# Modgrid: a business in power-constrained computing

**Research cutoff: November 4, 2026. Prepared for Preetham Kyanam.**

**Recommendation:** Start Modgrid as a power-readiness business for independent data-center developers and the engineering firms serving them. Sell a specific decision: which combination of utility service, flexible demand, storage, and on-site generation can support a proposed load, under what conditions, and with what financial exposure. Deliver this initially as a paid, expert-reviewed service. Build reusable software from the work. Expand into flexibility measurement and contract compliance only after customers and utilities accept the evidence.

This is an entry into supplying energy infrastructure through development and operating partners. It does not require Preetham to finance a power plant, become an electricity retailer, or guarantee a data center's uptime at inception.

The strongest alternative is a storage and demand-response development business for existing commercial customers in constrained regions. Those projects can help absorb data-center growth without selling directly to hyperscalers. Google's recently expanded, **1 GW** agreement with Voltus is concrete evidence of this route, although it is an announced portfolio buildout, not proof that a new developer can obtain the same contract. [Google, October 22, 2026][google-vpp]

## How to read this report

- **Scope:** Primarily the United States, with global demand context. PJM, Ohio, Virginia, and Texas receive extra attention because they expose different commercial and regulatory constraints. State examples are selective, not a nationwide legal survey.
- **Evidence:** Laboratory and grid-operator research, regulatory orders, utility documents, company filings, and first-party announcements. Company announcements verify what a company announced. They do not independently verify performance or guarantee completion.
- **Status matters:** Operating capacity, contracted capacity, financing frameworks, reserved factory slots, development pipelines, and forecasts are different things.
- **Estimates:** Business budgets, prices, scores, customer targets, and schedules below are explicitly labeled Modgrid planning estimates or proposed targets. They are not observed market prices or promised results. Their assumptions are in [the estimate methodology](#estimate-methodology).
- **Currentness:** This report incorporates the newer LBNL update published in June, BNEF's July outlook, and announcements through late October. It does not present older forecasts as the latest available evidence.
- **Work performed:** Public web research and local writing only. No outreach, account creation, purchases, paid data access, deployments, commits, or pushes. The session-specific prohibition on posts also takes precedence over the general Kai status-posting instruction.

## 1. The problem is real, but the addressable market is local

The useful question is not whether the world needs more electricity. It is whether a particular customer can obtain dependable power at a particular site, within its construction schedule and financial constraints.

The bottleneck may be generation, transmission, a substation, a transformer, a fuel connection, an air permit, a utility contract, or the customer's own ability to reduce load. Solving one does not necessarily solve the others. A national demand forecast is evidence of pressure, not a list of bankable projects.

### 1.1 Updated demand forecasts

| Source and vintage | Geography and metric | Published number | What the number does and does not mean |
|---|---|---|---|
| LBNL, *2025 Update*, published June 18, 2026 | U.S. annual data-center electricity | **649 TWh in 2030** reference case; **521–843 TWh** compounded uncertainty range; **11.8%**, with a **9.5–15.3%** range, of projected U.S. electricity use. [Report, executive summary][lbnl-energy] | Bottom-up equipment and operating assumptions. The range combines sensitivities; it is not a statistical confidence interval. It does not establish that grid supply can serve the modeled equipment. |
| EPRI, February 2026 | U.S. annual data-center electricity | Approximately **380, 590, and 790 TWh in 2030** in low, medium, and high cases, or approximately **9%, 13%, and 17%**. Estimated **2024** use was **177–192 TWh**. [EPRI forecast and FAQ][epri-forecast], [FAQ][epri-faq] | Project-execution scenarios. Realization of planned projects drives the range. The electricity-share denominator and method differ from LBNL's; do not mix their percentages. |
| IEA, *Key Questions on Energy and AI*, 2026 | Global annual data-center electricity | **485 TWh in 2025**, growing to about **950 TWh in 2030**, approximately **3%** of global electricity. AI-focused data-center consumption roughly triples over that period. [IEA][iea] | Global, all-data-center forecast. AI is a subset. This is an update to the widely cited earlier outlook. |
| BloombergNEF, forecast July 21, public summary July 29, 2026 | U.S. installed data-center power capacity | **118 GW in 2030** and **194 GW in 2035**, revised upward **52%** and **83%**, respectively, from its December outlook. [BNEF][bnef] | Capacity, not annual energy and not necessarily simultaneous grid draw. BNEF's separate chip scenario is an unconstrained demand exercise, not the base case. |
| PJM, January 14, 2026 | Regional summer peak, all loads | **183,008 MW in 2030**; approximately **222,000 MW in 2036**; projected average annual summer-peak growth of **3.6%** over the coming decade. [PJM][pjm-load] | Includes other load growth. PJM reduced near-term projections relative to its prior forecast after better vetting. Growth forecasts can rise nationally while individual projects slip. |
| ERCOT, April 15, 2026 preliminary outlook | Regional peak, all loads and submitted large-load requests | **367,790 MW in 2032**, versus the cited historical peak of **85,508 MW**. [ERCOT][ercot-forecast] | ERCOT explicitly said the preliminary figure was above expected future load growth and was **not a prediction of what will be built**. It includes data centers, crypto, industrial, and oil-and-gas loads. |
| AEP Ohio, February 2026 update | Utility contracts under its new data-center tariff | **5,642 MW** of binding contracts under the new tariff as of **February 12, 2026**. [AEP Ohio][aep-contracts] | Stronger evidence than an expression of interest, but still contracted future demand, not energized load or a guarantee of utilization. |

