Sify IPO injects competition by funding two major expansions and enabling faster rollout of NVIDIA-certified capacity in Chennai and Navi Mumbai, where demand is acute.

Moreover, public-market accountability forces transparent reporting of utilization, pricing, and customer mix, which private operators can obscure. That transparency benefits startups negotiating contracts because they can benchmark rates and service levels against disclosed metrics.

What’s next?

Sify’s DGX-Ready certification means its facilities meet NVIDIA’s rigorous standards for high-density AI workloads, 130-kilowatt-per-rack configurations with liquid cooling, so AI startups can deploy the latest GB200 and Blackwell-generation chips as they become available.

The company’s pay-per-use model also extends to its CloudInfinit+AI managed service, letting startups rent GPU instances by the hour without managing physical servers, bridging the gap between full colocation and hyperscale cloud.

In parallel, government initiatives like the IndiaAI Mission allocate subsidized GPU clusters, roughly 18,000 H100/H200 units distributed via Yotta, E2E Networks, and Tata Communications, offering another avenue for early-stage teams. However, these programs serve a limited cohort through competitive application processes, so the broader market still depends on commercial colocation and cloud capacity.

As India’s data center capacity scales toward nine gigawatts, the mix of neutral colocation, funded by offerings like the Sify IPO, and proprietary hyperscale sites such as the Google AI hub will determine whether hundreds of AI startups can access compute on fair terms or whether a few giants set prices and priorities.

For now, Sify’s expansion represents a tangible commitment to keeping infrastructure open, affordable, and compliant with India’s data localization mandates, qualities that matter most when startups have more ideas than capital.

Stay ahead of the curve and follow IndiaTechDesk on Facebook, Twitter and Linkedin for in-depth news of market trends, funding updates, and regulatory changes affecting startups in India.