Andhra's Data Centre Policy: What's New, What's Missing, What Can Be Fixed

In April 2026, the Andhra Pradesh government did something no Indian state had done before. It granted a power distribution licence, the kind traditionally held by state utilities, to a technology company.

Google can now act as its own electricity distributor for a 1 GW data centre cluster near Visakhapatnam. The company will break ground on April 28. The investment is approximately $15 billion, making it the largest single foreign direct investment in India’s data centre sector.

The policy that enabled this is broader than a single company-specific facilitation. Any private firm with a power requirement exceeding 300 MW can now apply for a similar licence. Policy documents list Reliance, RMZ, and other large data centre developers as early beneficiaries. The government has cited reliable power, cost reduction, and a mandated 51% renewable energy procurement as the policy’s core objectives.

This is a departure from the status quo, but information available in the public domain leaves several critical questions unanswered. This article examines what is known, what is missing, and whether any tools already exist to address the gaps.

What the Licence Does

Under India’s Electricity Act, 2003, a distribution licence allows the holder to supply electricity within a designated area. Traditionally, these licences have been held by state-owned discoms and, in some limited geographies, by private power utilities such as Tata Power.

The Andhra policy extends this right to non-power entities. Google can now procure electricity directly from generators (renewable projects, captive plants, or power exchange) without routing payments or approvals through the state distribution company. It will use the state’s transmission network, owned by APTRANSCO, and pay a service fee for that access.

The policy includes safeguards to protect existing discoms. Restrictions prevent current consumers from migrating to the new licencee. Standby power arrangements are also required. These provisions suggest that the state has anticipated concerns about revenue leakage.

What the licence does not require is equally important. There is no mandate for Google to build dedicated power plants. No requirement for battery storage. No obligation to feed surplus power back to the grid during peak demand. And the details of the service fee, along with the related terms and conditions, are undisclosed.

To understand whether this is pathbreaking, it helps to distinguish it from existing mechanisms. Under open access, available for years in states like Telangana and Tamil Nadu, a large consumer can buy power directly from a generator but remains within the discom’s regulatory framework. The discom still charges wheeling fees, manages scheduling, and remains the licencee of record. The consumer cannot, for example, set up its own internal distribution network or negotiate directly with the regulator on tariff structures.

The discom licence goes further. It makes the consumer a licencee. In theory, this allows greater operational control and direct regulatory engagement. In practice, given that Google will continue using APTRANSCO’s network rather than building its own poles and wires, the functional difference may be smaller than the legal one. But the precedent of a non-power entity holding a distribution licence is new.

What Remains Unclear

Three gaps stand out in the information available as of now.

Water. Visakhapatnam draws its water from the Yeleru reservoir and groundwater sources. Central groundwater board assessments indicate declining water tables in several blocks of the district. A 1 GW data centre typically consumes between 1.1 and 1.5 billion litres of water annually for cooling. For comparison, a medium-sized Indian city of 500,000 people uses approximately 2.5 to 3 billion litres per year for domestic purposes. By 2028, industry estimates suggest, global data centre water use will equal the indoor water consumption of approximately 18.5 million US households.

The state has issued a power distribution licence but it did not issue a water extraction permit. Under the Andhra Pradesh Water, Land and Trees Act, 2002, groundwater extraction is regulated. No public record exists of a permit granted to Google for this project. The absence of a permit does not necessarily mean extraction is prohibited, as land leases sometimes include implicit water rights, but it does mean there is no publicly disclosed cap, monitoring requirement, or penalty mechanism. The regulatory status of the project’s water use is, therefore, ambiguous.

Grid integration. No public document specifies how Google’s 1 GW load will be integrated into APTRANSCO’s network. There is no disclosed requirement for hourly load forecasting, remote disconnect capability, or real-time data sharing. Alongside, there are reports that households and farms near the project sites already experience power cuts — these predate the data centre project, which is not yet operational.