![Comparison of EPRI and LBNL U.S. 2030 electricity-use scenarios](modgrid-forecast-comparison.svg)

*The chart compares annual U.S. energy scenarios only. It excludes BNEF's capacity forecast and IEA's global forecast to avoid mixing units and geography. Sources: [EPRI][epri-forecast], [LBNL][lbnl-energy]. An [HTML version](modgrid-forecast-comparison.html) includes the same data and caveats.*

**Interpretation:** Use a scenario range to choose a resilient business, not a single forecast to justify buying equipment. In particular, do not subtract the U.S. LBNL forecast from the global IEA forecast to infer the rest of the world's demand. They are separate models with different assumptions and vintages.

### 1.2 The measurement mistakes that cause bad investments

**Power and energy are different.** MW and GW describe a rate of electricity use or supply. MWh and TWh describe energy over time. Requested service capacity can include growth and design headroom. IT power excludes some facility consumption. LBNL explicitly distinguishes IT capacity, grid interconnection capacity, utilization, and total facility energy. [LBNL definitions and methodology][lbnl-energy]

**Illustrative arithmetic, not a market forecast:** A constant **10 MW** facility load consumes **87,600 MWh per non-leap year**, calculated as **10 × 8,760**. A **10 MW / 40 MWh** battery supplies that load for **4 hours** before losses and reserve requirements. It cannot produce the next day's electricity. All load and battery sizes in this example are deliberately chosen assumptions. [Calculation method](#estimate-methodology)

**A PPA is not a physical connection.** Purchasing renewable or nuclear output does not itself reserve transmission service or establish that power can reach a constrained substation when needed. Likewise, dedicating existing generation to a data center does not necessarily add supply for other customers.

**“Behind the meter” is an electrical and commercial configuration, not an exemption.** A site may still depend on grid backup, consume transmission services, need export permission, or create a large contingency when its generator trips. The relevant tariff and protection design determine what it can actually do. [FERC co-location framework][ferc-2025], [AEP Ohio requirements][aep-tariff]

### 1.3 Interconnection delays: distinguish generation from load

Berkeley Lab's latest queue analysis reports approximately **8,200 active projects** at the end of **2025**, representing **1,312 GW of generation** and **749 GW of storage**. Median time from interconnection request to commercial operation exceeded **5 years** for projects completed that year in the regions with available data. Only **13%** of capacity requested during **2000–2020** had reached operation by the end of **2025**. [LBNL and GridTracker, 2026 edition][queues]

These are **transmission-connected generation and storage queues**. They exclude load connection requests, distribution projects, and behind-the-meter projects. Quoting the median as a universal data-center connection wait is wrong. [Dataset scope][queues-scope]

