Infrastructure Intelligence No. 04: Flexibility Is Becoming Infrastructure

Energy. Power. Development. Capital. | September 22, 2026 | Issue No. 04

THE WEEK IN INFRASTRUCTURE

Over the past three weeks, the large-load market has moved from power availability, to project readiness, to the electrical behavior of the load itself. This week, another concept moved to the forefront: flexibility is becoming a power asset.

On September 16, Google, NVIDIA and Emerald AI launched the AI Energy Management Alliance, or AEMA, around a simple premise: large AI facilities should be capable of dynamically changing how they use electricity in response to grid conditions. The alliance is advocating measurable ride-through, curtailment and contingency-response capabilities, along with faster, risk-adjusted interconnection pathways for projects that can make credible flexibility commitments. AEMA is an industry advocacy group; its principles are not binding on any utility or grid operator.

One day later, the concept became remarkably tangible. PJM activated demand response across much of its footprint during unusually warm September weather, with approximately 36,000 MW of generation outages still in place for maintenance season. DOE also granted emergency authority allowing PJM, if necessary, to direct backup generation resources to operate as a last resort; PJM specifically described the authority as potentially allowing data centers and other large loads to move onto backup generation.

The implications extend well beyond one operating event. The large load is beginning to look less like a fixed demand on the grid and more like a controllable part of the power system. For developers, that could materially change how sites are designed, studied and ultimately valued.

01 | THE LEAD

The Data Center Is Becoming a Grid Resource

Historically, a 300 MW data center requesting utility service was fundamentally treated as a 300 MW load. The emerging model is more nuanced.

A facility may be capable of shifting noncritical compute workloads, dispatching battery storage, operating paired generation, reducing grid imports during constrained periods or remaining connected through disturbances rather than abruptly dropping load.

AEMA's stated goal is to turn those capabilities into measurable commitments that utilities and grid operators can actually rely upon. Its principles call for ride-through, curtailment and emergency-response obligations to be established before interconnection, standardized performance metrics and operational data sharing, and potentially faster interconnection pathways for customers that can demonstrate credible and verifiable flexibility.

The timing is significant. On September 17, PJM forecast an approximately 132,000 MW peak while roughly 36,000 MW of generation remained out of service during the beginning of fall maintenance season. PJM called both pre-emergency and emergency demand response in portions of its system and obtained temporary federal emergency authority.

PJM did not report that it ultimately had to order data centers onto backup generation. But the fact that federal emergency authority explicitly contemplated backup generation from large-load customers demonstrates how quickly the operating relationship between data centers and the grid is evolving.

Developer Takeaway

Flexibility may become interconnection currency.

Two 300 MW campuses may look identical on a utility load letter but very different to a system operator if one can verifiably reduce 50 MW within minutes, sustain the reduction for several hours, ride through defined disturbances and coordinate batteries or onsite generation.

That difference could eventually influence study assumptions, upgrade requirements, service structures and energization timelines. Developers should increasingly think of flexible IT load, BESS, onsite generation, controls and telemetry as one integrated power architecture, rather than separate systems designed at different stages of the project.

02 | POLICY & GRID SECURITY

California Makes Data Center Infrastructure Responsibility State Law

California moved decisively on September 21, when Governor Gavin Newsom signed a package of seven data-center bills addressing electricity, water, land use, reporting and rate structures.

The state says the new framework is intended to require data centers to bear appropriate grid-upgrade costs, prevent those costs from being shifted to other ratepayers, meet state energy-procurement requirements and provide local governments and water suppliers with additional information about water consumption, efficiency and drought planning. The package also changes how data centers qualify for certain environmental-review streamlining.

The legislation includes AB 1577, AB 2383, AB 2469, AB 2619, SB 886, SB 887 and SB 1168.

California's action follows a similar policy movement in Washington. On September 16, the U.S. House passed the Ratepayer Protection Act, H.R. 9340, by 417-3. The bill would require state utility commissions to consider standards for data centers above 100 MW intended to recover the full incremental cost of generation, transmission and distribution infrastructure required to serve them. The measure has not passed the Senate and is not yet law.

