For two years the conversation about artificial intelligence has been a conversation about silicon — who gets the GPUs, how many, and how fast. But the more important scarcity may be invisible in the spec sheet. Every training run and every inference query is, at bottom, an electricity bill, and the machines that run them are now landing on power grids that were never planned for them. In June 2026 that tension stopped being a forecast and became a regulatory event. The story is no longer just about chips. It is about megawatts, who pays for them, and which countries can supply them.
The grid pushes back
On 18 June 2026, the US Federal Energy Regulatory Commission (FERC) issued ‘show cause’ orders to all six of the country’s regional grid operators — every one except Texas, which runs its own grid — according to reporting from Build Fast with AI. A ‘show cause’ order is regulatory shorthand for: explain yourself or change your rules. In this case, FERC wanted the operators to defend, or reform, how they connect very large electricity consumers — chief among them AI data centres — to the grid.
The order matters because the interconnection process has quietly become the chokepoint of the AI era. A hyperscaler can sign a lease, order servers, and stand up a campus far faster than a utility can build the transmission lines and substations to feed it. Operators have responded inconsistently: some fast-track large-load requests to win the economic prize of a gigawatt-scale customer; others stall, wary of committing scarce capacity to one buyer. FERC’s intervention is an admission that this ad-hoc approach is no longer tenable when single facilities can demand as much power as a mid-sized city.
The thorniest question buried inside these orders is deceptively simple: who pays for grid expansion? When a data centre triggers the need for new transmission, that cost can either be socialised across all ratepayers or assigned to the customer that caused it. Historically, much grid investment has been spread across everyone. That worked when load grew slowly and predictably. It looks far less defensible when the new demand comes from a handful of trillion-dollar companies building speculative compute.
Costs land on someone
Arizona is where that abstract question turned into a number. Arizona Public Service has proposed an electricity-rate increase of roughly 45% aimed specifically at data centres, per Build Fast with AI’s reporting. The framing is pointed: data centres should ‘pay their fair share’ for the grid capacity they consume, rather than passing the cost to households and ordinary businesses.
It is a politically shrewd move and, arguably, an economically sound one. If a new data centre forces a utility to build generation and transmission, charging that facility for the burden it creates is closer to first principles than smearing the cost across every retail customer in the state. The ‘fair share’ language also pre-empts a backlash that utilities can see coming — voters tend not to forgive rate hikes that subsidise the AI ambitions of distant corporations.
But the ripple effects are real and worth watching. A 45% surcharge changes the maths of where to build. Operators will route capital toward states and countries with cheaper, faster, cleaner power, accelerating a kind of regulatory arbitrage. And the ‘fair share’ principle is harder to apply than to assert: data centres can plausibly argue that the grid upgrades they fund also benefit everyone else, and that singling them out chills investment. The likely outcome is not a clean resolution but a patchwork — some jurisdictions courting data centres with cheap power, others taxing them as cost centres, and consumers caught in the middle of both bets.
China’s parallel strain
None of this is uniquely American. China, the other pole of the AI race, is staring at the same arithmetic from the opposite direction. Chinese data-centre power demand is projected to rise by roughly 300 to 500 billion kilowatt-hours between 2026 and 2030 — about a fifth of the country’s total electricity-demand growth over that period, according to figures reported by The Next Web via TechTimes. That is a staggering share of new load to attribute to a single category of facility.
The strain reveals a mismatch at the heart of every national AI strategy: the buildout of compute is racing ahead of the buildout of power. China has spent years adding renewable and nuclear capacity at world-leading speed, yet even that may not keep pace with both its industrial electrification and its AI ambitions simultaneously. Add domestic chip mandates designed to reduce reliance on imported silicon, and Beijing is effectively trying to scale three constrained inputs — chips, power, and grid — at once.
The takeaway transcends any one country: energy strategy is now AI strategy. The nation that can deliver abundant, cheap, reliable electricity to compute clusters has a structural advantage that no amount of chip-design brilliance can offset. When the binding constraint shifts from semiconductors to electrons, the competitive map redraws itself around power plants and transmission corridors rather than fabs alone.
The India implication
This is where India’s opportunity — and its risk — come into focus. India has been pitching itself as a data-centre destination on the strength of cheap land, a vast talent pool, data-localisation rules that require certain workloads to stay onshore, and an aggressive renewable-energy programme. The clean-power angle is genuinely attractive: hyperscalers under pressure to decarbonise want gigawatts that come with a credible green narrative, and India’s solar buildout offers exactly that positioning.
But the lesson from the US and China is unambiguous: energy is the gating factor for local compute, not ambition or policy intent. India’s grid already contends with peak-demand stress, transmission bottlenecks, and the intermittency that comes with leaning heavily on solar. A data centre needs power that is not just clean but constant — 24×7, year-round, with the kind of reliability that intermittent renewables alone cannot guarantee without massive storage or firm backup. The gap between ‘India has lots of solar capacity’ and ‘India can deliver firm gigawatts to a campus on demand’ is precisely where the buildout will succeed or stall.
For operators, marketers, and founders watching this space, a few things deserve close attention:
- Interconnection timelines, not just tariffs. The FERC orders show that the speed of getting connected can matter more than the headline price of power. Ask how long a grid connection actually takes in a given Indian state before celebrating a cheap-power announcement.
- Who bears expansion costs. Watch whether Indian regulators follow the ‘fair share’ logic and load grid-upgrade costs onto data centres. That single policy choice can swing a project’s economics by double-digit percentages.
- Firm vs. intermittent supply. Clean-power marketing is easy; firm clean power is hard. Storage capacity, nuclear, and round-the-clock renewable contracts are the real differentiators.
- Regional arbitrage. Just as US capital is migrating to favourable jurisdictions, compute will flow to whichever Indian states pair fast interconnection with credible firm power. Location strategy is now an energy decision.
The romance of the AI era has been about intelligence — models that write, reason, and create. The reality underneath it is industrial: turbines, transmission lines, and rate cases argued before regulators. FERC’s show-cause orders, Arizona’s 45% proposal, and China’s looming demand surge are early tremors of the same shift. The countries and companies that treat electricity as a first-class strategic input, rather than a line item, will own the next phase of the buildout. For India, the prize is real — but it will be won at the substation, not the server rack.