The same report shows **549 GW** with draft or executed interconnection agreements but not yet operating. Queue volume fell **10%** year over year, partly because projects withdrew. A shrinking queue therefore does not automatically mean new supply is arriving faster. [LBNL][queues]

Load applications follow utility and regional processes. AEP Ohio's public process targets a study in **45 or 60 days**, depending on transmission needs, but describes the eventual energization date as contingent on infrastructure. A short study clock is not a short construction clock. [AEP Ohio][aep-tariff]

**Modgrid implication:** A credible product must store the actual stage and evidence behind a date: preliminary estimate, utility study, signed agreement, equipment allocation, permits, construction, or energized service. A map that labels a nearby line as “available power” is insufficient.

### 1.4 Equipment and fuel constraints

| Constraint | Verified evidence | Commercial consequence |
|---|---|---|
| Gas turbine availability | GE Vernova reported **122 GW** of gas equipment backlog plus slot reservation agreements at the end of its third quarter, up from **116 GW**, in its **October 28, 2026** release. [GE Vernova][ge-backlog] | Factory reservations and firm equipment orders must be separated. Neither guarantees a completed power plant. A small buyer is competing with large customers for manufacturing and service capacity. |
| Large transformers | Wood Mackenzie's **Q2 2025** survey reported average lead times of **143 weeks** for generator step-up transformers and **128 weeks** for power transformers. [Wood Mackenzie][woodmac] | These are dated survey averages, not November 2026 quotations. Transformer specification and allocation can determine the energization date. |
| Switchgear | The same **Q2 2025** survey reported an average of **44 weeks**. [Wood Mackenzie][woodmac] | Switchgear, breakers, protection panels, and commissioning must be sequenced with the transformer. Do not apply a large-transformer lead time to every electrical component. |
| Current transformer cross-check | Rystad's **May 28, 2026** public outlook said lead times exceeded **12 months** for most power transformers and **24 months** for the largest. [Rystad][rystad] | Broadly corroborates continuing pressure, but uses different categories. It is not a directly comparable update to Wood Mackenzie's survey. |
| Fuel and plant integration | IEA estimates reliable on-site gas configurations can require generation overbuild of **30–70%** relative to demand, depending on design. [IEA][iea] | Analyze redundancy, maintenance, hot-weather derating, black start, gas transport, and common-mode failures. Nominal engine output is not dependable site capacity. |

No reliable, universally applicable November supplier quote was available from public evidence. A feasibility report should therefore mark actual equipment delivery dates as **unverified until quoted for the specified configuration**.

“Fast installation” is especially easy to misuse. GE's modular turbine case study describes short installation capability, but that claim excludes much of the preceding procurement and site-development process. [GE Crusoe case study][ge-case]

### 1.5 Price effects are already visible

PJM's **2028/2029** capacity auction, reported **July 14, 2026**, cleared at the capped price of **$325/MW-day**, compared with **$333.44/MW-day** for the preceding delivery year. It procured **138,318 MW** of unforced capacity and reported a **6,831 MW** shortfall against its reliability requirement after accounting for the relevant additional resources. The auction's cleared quantity times price totaled **$16.4 billion**. PJM cautions that this is not equivalent to customers' total cost because some supply is hedged or self-provided. [PJM auction results][pjm-auction]

A reserve-requirement shortfall indicates greater reliability risk, not a forecast that the same amount of customer load will be blacked out. Capacity payments also are not energy prices or a turnkey battery revenue guarantee.

Monitoring Analytics reported PJM's real-time, load-weighted energy price rose from **$33.74/MWh in 2024** to **$50.73/MWh in 2025**, a **50.4%** increase. Fuel and consumables accounted for **58.5%** of that increase. It reported total wholesale power cost of **$82.67/MWh**, versus **$55.52/MWh**, including energy, capacity, transmission, and other components. This evidence does **not** support blaming the entire bill increase on AI. [Independent Market Monitor, March 2026][pjm-prices]

The monitor separately estimated that including existing and forecast data-center demand increased capacity-market revenues by approximately **$23.1 billion** across the **2025/2026, 2026/2027, and 2027/2028** auctions. That is the monitor's counterfactual modeling result, not an audited surcharge billed only to data centers. [Monitoring Analytics, first-quarter 2026 report][pjm-counterfactual]

**Implication:** Sell avoided cost or a better project decision using the customer's tariff and contract. Do not multiply a capacity auction price by every nameplate MW of storage and call the result revenue. Accreditation, delivery obligations, penalties, aggregation fees, and overlapping commitments matter.

