The New Utility

Google closed a $4.75 billion acquisition of Intersect Power in March 2026. In doing so, it became the first hyperscaler to cross from offtake into ownership of the developer platform and its generation pipeline. The result: a hyperscaler owning the development platform that originates generation, not just contracting for the electrons it produces, and thus, blurring the line between the hyperscaler and utility business model.

It’s another chapter in the grid defection and grid bifurcation (see Tale of Two Grids) story. This new model is not quite the Rockefeller industrial vertical monopoly applied to AI in which AI industrials consolidate the full physical and digital stack. It isn't the Insull economies-of-scale (natural monopoly) utility in which hyperscalers morph into vertically integrated IOUs either. It's a third corporate form — a new utility model that owns generation and delegates operations. It involves the assumption of ownership in developers while delegating the operating risk and leaving the regulatory interface for somebody else.

The emerging model is novel enough that the regulatory framework that would govern it does not yet exist. This gap has critical implications for the AI infrastructure buildout for the next 24-36 months. It's yet another data point signaling just how scarce power supply is in the AI infrastructure boom today and how the compute arms race is fueling an increasingly aggressive push to activate all available electrons.

Hyperscalers "buying generation" generally refers to seven distinct legal and financial instruments:

You can plot every verified hyperscaler power transaction on a "degree of control" spectrum running from PPA to capacity tolling to offtake-with-equity to minority equity to acquiring the developer platform to OEM integration. The pattern of the last eighteen months is migration up that spectrum, not occupation of any single tier. In other words, the hyperscalers are getting increasingly creative (and ambitious) in their scramble for electrons, including integrating an increasing share of the physical stack into their corporate structure.

What hyperscalers are conspicuously not buying: operating risk, regulatory interface, local political risk, decommissioning liability. The decommissioning point matters most on the 60-year nuclear tail, which is exactly the layer hyperscalers are happy to leave with the operator. Nor are they likely to acquire OEMs directly.

Microsoft and Amazon hold the procurement tier with Constellation TMI and Talen Susquehanna respectively. Google closed its $4.75 billion acquisition of Intersect Power in March 2026, which is the first hyperscaler transaction to cross from offtake into ownership of the developer platform and its generation pipeline. The cleanest single-sentence description of the stack actually being assembled: long-dated electron rights, equity in the origination platform, queue position at constrained nodes, upstream supply optionality, wrapped in a structure that leaves operating, regulatory, and decommissioning risk with somebody else.

The closest historical analog is the IPP wave in the 1990s. Merchant generators were built against long-term industrial offtakes from a small number of large industrial counterparties, financed against the offtake rather than the regulated rate base. PJM's gas fleet and the Texas merchant build were largely structured this way. The IPP form was novel at the time, the regulatory framework, which remains thin, took most of a decade to catch up mostly in capacity markets, FERC merchant rules, and the long fight over Minimum Offer Price Rule (MOPR).

What's different this time around is the offtake counterparty. The 1990s IPP build had a fragmented industrial buyer base; the New Utility counterparty set is six balance sheets — Microsoft, Google, Amazon, Meta, Oracle, SpaceX — the only firms with both the capital base and the AI roadmap to underwrite 24/7 multi-decade generation offtake at scale. The equity layer extends into the developer platform rather than the project. The demand profile is uniform 24/7 rather than load-following. This model is indeed closer to a 1990s IPP wholesale model than either Rockefeller or Insull, and the regulatory reception will likely come through a similar set of forums on a substantially shorter clock.

Earlier merchant cycles still operate largely outside a coherent regulatory framework decades after the IPP emerged: the 2014 second-wave gas builds were still finding their financing structures inside organized markets long after deregulation. The New Utility cycle is different. The regulatory response is already taking shape as the cost-assignment debate (aka ‘who foots the bill for the corresponding grid upgrades’) heats up.

NERC has issued an alert on data center load reliability with a mandatory response deadline this summer. FERC's advance notice of proposed rulemaking on large-load interconnection (RM26-4) is on track to advance to a notice of proposed rulemaking (NOPR) by mid-2026. The White House Ratepayer Protection Pledge is voluntary, but its publication in the Federal Register gives state PUCs a federal reference document, and that document landed in March. Colorado and Pennsylvania PUCs have adopted model tariffs that assign new infrastructure cost to the requesting load. Over 300 data center bills have been introduced across 27 state legislatures in the current session; most will die in committee, but the directional pressure is unambiguous and the next twenty-four months of rulemaking will determine the shape of mandatory cost-assignment in the most affected jurisdictions.

The voluntary to mandatory transition is real. The timeline is patchwork. State PUC proceedings move on cycles of twelve to twenty-four months unevenly, and the actual landscape will remain uneven through at least 2028 (and likely beyond). But the strategic implication is already clear: a hyperscaler signing a PPA today inherits cost-assignment exposure in the jurisdiction where the new generation gets built.

Owning the developer changes the calculus. You are paying for what you would have been assigned anyway, and you are capturing the upside on capacity that will be sold into a capacity-constrained grid for two decades.