Renewables firming. The 51% renewable mandate is a floor. But meeting it in practice requires either storage or continued reliance on fossil generation when renewable output falls. India’s grid has limited storage capacity. The Central Electricity Authority’s own planning documents acknowledge that renewable generation must be backed by firm dispatchable capacity, currently coal and gas, to meet 24/7 demand. The licence does not require Google to provide storage or to disclose what share of its power will be backed by fossil fuels during low-renewable periods.

How Other Jurisdictions Have Approached This

China and the United States, both of which are at the forefront of data centre development, have taken sharply different approaches.

China launched the “Eastern Data, Western Computing” framework in the early 2020s. The infrastructure investment (transmission lines, renewable generation, and fibre optic cables) was pursued primarily to balance renewable energy across regions and develop western provinces. Data centres have since been incentivised to locate in those western areas, where solar and wind farms are abundant. The computation results travel east via fibre optic cables while the power does not travel at all. This model rests on the back of approximately a decade of infrastructure investment and now serves as a template for co-locating computation with renewable generation.

The US presents a cautionary tale of grid constraints. According to a report from commercial real estate firm JLL in 2025, the average wait time for a grid connection across North America is now four years. Vacancy rates in North American data centre markets have fallen to 2.3%, down from nearly 10% in 2020, with 73% of all new construction already pre-leased. The strain is such that $98 billion worth of projects were cancelled or delayed in the second quarter of 2025 alone, driven largely by community opposition and utility capacity constraints. Additionally, the Federal Energy Regulatory Commission has implemented rules for what are termed “High Impact Large Loads,” requiring customers drawing more than 300 MW continuously to provide hourly load forecasts and accept remote disconnection if grid stability is threatened.

In contrast, Andhra’s model keeps the load on the existing grid, imposes no backup generation mandate, and does not require load forecasting or remote disconnect capability. The service fee arrangement assumes that the existing transmission network can absorb 1 GW of new continuous load without significant upgrades or new customer obligations. Whether this assumption holds will become clear only after the project becomes operational.

Employment and Other Ancillary Benefits

Beyond power and water, what major secondary benefits accrue from the policy?

For a state with Andhra’s workforce size, the employment yield of any large investment is a metric of interest. By that measure, data centres deliver relatively little.

Data centres are not labour-intensive. Large campuses in the United States employ few permanent staff relative to their investment size. Construction employment is larger but temporary. Specialist technical staff are typically not hired locally.

During the building phase, data centres do create demand for construction, security, catering, and transport. Local suppliers of concrete, steel, and electrical equipment may see a temporary uptick. Once operational, however, the facility requires mainly facility managers, security personnel, and remote monitoring staff - none of these categories generates large numbers of jobs. Unlike manufacturing, a data centre does not anchor a supply chain of component suppliers or logistics providers.

So for a state with Andhra’s employment needs, a 1 GW data centre is unlikely to be a significant source of permanent jobs.

But employment is not the only metric. The state gains three other benefits which are worth noting.

Reputational. Landing the Google deal has placed Andhra in direct competition with Karnataka, Tamil Nadu, and Telangana for large-scale digital infrastructure. The political rhetoric, including public taunts about “neighbours feeling the burn”, suggests that the state values the signalling effect as much as the direct returns. Whether this reputational boost translates into follow-on investment will depend on delivery.

Agglomeration. Data centre clusters, once established, attract ancillary industries: construction, security, facilities management, electrical equipment, and eventually related tech services. The UK’s Slough cluster (approximately 1 GW of capacity) supports roughly 14,000 jobs across direct, indirect, and induced employment. However, the same cluster has strained West London’s grid so severely that it limits new housing development, a trade-off Andhra policymakers may need to consider. Tamil Nadu’s policy explicitly incentivises local MSMEs to provide maintenance services. Andhra has not yet announced similar measures. Agglomeration is possible but not automatic, and it often comes with grid constraints that are rarely discussed upfront.

Connectivity and Location Advantages

Beyond employment and ancillary industries, Vizag offers two location-specific advantages worth noting.