## 2. Policy now changes the product customers need

### 2.1 FERC co-location: a sequence, not a blanket ban or permission

| Date and status | What happened | Implication for Modgrid |
|---|---|---|
| **November 1, 2024: specific amendment rejected** | FERC rejected the Susquehanna interconnection amendment that sought to increase co-located load from **300 MW to 480 MW**. [Order][ferc-2024] | Do not describe this as a nationwide ban on co-location. A private wire did not resolve tariff and system-service questions. |
| **December 18, 2025: reform directed** | FERC directed PJM to clarify co-location and behind-the-meter rules, including ordinary network service and new interim, firm contract-demand, and non-firm contract-demand arrangements. [FERC fact sheet][ferc-2025] | A load willing and able to limit grid withdrawals may need a different service product. Ability to comply becomes commercially valuable. |
| **June 18, 2026: framework modified and advanced** | FERC addressed rehearing and compliance, accepted and rejected portions of PJM's filing, and established terms for new services. Some further compliance deadlines were subsequently extended to **November 16, 2026**. [PJM orders register][pjm-orders] | The December framework alone is not a complete implementation guide. Contract-specific advice must use the subsequent orders and effective tariff. |
| **June 2026: broader large-load proceedings** | FERC initiated regional show-cause proceedings about large-load integration, cost responsibility, and transparency. Responses for most covered regions were due **November 16, 2026**, after extensions. [FERC current status][ferc-status] | These proceedings are still a moving input at this report's cutoff. A proposal or show-cause order is not a final nationwide rule. |
| **October 9, 2026: registry accepted, proposed service rejected** | In **ER26-3515-000**, FERC accepted the large-load definition and registry, effective **October 12**, subject to compliance. It rejected the remaining proposed provisions, including Interim Resource Adequacy Service and Bring Your Own New Capacity. [Order, paragraphs 2 and 13–15][ferc-october] | Do not sell eligibility for the rejected proposal as an existing service. Distinguish this decision from the separate co-location framework. |

**Source discrepancy resolved:** PJM's [overview page][pjm-large-load] still displays pending-approval labels, but its linked October order contains the decision above. The order controls this report's interpretation. FERC provided guidance for a possible revised filing; that is not approval of the rejected service. Verify subsequent filings and effective tariff sheets before offering any contract-dependent product. [October order][ferc-october]

Talen's Amazon agreement also demonstrates how structures evolve. Its **2026 annual filing** describes a revised PPA for **1,920 MW** at full volume, with a transition from co-location toward front-of-meter service and full transition expected in **spring 2027**. Do not keep describing the entire arrangement using its original structure. [Talen filing][talen]