The following day, competing Senate proposals exposed disagreement over how prescriptive the federal government should be. Senator Jon Husted sought unanimous consent for the House approach; Senator Martin Heinrich objected and offered a separate proposal that would make large-load customers directly responsible for facilities needed to connect them. Neither proposal advanced by unanimous consent.

What is notable is not simply the legislative disagreement. It is the increasingly broad agreement around the underlying principle: large loads should bear more of the infrastructure costs created by their development.

Developer Takeaway

Cost allocation is becoming a core component of site economics. Developers should increasingly expect combinations of infrastructure reimbursement, minimum bills, long-term service commitments, collateral, deposits, generation requirements and stranded-cost protections.

The question is moving away from whether large loads will be expected to bear more infrastructure cost. It is becoming how those obligations will be structured, financed and allocated among the developer, tenant, utility and capital provider.

03 | POWER & GRID

NERC Moves Computational Loads Toward Formal Reliability Standards

The reliability framework around large computational loads is also advancing quickly. Issue No. 03 covered the two Standard Authorization Requests that opened this process; the first of the resulting standards has now cleared an initial ballot.

On September 19, NERC announced preliminary results indicating that its foundational Computational Loads Reliability Standards passed their initial ballot.

NERC said the proposed standards draw from recent incident reviews, its Level 3 alert and work by the Large Loads Working Group, and are intended to establish baseline reliability requirements for the distinctive behavior of large computational facilities. NERC is still validating the ballot results and has not yet completed the standards process. These are not yet enforceable standards.

The direction, however, is clear. Data centers are moving from being treated primarily as utility customers toward becoming facilities whose modeling, protection systems and response to grid disturbances may be governed by formal reliability requirements.

Developer Takeaway

Electrical performance requirements need to migrate much earlier into design. Ride-through capability, protection coordination, telemetry, load-shedding logic, UPS behavior, backup-generation transitions and model quality are increasingly likely to affect interconnection.

For developers, these are no longer merely commissioning or operating issues. They are becoming development inputs.

04 | DATA CENTER / MISSION CRITICAL

207 MW of Contracted Power Comes With Nearly $200 Million of Credit Support

A new Texas transaction provides a useful look at what "secured power" can actually require.

On September 18, a subsidiary of New Era Energy & Digital entered into a 20-year power purchase agreement with Luminant, a Vistra affiliate, for at least 200 MW and up to 207 MW for the first phase of the Texas Critical Data Center near Odessa. The power can come from Vistra's adjacent 1,180 MW natural-gas facility or other available sources, including the ERCOT grid.

But the commercial obligations behind those megawatts are equally noteworthy. New Era must post a $116 million letter of credit within 15 business days of the PPA and provide as much as another $82.8 million in security by the delivery date, potentially approaching $199 million of total credit support.

The related development agreement also provides Vistra rights around future onsite generation and certain power and storage projects, requires reimbursement of specified substation and transmission construction costs, and calls for Vistra to receive a 5% non-voting equity interest in the relevant project company after power delivery begins.

That is a much more sophisticated economic relationship than a simple utility reservation.

Developer Takeaway

Power certainty is being financed like infrastructure. A signed power agreement can consume substantial balance-sheet capacity, require nine-figure collateral, create infrastructure-reimbursement obligations and affect future development rights and project ownership.

For developers and investors, "contracted power" should never be treated as a single diligence checkbox. The underlying security package, conditions precedent, transmission obligations, generation rights and capital requirements may materially affect the economics of the entire development.

05 | ENERGY INFRASTRUCTURE

Behind-the-Meter Power Is Becoming a Deployable Service

The behind-the-meter market continues to mature from project concept into contracted infrastructure.