First, undersea cable connectivity. Visakhapatnam is becoming a landing point for multiple submarine cable systems — Google’s own network, Meta’s 50,000-kilometre Waterworth cable, and an open cable landing station developed by Sify. This complements existing landing stations in Mumbai and Chennai, providing route diversity and reducing latency for data transmission between India and Southeast Asia.

Second, port access for equipment. Vizag’s deep-water port, the only 18-metre draft port on India’s east coast, allows heavy lift infrastructure to arrive directly by ship. For data centre developers importing servers, cooling systems, and electrical gear, this reduces logistics costs compared to inland locations.

What about water? The coastal location offers no inherent cooling advantage. Seawater is abundant but saline; freshwater remains the standard for most cooling systems. The state government has claimed that “surplus monsoon water” draining into the Bay of Bengal could be captured for cooling, but no technical detailing or implementation plan has been released. The Human Rights Forum counters that Vizag’s hot climate will require even more water-intensive cooling, worsening existing groundwater stress. The coastal advantage here is theoretical, not proven.

Competitive positioning. Andhra’s innovation, the discom licence, is a regulatory differentiator. Other states compete on tax breaks, power tariffs, and clearance speed. Maharashtra offers 20-year electricity duty waivers and a 100% green energy mandate for dedicated parks. Tamil Nadu offers industrial power at ₹7.25 per kWh and priority renewable allotment. Telangana offers 100% stamp duty reimbursement. No other state has issued a discom licence to a non-power entity. Whether Andhra’s model proves superior or merely gimmicky is an open question.

This is not a criticism of the policy but a clarification of what the policy does and does not deliver (and also its dependency on what other states are failing to deliver).

The Tools That Already Exist

Five gaps in the current licence, i.e., captive power compliance, grid stability, technical standards, water regulation, and grid-sharing, can each be addressed by tools already available. None require new legislation.

The captive power rules gap. The central government recently amended the Captive Power Rules. The updated Rule 3 requires proportionality between ownership and consumption for captive generating plants. If Google’s arrangement qualifies as a captive plant, a reasonable reading, given the scale and direct procurement, then Google must own a demonstrable share of the generation it uses. Paper purchase agreements without ownership would not satisfy the rule. The Central Electricity Regulatory Commission or the state commission can enforce this as the rule is already in effect. No one has publicly announced its application to Google’s project. This fix would ensure that Google’s renewable procurement is backed by actual generation assets, not just contracts.

The grid stability gap. CERC could classify large loads as "High Impact Large Loads" (HILL), a concept proposed in the draft Electricity (Amendment) Bill, 2025. A HILL designation could require Google to provide hourly load forecasts, allow remote disconnect by the grid operator, and publicly disclose draw patterns. This fix would address the absence of grid integration requirements in the current licence and protect the grid during peak stress.

The technical standards gap. Power Grid Corporation of India is already pursuing its own data centre pilot. In September 2025, its subsidiary PowerTel awarded a contract valued at approximately ₹3.60 billion to Invenia-STL Networks for a greenfield Tier III data centre at Manesar. The project includes design, construction, commissioning, and maintenance of IT infrastructure, plus a disaster recovery facility. Once operational, this pilot is expected to generate technical standards for grid-connected data centres: cooling efficiency, storage co-location, and grid interaction protocols. The Ministry of Power could then mandate these standards for any data centre drawing from the interstate grid. State policies cannot override central technical standards for interstate transmission. This fix is the most likely path to systematic regulation, as it applies uniformly and does not single out any state or company.

The water gap. The Andhra Pradesh Water Resources Department has statutory authority under the Andhra Pradesh Water, Land and Trees Act, 2002, to regulate groundwater extraction. It could cap Google’s water extraction, require real-time monitoring, impose penalties for over-extraction, or mandate alternative cooling technologies such as air-cooled or liquid immersion systems. No such order has been publicly issued. This fix would bring the project’s water use under the same regulatory framework that applies to other large industrial consumers.