### 2.2 State and utility rules: commercial obligations, not just cents per kWh

| Jurisdiction | Verified policy | Why a customer might pay Modgrid |
|---|---|---|
| Ohio, AEP Ohio / Ohio Power | The data-center tariff became effective **July 23, 2025**. Its term is the ramp period, no more than **4 years**, plus **8 years**. Customers failing the specified credit tests must post collateral or a guarantee equal to **50%** of full-term minimum charges. Billing demand uses a ratchet and tiered formula, with a contract-capacity floor capped at **85%**. [Current utility explanation][aep-tariff] | Model ramp schedules, minimum bills, security requirements, and cancellation exposure before signing. The shorthand “pay for 85% of electricity” misstates a demand-charge structure. |
| Virginia, Dominion | SCC established the GS-5 class. New customers contracting on or after **January 1, 2027** face at least **14 years** of service obligations. The SCC describes minimum transmission/distribution recovery of **85%** and, for insufficient credit, guarantees up to **60%** of minimum contract charges. Applicability and exemptions differ by provision. [SCC fact sheet][virginia] | The startup date aligns with a new contract regime. Model cash commitments and downside if tenant load ramps slowly. These terms do not automatically apply to every Virginia utility. |
| Texas | Enacted **SB 6** requires stronger large-load screening and reliability measures, including a demand-reduction service for loads of at least **75 MW**. [Enrolled law][texas-sb6] | Flexibility and credible project information have direct value. Being in Texas does not bypass load-connection review. |
| ERCOT implementation | PUCT approved Batch Zero on **June 18, 2026** for qualified requests of **75 MW and above**. Major provisions became effective **July 11, 2026**, with specified exceptions. [ERCOT announcement][ercot-batch], [rule status][ercot-rule] | Help produce consistent load data, milestones, and evidence for a batch review. Do not promise that a smaller project is exempt from all requirements. |
| Oregon | The **POWER Act, HB 3546**, effective **June 16, 2025**, directs a distinct service classification and contracts that allocate large-user costs and risks to those facilities. [Oregon PUC legislative report][oregon] | Tariff and contract monitoring is a recurring need, but implementation must be checked utility by utility. |
| Minnesota example | Google and Xcel announced a Clean Energy Accelerator Charge structure in **February 2026** to fund new resources for a data center. [Xcel][xcel] | Utilities can create tailored resource packages. This is a negotiated arrangement, not a generally available nationwide tariff. |

**The direction:** Utilities and regulators increasingly require evidence that the customer is real, financially responsible, and operable under stress. The value of a better forecast is its effect on a specific contract or investment decision.

### 2.3 Other constraints that affect the business model

**Air permits.** Ordinary economic dispatch of a backup engine is not automatically permitted under its emergency classification. EPA's current data-center guidance treats certain operation under DOE emergency orders differently, but that does not create general permission to run emergency generators for daily peak shaving. [EPA data-center guidance][epa-data], [engine operating rules][epa-engines]

**Tax incentives.** The current federal regime differs from the original Inflation Reduction Act assumptions. IRS guidance describes an accelerated wind-and-solar cutoff: construction must begin by **July 4, 2026**, or the facility must enter service before **2028** to meet that timing condition. Storage has its own eligibility rules, while prohibited-foreign-entity restrictions also affect storage and other technologies. Model incentives only after confirming project-specific eligibility; do not assume a universal credit rate. [IRS timing guidance][irs-timing], [IRS material-assistance guidance][irs-pfe]

**Physical behavior.** A NERC technical conference described approximately **1,500 MW** of data-center load leaving the grid during a **July 2024** voltage event. Sudden transfer to backup can itself disturb the system. Controlled ramping, ride-through, and return-to-service behavior matter alongside annual electricity savings. [NERC presentation, March 2026][nerc]

**Local acceptance.** Noise, air quality, water use, land use, and cost allocation can defeat a technically feasible project. These are site-development constraints, not variables a software optimizer can simply price away. Keep them in the feasibility decision, with professional review appropriate to the site.