On September 21, Kodiak Gas Services announced a six-year agreement to provide 76 MW of baseload behind-the-meter power to a West Texas data center using approximately 40 natural-gas reciprocating-generation units. Deployment is expected to begin in the fourth quarter of 2026 and scale into the first quarter of 2027. Kodiak says the data-center operator is contracted with an investment-grade hyperscaler and that a GPU designer guarantees the data-center lease.

At a much larger scale, INNIO announced September 17 that a U.S. energy company had placed a firm 450 MW order for containerized Jenbacher engines intended for large-scale North American data-center projects. Those units are expected to be delivered by 2028.

The two announcements illustrate both the opportunity and the constraint. Existing distributed-power fleets can provide relatively rapid bridge or primary power in certain locations. But larger deployments still depend on factory capacity and long-lead equipment.

Developer Takeaway

Behind-the-meter generation can reduce dependence on utility energization schedules. It does not eliminate development scheduling.

Fuel supply, air permitting, emissions controls, generation equipment, switchgear, transformers, interconnection configuration and construction all become part of the critical path.

The emerging lesson is straightforward: fast-to-power increasingly begins with fast-to-procurement.

06 | EQUIPMENT & SUPPLY CHAIN

800 VDC Moves From Roadmap to Product Portfolio

The electrical architecture inside AI facilities is also changing. On September 21, ABB launched Infinitus, which it describes as the industry's first integrated source-to-rack direct-current portfolio for AI data centers.

The platform is built around medium-voltage-to-800 VDC conversion using solid-state transformer technology, together with DC power quality, distribution, protection and cooling optimization.

ABB says its architecture can eliminate conversion stages and reduce heat loss and electrical footprint. The company estimates that a DC architecture can provide more than 5% end-to-end efficiency improvement, though that figure is an ABB estimate rather than an independently established industry benchmark.

Why does a few percentage points matter? At hyperscale, efficiency is no longer merely an operating-cost issue. It determines how much of a constrained electrical service can actually reach revenue-producing compute.

Developer Takeaway

Facility electrical architecture is becoming a site-development variable. At 500 MW, even modest improvements in conversion efficiency can represent meaningful additional megawatts available for IT load without increasing the site's gross power requirement by the same amount.

As rack densities rise, developers will increasingly need electrical engineers, utility teams and data-center designers working from a common power budget much earlier in development. The boundary between building electrical design and site power strategy is disappearing.

07 | CAPITAL & DEAL FLOW

Nscale Takes the Power-First Model to the Public Markets

Perhaps the clearest capital-market expression of this week's theme came from Nscale. On September 18, the AI infrastructure company filed a Form S-1 for a proposed U.S. initial public offering and applied to list on the New York Stock Exchange under the symbol NSCL. The number of shares and price range have not yet been established, and the registration statement is not yet effective.

More interesting than the IPO itself is the development thesis Nscale is presenting to investors. Its SEC filing describes reliable, contiguous power as the primary gating factor for AI infrastructure deployment and says the company has built a portfolio exceeding 10 GW of owned and controlled power. Nscale has also created an internal Energy & Power division to originate and develop generation and describes its strategy as power-first.

Its approximately 2,250-acre Monarch Compute Campus in West Virginia is planned around as much as 8 GW of gross behind-the-meter power, with potential capacity exceeding 6.5 GW of IT load. These are the company's own planning figures, disclosed in a registration statement, not built or contracted capacity.

Nscale's prospectus is therefore not presenting AI infrastructure merely as data centers plus GPUs. It is presenting land, power generation, data-center development, compute and software as a vertically integrated platform.

Developer Takeaway

Capital markets are beginning to evaluate power-development capability as part of the digital-infrastructure platform itself.

That matters for developers because powered land, generation strategy, interconnection capability and infrastructure execution may increasingly influence enterprise value rather than simply support it. The distinction between a data-center company, an energy developer and an infrastructure platform continues to narrow.