The grid-sharing gap. The Andhra Pradesh Electricity Regulatory Commission has the authority to amend licence conditions. It could require that any storage capacity Google installs be made available for peak-hour grid support (typically 6-9 PM), with compensation at prevailing tariffs. This fix would turn the data centre from a pure load into a distributed resource that supports, rather than just strains, the grid.

The Way Forward?

A scenario based on how similar regulatory gaps have been addressed in other sectors and jurisdictions could imply the following:

Immediate term. The groundbreaking proceeds. No fixes are announced. The service fee and water extraction arrangements remain undisclosed. The state celebrates the FDI.

Next three to six months. CERC begins informal inquiries into Captive Power Rules compliance, not as a public enforcement action but as standard regulatory due diligence. PGCIL publishes initial technical standards from the Manesar pilot. APERC issues a consultation paper on grid-sharing for large loads, inviting public comment.

Six to twelve months. The Ministry of Power issues a clarification that HILL classification applies to data centres above a specified threshold. This is framed as implementing existing law, not targeting any company. Andhra announces a water monitoring framework. Google announces voluntary battery storage installation, without disclosing whether the decision was influenced by regulatory inquiries.

Two to three years. A stress event occurs: a summer heatwave strains the grid; a dry year drops the water table further; a public interest litigation is filed. Courts impose interim restrictions. The fixes listed above are implemented under pressure, not by design.

The most likely path to systematic regulation runs through PGCIL. As the central transmission utility, PGCIL can establish technical standards that apply across all states. This approach would also be politically sustainable because it would not single out Andhra or Google but would apply uniformly across the board.

What Remains Unanswered

Some questions have no clear answers.

First, what is the service fee? Google will use APTRANSCO’s transmission network, built and maintained with public investment. The fee has not been disclosed. Transparency on this point would clarify whether taxpayers are subsidising the project’s grid access.

Second, what is the water baseline? Without a permit or cap, it is unclear whether Google’s extraction will be monitored at all. The absence of regulation could mean extraction is implicitly permitted, or it could mean existing law is simply not being enforced. Either interpretation has implications for a water-stressed region.

Third, what happens when renewable output falls? The 51% mandate does not require storage. Without disclosure requirements, there is no public way to know what share of Google’s power is backed by fossil generation during low-renewable periods.

China’s approach required a decade of infrastructure investment that was not originally intended for data centres but now serves them well. America’s experience shows that grid constraints can stall or kill projects ($98 billion worth in a single quarter!).

Andhra’s approach took weeks and required no state funding or upfront support. But while speed has tremendous value, speed without guardrails is not reform.

The Takeaway

When Google breaks ground in Visakhapatnam at month end, the photographs will show officials and hard hats; the press release will call it a milestone. What the photographs will not show is the groundwater extraction rates, hourly grid load data, the service fee amount, or the employment numbers five years from now.

Andhra, by giving Google a power licence, has moved fast and broken a monopoly. It attached a renewables mandate and included safeguards for existing discoms, suggesting that policymakers were aware of at least some of the risks.

But the licence has arrived without public enforcement of the Captive Power Rules that already exist. These rules, amended in March 2026, just weeks before Google’s licence was granted, require that any captive generating plant meet two conditions: captive users must hold at least 26% ownership in the plant and consume at least 51% of its generation. The rules apply if Google co-owns any generation capacity, as it has indicated it may. But no public verification has been announced. Pending verification, surcharges are not levied, but if compliance is later found lacking, they become payable with interest.

The absence of enforcement leaves unclear whether Google’s arrangement is genuinely compliant, deliberately structured to avoid the rules, or simply awaiting verification. Either way, the lack of transparency benefits the company while other consumers may have to bear the cross-subsidies that captive status is designed to protect.

However, the fixes suggested above are already sitting in CERC’s rulebook, in PGCIL’s pilot project, in APERC’s regulatory authority, and in the state water department’s statutory powers. Whether they remain there, or are brought to bear, will determine whether this bold policy move is remembered as a reform or as an expensive photograph.