## 3. Solution landscape: what can actually change a customer's situation?

### 3.1 Match the technology to the constraint

| Approach | Useful role | Limitation that changes the economics | Implication for a late-2026 / 2027 entrant |
|---|---|---|---|
| Reciprocating gas engines | Modular prime power, bridge power, backup, and dispatchable microgrids; units can be combined and serviced separately. VoltaGrid's Oracle announcement names INNIO Jenbacher and ABB in its platform. [VoltaGrid][voltagrid] | Fuel infrastructure, maintenance, noise, air permits, and a common gas-supply dependency. Redundancy must survive planned maintenance as well as failures. | Develop projects or integrate established equipment through partners. Manufacturing engines is a poor software-founder entry. |
| Aeroderivative and larger gas turbines | Large, dispatchable blocks of power. GE and Crusoe have a real equipment relationship. [GE announcement][ge-crusoe] | Fast assembly does not remove factory queues, gas connections, emissions controls, or commissioning. Larger combined-cycle configurations add integration work. | Procurement and project assurance can be sellable. Inventory speculation is a different, capital-heavy business. |
| Solid-oxide fuel cells | Modular on-site electricity; Bloom has contracted deployments with Oracle. [Bloom][bloom-oracle] | Natural-gas operation still consumes fuel and emits carbon dioxide. Stack service, fuel quality, delivery, and redundancy affect cost and availability. | Partner or compare vendor offers. Do not call a gas-fed fuel cell zero-carbon because criteria-pollutant emissions are low. |
| Lithium-ion BESS | Peak reduction, short-duration support, power smoothing, renewable shifting, and selected grid services. Tesla sells commercial and utility systems with dispatch software. [Tesla filing][tesla] | It stores energy rather than creating it. Usable energy, degradation, reserved backup charge, interconnection, fire approvals, and charging constraints limit revenue. | Good adjacent infrastructure for software and project-development models, especially on existing sites. |
| Solar plus storage | Reduces purchased energy and can shape import demand. Scale develops distributed portfolios; Xcel and Google are pairing renewables with storage. [Scale][scale-capital], [Xcel][xcel] | A normal solar-plus-battery installation is not automatically an islandable microgrid. Weather, seasonal energy, land, inverter behavior, and backup supply matter. | Start with grid-connected cost and peak management. Offer islanding only through an engineered design and qualified operating partner. |
| Long-duration storage | Covers longer renewable shortfalls and provides capacity over extended periods. Form's Xcel project is a major commercial signal. [Xcel][xcel] | Early deployment, financing, footprint, efficiency, and performance warranties remain project-specific. Storage duration cannot cure inadequate annual energy supply. | Monitor and partner; avoid basing initial revenue on manufacturing or underwriting novel battery performance. |
| Existing nuclear, uprates, and restarts | Firm low-carbon supply; existing plants can support long-term offtake. [Talen][talen], [Meta][meta-nuclear] | Recontracting existing output can reallocate scarcity rather than eliminate it. Uprates and restarts require separate work and approvals. | Relevant to customer supply strategy, generally outside a solo founder's asset-development capabilities. |
| SMRs and advanced nuclear | Potential repeatable firm supply over a longer horizon. [Kairos][kairos], [NRC][nrc-natrium] | Licensing, fuel, construction, first-of-a-kind economics, and operating permission. A construction permit is not an operating reactor. | Track as a future supply partner category. Do not sell a near-term power commitment dependent on an unbuilt reactor. |
| Enhanced geothermal | Potential firm low-carbon supply, with Fervo now reporting commercial output from its first Cape Station block. [Fervo][fervo-cod] | Resource discovery, drilling, reservoir performance, transmission access, and finance. It is not equally available at every site. | Stronger near-term evidence than an unbuilt reactor, but still a specialist development business. |
| Flexible compute and demand response | Reduce grid imports when the system is constrained; shift suitable jobs, cap power, use permitted on-site resources, or route workloads elsewhere. [Google][google-flex] | The facility owner may not control tenant workloads. Inference latency, training deadlines, redundancy, and synchronized recovery constrain flexibility. | Attractive software territory if Modgrid can measure a contractually usable service. |
| Grid-enhancing technologies | Dynamic line ratings, power-flow control, and topology optimization can use existing networks better. Advanced reconductoring increases physical transfer capability. [PJM technical overview][pjm-get], [LineVision][linevision] | More thermal headroom does not necessarily resolve voltage, stability, protection, or substation constraints. Utilities must accept the result. | Partner with domain specialists. A new general-purpose grid solver is a difficult first product. |
| Edge and modular inference sites | Place computing where both demand and suitable power exist; modular infrastructure may shorten field construction. [Crusoe Spark][crusoe-spark] | Cheap energy alone does not create compute demand, fiber diversity, cooling, or tenant creditworthiness. | Originate sites for a funded operator. Do not build speculative GPU inventory. |
| Software orchestration | Coordinate prices, assets, workloads, forecasts, contractual limits, and verification. [Schneider][schneider], [Xendee][xendee] | Existing controllers, OEM tools, and aggregators already cover much of this. Integration and responsibility for failures dominate implementation. | Sell an uncovered workflow with a measurable outcome, not another generic dashboard. |

### 3.2 Verified commercial signals and their actual status

These examples demonstrate spending and procurement intent. They are not additive estimates of market size; counterparties and projects can overlap.