The Subnational Sovereigns mapped the bifurcation of US state regimes around AI infrastructure — a small set of jurisdictions (Virginia, Texas, Ohio, Georgia among them) have become the de facto policy authors while Washington remains gridlocked. The cost-assignment mechanism described here is what enforces that fracture in dollar terms: same political geography, now expressed as who pays for the wires.

The New Utility model will expand materially over the next five years. Two falsifiable predictions, both observable by the end of 2027:

If all the announced material hyperscaler power deals through the end of 2026 are pure offtake or PPA without direct investment in the developer, then the equity-coupled offtake form is unlikely to expand further in the near-term and the forecast is invalidated. If any hyperscaler takes majority equity in a top-ten equipment OEM, then the dominant model becomes Rockefeller-style industrial conglomerate after all, and again, the forecast is invalidated.

The industrial monopoly model remains unlikely. Transformer and switchgear lead times are real and delaying several in-flight hyperscaler builds. The following realities are more than a footnote: Prolec GE backlog at $5 billion with factory booking out three to four years; Eaton order book at $14.5 billion, up 44 percent year-over-year as of the most recent quarterly report; Siemens Energy order book past €146 billion; GE Vernova orders up 71 percent above prior-year levels in Q1. Equipment supply chains have become one of the genuine risks to the AI infrastructure buildout over the next three to five years, which explains an increasingly popular trade-press hypothesis that hyperscalers will invest in the OEMs themselves.

The bottom line: the factory is a constraint, but not the strategically decisive one. Supply follows demand and OEMs will ramp their supply chains in time. Meanwhile, the queue position remains the most critical bottleneck. Buying the OEM does not move the queue. Booking three years of factory output does, which is why hyperscalers will not likely buy OEMs. The rational response is the purchase contract, not the equity.

The buy case for hyperscaler power buildouts assumes the regulatory framework either remains the current status quo (built for utilities serving customers) or evolves toward an industrial-host framework (built for IPPs serving offtakers). Neither fits.

The New Utility model has no historical playbook because the regulators who would write it are still grappling with how to characterize it. Most likely is a regime that aligns with state PUCs wrestling with cost-assignment, FERC reasserting jurisdiction, and an antitrust bench newly interested in the cloud–power adjacency.

The lag between a new corporate form taking shape and a framework arriving to govern it is measured in years, not decades. The scale of the regulatory response will reflect the scale of the New Utility model's success. The faster the New Utility model takes hold, the faster the regulatory reaction coalesces. In other words, the regulatory response to the New Utility is very likely coming soon and probably this decade, not next. When it arrives, it won't look like the current utility framework or the IPP framework and will very likely shift the investment thesis.

What does the New Utility tell us about where we are in the Energy Transition x AI Infrastructure buildout? The New Utility is the political-commercial response that enables hyperscalers to convert strong balance sheets into power to sustain compute growth while regulators ring-fence the externalities through cost-allocation rules. But how durable is this current state? Not very, for several reasons outlined below.

Aroko’s Grid-Silicon Order framework defines four possible geopolitical futures in the AI infrastructure x Energy Transition buildout defined by the the degree to which: a) geopolitical architecture is Integrated versus Fragmented (does the world coordinate on standards, capital flows, and technology transfer — or harden into competing blocs with parallel supply chains?); and b) velocity buildout is Accelerated versus Constrained (does generation buildout and compute capacity keep pace with AI demand — or do permitting, materials, and power become binding constraints?).

Possible futures include:

1) Open Abundance (Integrated + Abundant) — Abundant power delivered by a portfolio (gas as bridge, renewables at scale, nuclear restart and SMRs coming online, transmission queues unblocked) plus cross-border AI standards. Compute trades like a commodity; climate goals tracked, but on a longer glide path than the headline 2030 plans suggest, and methane regulation matters more than CO₂ in this decade. Winners: integrated grid operators, gas-turbine OEMs with credible low-carbon retrofit roadmaps, scaled cloud, multilateral standards bodies.

2) Sovereign Stacks (Fragmented + Abundant) — Each bloc builds its own energy-compute stack including the fuel layer. US AI runs on Permian and Marcellus gas plus a nuclear restart; the Gulf runs on its own gas plus solar; China runs on coal plus nuclear. Gas turbine OEMs (GE Vernova, Siemens Energy, Mitsubishi) become strategic national assets and their multi-year backlogs become geopolitical bottlenecks. LNG export policy is industrial policy. Winners: national champions, friend-shored fabs, integrated gas-to-power-to-compute corridors (Permian–Texas, Qatar–Saudi, Marcellus–Virginia).

3) Coordinated Scarcity (Integrated + Constrained) — Power and silicon bottlenecks force allocation, with gas as the swing fuel everyone needs but no one is comfortable building. Methane regulation, stranded-asset risk, and ratepayer fights over who pays for new CCGTs become the political flashpoints — and these are the cost-assignment battles the New Utility piece is actually describing. A compute climate club may emerge but cohabits messily with gas-heavy build. Winners: efficiency-frontier labs, edge AI, demand response, and firms that can credibly run gas under tightening methane rules or pair it with capture.