08 | THE INTERCONNECTION DESK

ERCOT

Texas is moving beyond interconnection engineering into broader project diligence. The State and Community Impact RFI, introduced in Issue No. 03, applies to data-center projects of 25 MW or greater pursuing interconnection and asks developers for information including grid dependency, onsite generation, water sources and consumption, cooling technology, public incentives, ownership and local impacts. Responses are due October 12.

On September 21, ERCOT modified its RFI portal to further restrict visibility of submitted information after receiving participant feedback, while continuing its Batch Zero verification process.

This is another indication that large-load diligence is expanding from whether the grid can serve a project toward what exactly is being developed, by whom, and with what broader infrastructure impact.

MISO

MISO is moving toward more detailed real-time visibility of large computational loads. The Large Load Working Group has proposed telemetry and phasor-measurement-unit requirements, with stakeholder feedback due September 23.

For developers, the significance is practical: high-resolution monitoring and operational visibility may ultimately become part of the facility design and interconnection package rather than something added after energization.

CAISO

California ISO is approaching a major milestone in its Large Loads Initiative. CAISO plans to publish its draft final proposal and draft tariff language on September 24, followed by an October 1 stakeholder meeting and an October 15 comment deadline before eventual Board and FERC consideration. Nothing has been filed with FERC.

The California process is particularly important following this week's state legislation because tariff design, cost responsibility, flexible service and state policy are increasingly converging around the same large-load developments.

NERC

NERC's preliminary ballot result on the first foundational Computational Loads Reliability Standards adds another layer above the regional processes.

The direction across ERCOT, MISO, CAISO, PJM and NERC is becoming increasingly consistent: large-load interconnection is becoming more technical, more transparent, more performance-based and less tolerant of speculative development.

09 | ON OUR RADAR

CAISO, September 24

The draft final Large Loads Initiative proposal, flagged in Issue No. 03, could provide one of the clearest indications yet of how a major ISO intends to reconcile speed-to-power, flexible service, cost allocation and system reliability.

ERCOT's October 12 Data Center Deadline

Texas developers of qualifying data-center projects will have to provide information extending well beyond electrical demand, including water, cooling, ownership and community impacts. That could become a model for how other states evaluate large-load development.

EO 14421 Implementation

The DOE implementation process for the bulk-power security executive order remains open for written responses through October 9. Equipment provenance, foreign ownership, software, firmware and supply-chain security remain issues developers should be incorporating into procurement diligence now, rather than waiting for the final implementing rules.

Flexible Load Becomes Contractual

AEMA's significance will ultimately depend on whether utilities and ISOs convert the concept into enforceable tariffs and interconnection agreements. If they do, metrics such as response time, duration, available megawatts and telemetry could become commercially valuable project attributes in the same way that firm generation or transmission access are today.

THE INTERFACE VIEW

Over the first four issues of Infrastructure Intelligence, a progression has emerged.

First: the data center business is becoming the power business. Then: interconnection readiness is becoming development readiness. Then: electrical behavior is becoming part of development responsibility. This week adds another principle: flexibility is becoming infrastructure.

For years, flexibility was largely viewed as an operating feature to be considered after a facility had obtained power. That sequence is changing.

A 300 MW data center capable of reliably shedding 50 MW, shifting compute, dispatching storage, synchronizing onsite generation and riding through a grid disturbance is fundamentally different from 300 MW of static demand.

Grid operators are beginning to recognize that distinction. Policymakers are beginning to assign costs based on actual infrastructure impacts. Capital providers are beginning to value platforms that control both energy and compute. And equipment manufacturers are redesigning the electrical system from the transmission interface all the way to the rack.

The development stack is evolving again:

Land + Power + Interconnection + Generation + Equipment + Capital + Flexibility + Execution

The next competitive advantage may not belong to the developer that requests the most megawatts. It may belong to the developer that can demonstrate that the megawatts it needs are credible, controllable, financeable and executable.

Power availability created the opportunity. Development readiness established credibility. Flexibility may increasingly determine who gets connected, and how quickly.

Interface Holdings and Development Firm

Building the Future of Infrastructure Through Strategy, Capital, and Integration.