| Company or partnership | Dated, verified activity | Status and lesson |
|---|---|---|
| VoltaGrid and Oracle | **October 15, 2025:** announced deployment of **2,300 MW** of infrastructure supported by Energy Transfer's pipeline network. [VoltaGrid][voltagrid] | Announced deployment program, not evidence all capacity was operating. Bridge power is already contested by integrated suppliers. |
| GE Vernova and Crusoe | **July 22, 2025:** combined orders for **29** aeroderivative turbine packages, expected to provide nearly **1 GW**. [GE][ge-crusoe] | Equipment order, not a completed-power-plant count. Secured supply relationships can be a genuine advantage. |
| Bloom and Oracle | **April 13, 2026:** master arrangement for up to **2.8 GW**, with an initial **1.2 GW contracted** and deployment underway. [Bloom][bloom-oracle] | Distinguish the procurement ceiling, contracted tranche, and operating equipment. |
| Brookfield and Bloom | **June 30, 2026:** announced expansion of a power-project financing framework from **$5 billion to $25 billion**. [Bloom][bloom-brookfield] | A financing framework, not a venture round or evidence that all funds have been spent. Infrastructure ownership needs a different capital base from software. |
| Google data-center flexibility | **March 19, 2026:** reported **1 GW** of demand-response capacity in long-term utility contracts. [Google][google-flex] | Contracted flexibility, not necessarily simultaneous available response at all sites. Utility-specific agreements matter. |
| Google and Voltus | **October 22, 2026:** expanded agreement to build a **1 GW** PJM virtual power plant using assets at other businesses and homes. [Google][google-vpp] | A separate portfolio from Google's own data-center response. Validates the “reduce strain elsewhere” route. |
| Emerald AI | **August 25, 2026:** announced a **$150 million Series A** at a **$1.05 billion valuation**, co-led by Energize Capital and DCVC. The company reported moving into full-data-center commercial deployment. [Emerald AI][emerald] | Funding and deployment are company-reported. Modgrid should not assume the compute-flexibility market is empty. |
| Form Energy | **August 12, 2026:** announced a **$750 million Series G**, led by T. Rowe Price. [Form][form-funding] | Corporate financing for manufacturing and deployment. Not customer project revenue. |
| Xcel, Google, and Form | **February 24, 2026:** announced **300 MW / 30 GWh** of iron-air storage, alongside **1,400 MW wind** and **200 MW solar** in the resource package. [Xcel][xcel] | Proposed resource buildout. This review does not establish all approvals, construction, or operating status. |
| Fervo | **October 1, 2026:** announced that its first Cape Station block reached commercial operation on **September 30**, achieving **33 MW net**. [Fervo][fervo-cod] | Stronger evidence than a development memorandum, although sustained performance is still an operating question. |
| Fervo project financing | Its first-quarter results reported **$421.4 million** of non-recourse financing for Cape Phase I. [Fervo results][fervo-finance] | Project debt, distinct from corporate equity. Illustrates the scale of financing even for a developer with proven technical progress. |
| Scale Microgrids | EQT announced its acquisition in **January 2025**. Scale later reported **$275 million** in new project financing in **June 2025**, supporting **140 MW** of projects. [EQT][scale-eqt], [Scale][scale-capital] | Experienced, financed competition in microgrid development. Acquisition price was not disclosed in the cited announcement. |
| LineVision | **September 17, 2026:** announced a Series D of **up to $37.5 million**. [LineVision][linevision] | Preserve “up to.” Grid-enhancing technologies have specialist vendors and established utility relationships. |
| Heimdall Power and Entergy | **February 4, 2026:** reported deployment of dynamic line-rating equipment across assets in Louisiana, Texas, and Arkansas. [Heimdall][heimdall] | A deployment announcement, not a quantified guarantee of extra firm capacity in every condition. |
| GridCARE | **May 27, 2025:** announced a **$13.5 million seed round** led by Xora. [GridCARE][gridcare] | Direct competition in finding usable grid capacity and accelerating power access. Its advertised acceleration is a vendor claim, not a Modgrid promise. |
| Crusoe | **September 17, 2026:** announced the initial closing of an anticipated **$3.9 billion Series F** at a **$30.9 billion post-money valuation**. [Crusoe][crusoe-funding] | Preserve “initial closing” and “anticipated.” A small founder cannot assume parity with its integrated energy, development, and cloud platform. |