4) Fortress Era (Fragmented + Constrained) — Scarce resources and closed borders. Natural gas is the primary energy weapon — LNG flow control, pipeline politics, gas turbine and compressor station export controls (precedent already exists from the Siemens Crimea episode). Critical minerals become secondary leverage. AI capability concentrates in two or three powers; the ability to fuel the compute matters as much as the ability to fabricate it. Winners: defense primes, sovereign wealth gas holdings, LNG terminal operators, strategic stockpilers of turbine components.

The New Utility is the corporate form emerging to address constraints and is a response best aligned with a Coordinated Scarcity world. But this is likely more fleeting than durable. Capital displacing utility rate base, the New Utility playbook outrunning regulators, and interconnection queues hardening into a self-reinforcing concentration moat are inherently destabilizing to the Coordinated Scarcity frame. The most probable 3-5 year outlook, especially with gas playing an increasingly central role, is a drift toward Sovereign Stacks. The same political pressure that's currently routed through PUC and FERC dockets eventually reframes hyperscalers as strategic national infrastructure rather than as utilities to be ring-fenced. In this scenario, the corporate form survives but the regulatory regime around it changes character entirely.

Three implications follow that shape the investment calculus.

First, the dependency arrow flips. The prevailing thesis ran one direction: silicon needs grid. The New Utility acknowledges the inverse: the grid buildout (specifically new generation and the transmission infrastructure that integrates supply) now needs hyperscaler balance sheets to get financed. Each side is a hostage of the other's roadmap, and that bidirectional dependency is a structural shift, not a phase of demand growth.

Second, hyperscaler generation capex is not net-new money flowing into the energy transition. It is capital redirected — from utility ratebase (regulated and slow) to hyperscaler corporate balance sheet (fast and concentrated to data center load). This is the grid defection narrative. For utilities this means losing growth runway exactly when they're ramping investment into grid modernization initiatives. The same dollars that accelerate AI's power supply decelerate the rest of the energy transition. That tradeoff is invisible in the gigawatt-announcement coverage and critical to understanding the macro picture.

Third, AI hyperscaler competitiveness moves beyond frontier model capabilities to interconnection queues. The seventh-place AI lab cannot compete regardless of model quality, because it cannot access power. AI industry concentration becomes self-reinforcing on infrastructure grounds, not just data and compute. Antitrust analysis has to extend into power markets, a domain DOJ and FTC do not currently staff.

Ultimately, hyperscalers are not building a traditional utility, and they are not building Standard Oil. They are building the New Utility out of necessity — and the regulatory framework that will govern it is being drafted right now, in state PUCs, in FERC dockets, and in the cost-assignment battles of dozens of state legislatures, by people who do not yet have a name for the form they are regulating. This form is born out of power scarcity and an increasingly opportunistic playbook for acquiring the necessary generation to sustain ambitious hyperscaler growth trajectories.

Two 2026 events will shape the New Utility future. FERC's RM26-4 trajectory toward a notice of proposed rulemaking by mid-year determines whether large-load cost-assignment becomes a federal floor or remains a state-by-state patchwork. The second material hyperscaler developer-platform equity transaction — the one that confirms Google–Intersect was a template rather than an outlier — locks the equity-coupled offtake form in as the dominant strategy for the rest of the decade.

Underneath, the structural force is older than AI. The U.S. electricity system was built to socialize the cost of long-lived generation across a broad ratebase on a regulated cadence. Hyperscaler demand is concentrated, 24/7, and on a timeline ratebase finance cannot meet. The same physics that produced the 1990s IPP wave — offtake-financed merchant generation underwritten by a small set of industrial counterparties — is producing the New Utility now, with the equity layer extended into the developer platform and the counterparty set narrowed to five balance sheets.

The form scales faster than the regulator. Cost-assignment fights in Colorado, Pennsylvania, Virginia, and Texas are skirmishes over who pays for the wires. The larger contest is over which institution regulates a counterparty that owns generation, controls the queue, takes equity in development platforms, and refuses operating and decommissioning risk. State PUCs have the police-power tools but not the cross-border jurisdiction. FERC has the jurisdiction but not the antitrust mandate. DOJ has the antitrust mandate but no staff in power markets.

The New Utility is not a deviation from utility form. It is the corporate response to a constraint that the traditional utility model cannot solve at the speed AI demand requires. The regulators are likely to catch up, but the response may very well be superseded by the Sovereign Stack future. The methods in which regulators respond — federal preemption, antitrust extension, state-by-state cost-assignment, or some hybrid — is what the next twenty-four months decide.

About Aroko: Aroko provides strategic advisory and macro trend intelligence at the intersection of energy transition, technology infrastructure, and geopolitical risk. Our analytical process combines proprietary evidence infrastructure with human-directed thesis formation. Every keystone claim is verified against primary sources, and all editorial judgment and capital allocation framing is conducted by Aroko’s team. The Letter is published biweekly.

Methodology: AI-assisted evidence infrastructure · Human-directed thesis · Primary-source verified

Disclaimer: This article is for informational purposes only and does not constitute financial, investment, legal, or tax advice. The opinions expressed regarding macro trends and infrastructure investments are solely those of the authors. Past performance does not guarantee future results. Readers should consult with a qualified financial professional before making any investment decisions.