For two years the story of artificial intelligence has been a story of abundance — more chips, more capital, more capacity announced in ever-larger round numbers. But the physical world is starting to push back. The constraint on AI is quietly shifting away from semiconductors and funding toward something harder to manufacture: electricity, water and the consent of the people who live next to a humming, fenced-off box the size of several football fields.
The emerging picture across the United States is one of plans colliding with grids, supply chains and local politics. That collision is instructive for India, which is racing to build out its own data-centre and AI-compute footprint and will face the same binding constraints — sooner than many in the industry expect.
The build-out meets reality
The headline numbers around AI infrastructure have always assumed a frictionless world: announce a campus, secure the chips, pour the concrete, plug in. Reality is proving messier. According to analysis from TechStartups citing Sightline Climate, a substantial share of planned US data centers — on the order of 30 to 50 percent of roughly 140 sites targeting around 16 gigawatts of capacity — may miss their 2026 timelines or be cancelled outright. (Zoho Social has not independently verified these figures; we flag them as the source’s estimate.)
The reasons are stubbornly physical. Large-scale data centers need high-voltage transformers, grid interconnections and increasingly on-site battery storage — and all three are in short supply. Lead times for big power transformers now stretch into multiple years, a global bottleneck driven by surging demand and limited manufacturing capacity. Grid-connection queues are similarly clogged; a developer can sign for land and silicon and still wait years for a utility to physically deliver the megawatts a hyperscale facility demands.
Construction itself is no longer the gating factor. A data hall can be built faster than the electricity to run it can be procured. That inversion — where the building is ready but the power is not — is the defining tension of this cycle. Capital is plentiful; transformers, interconnects and trained linemen are not. When supply-chain delay meets local resistance, projects don’t just slow. Some quietly disappear from the pipeline.
The political backlash
The second constraint is political, and it is arriving faster than the industry anticipated. Communities that once competed for data-centre investment with generous incentives are reconsidering the bargain. Per TechStartups’ reporting, some jurisdictions — Ohio among them — have moved to suspend data-centre tax incentives and have floated outright bans on new hyperscale facilities. Towns are putting these decisions to votes. The mood has shifted from courtship to scrutiny.
The core dispute is about who pays. Data centers consume enormous amounts of power, and the grid upgrades needed to serve them — new substations, transmission lines, reinforced distribution — are expensive. When those costs land on the ratepayer base, ordinary households can end up subsidising the energy bills of trillion-dollar technology firms through higher utility rates. That is a politically toxic outcome, and residents are increasingly aware of it.
The traditional defence — jobs — is wearing thin. A hyperscale facility employs a large construction crew for a year or two, then settles into a steady-state operation that may require only a few dozen permanent staff. Set against multi-decade tax abatements, strained local water supplies and higher power bills, the jobs-per-megawatt math no longer reassures town councils the way it once did. The result is a wave of incentive rollbacks, moratoriums and ballot-box rejections that the sector’s growth forecasts largely failed to price in.
Water becomes the flashpoint
If power is the slow-burning constraint, water is the one that ignites public anger fastest. Many large data centers use water-based cooling, and a single campus can consume volumes comparable to a small town. In regions already facing drought and utility stress, the optics of an AI facility drawing millions of litres while residents face restrictions are politically explosive.
The industry has noticed. According to TechStartups, Google has pledged to replenish more water than it consumes by 2030, an explicit response to mounting community pushback over data-centre water use. Such “water-positive” commitments are becoming table stakes for operators seeking social licence. The credibility of these pledges — whether replenishment happens in the same watershed that bears the consumption, and whether it is verifiable — will increasingly determine whether a project gets built at all.
The clearest signal of how seriously the sector now takes this came, per the same reporting, from SpaceX, which flagged water access as a risk factor in its IPO filing. When a frontier technology company names water alongside the usual capital and competitive risks, the message is unambiguous: hydrological reality has become a board-level concern. Water is no longer an operational footnote. It is a gating input, a regulatory pressure point and a community flashpoint rolled into one.
The India read
India is building data-centre capacity at pace, propelled by data-localisation rules, a swelling digital economy and a national ambition to host sovereign AI compute. Mumbai, Chennai, Hyderabad and emerging hubs around the National Capital Region are drawing billions in committed investment. The opportunity is real. So are the constraints — and they map almost exactly onto the American experience, with local intensifiers.
Power is the first. India’s grid is improving but remains uneven, and large parts of generation still lean on coal. A data-centre boom that simply bolts onto fossil-heavy power undercuts the clean-AI narrative buyers increasingly demand and exposes operators to future carbon costs. Reliable, round-the-clock power at hyperscale volumes is not a solved problem in every Indian metro, and grid-connection timelines here carry their own delays.
Water is the second, and arguably the sharper one. Many of India’s data-centre clusters sit in water-stressed regions; Chennai’s periodic water crises are a standing reminder of how fast supply can tighten. The same cooling-water demands that have inflamed American towns will surface in Indian municipalities — and India’s water governance is more fragmented and more politically charged. Operators that arrive with credible, watershed-level replenishment plans and air- or recycled-water cooling will face far less friction than those that treat water as an afterthought.
There is, however, a genuine advantage hiding in these constraints. India has abundant solar and growing wind capacity, and falling renewable costs make clean-power siting a competitive lever rather than a compliance burden. Co-locating data centers with renewable generation — in Rajasthan, Gujarat, the southern wind corridors — can address the power constraint, ease the carbon question and reduce reliance on water-hungry thermal cooling and generation all at once. Operators that lead with clean-power siting, transparent water accounting and early community engagement will build faster and face fewer reversals than those chasing the cheapest land and the largest subsidy.
The lesson from the US is not that the AI build-out will stall everywhere. It is that the easy phase is over. The winners in the next cycle will not be the firms with the most chips or the deepest balance sheets. They will be the ones that solved for transformers, water tables and town councils before the first slab was poured. For India, that is both a warning and an opening — the chance to build a data-centre industry that is constrained by design rather than caught out by it.