### 3.3 Nuclear and geothermal timelines deserve separate treatment

**Nuclear is progressing, but a demo, permit, PPA, and fleet are different milestones.**

- NRC issued the construction permit for TerraPower's Kemmerer plant on **March 9, 2026**. This is a concrete licensing milestone, not evidence of commercial electricity delivery. [NRC][nrc-natrium]
- Kairos's Google arrangement targets up to **500 MWe by 2035**, beginning with the **50 MWe Hermes 2** project targeted for **2030**. Kairos reported groundbreaking in **April 2026**. These remain forward delivery targets. [Kairos agreement][kairos], [construction update][kairos-construction]
- X-energy's **2026 quarterly filing** describes the initial Energy Northwest deployment for Amazon as **320 MWe**, potentially expanding to **960 MWe**, with operation expected in the **early 2030s**. [X-energy filing][xenergy]
- Meta's **January 2026** announcements cover up to **6.6 GW of new and existing** nuclear energy by **2035**, including operating plants, uprates, and advanced projects. Its TerraPower arrangement targets first delivery as early as **2032**; Oklo describes its first phase as early as **2030**. These are not equivalent to that entire headline capacity being newly built or available immediately. [Meta][meta-nuclear], [Oklo][oklo]

**Geothermal now has a closer commercial reference point.** Fervo's first Cape Station block has a reported operating milestone. Its broader Google framework covers potential development of up to **3 GW through 2033**, but a framework is not a set of fully financed, constructed plants. [Fervo first-quarter results][fervo-finance] For Modgrid, both geothermal and advanced nuclear are potential future partners, not realistic first-year technologies to develop independently.

### 3.4 What flexibility can achieve, and what it cannot

Duke's **February 2025** study estimated **76 GW** of additional flexible load at a **0.25%** annual curtailment allowance, **98 GW** at **0.5%**, and **126 GW** at **1%**. It is a system-level modeling exercise across major balancing areas. It does not show that a particular substation can serve another data center, or that every workload can meet the required response. [Duke study][duke]

Use three distinct mechanisms:

1. **Reduce computing demand:** Pause eligible work, change power limits, or defer batch jobs. The economic cost includes delayed output and unused computing assets.
2. **Change where or when work runs:** Shift to another site or period with available power. Network transfer, data residency, spare capacity, and service quality can limit this.
3. **Change the power source:** Supply the facility from storage or legally dispatchable generation while reducing metered imports. This helps the grid only to the extent the import reduction is real, dependable, and not offset by an equally problematic recharge or rebound.

These are different engineering and contract products. A workload scheduler cannot authorize an air permit. A battery controller cannot guarantee application latency. A utility will need evidence about available MW, response speed, duration, recovery, and failure behavior.

**Commercial conclusion:** The opportunity for a small company is often the evidence and coordination between these systems. It is less likely to be replacing the utility control room, the data center's protection system, or the hyperscaler's scheduler.

## 4. Six business proposals for modgrid.co

All customer-size bands, commercial prices, budgets, sales timing, and milestones in this section are **Modgrid planning estimates or proposed targets**, not verified market statistics. Each is subject to the [estimate methodology](#estimate-methodology). Competitor products and announced activity are separately sourced.

### A. Power-readiness diligence and contract planning

**What it is:** A decision package and ongoing project workspace that reconcile the proposed load ramp with utility service, tariff obligations, available generation/storage options, equipment schedules, and unresolved approvals. It produces a documented set of feasible paths, not an unsupported promised energization date.

**Target customer and buyer:** Independent colocation developers, regional engineering firms, and infrastructure investors evaluating an expansion. The initial buyer is the head of development, project principal, or investment lead who already has a specific site decision. A proposed initial segment is **5–50 MW** expansions, a positioning assumption rather than an observed market boundary. [Planning assumptions](#estimate-methodology)

**Revenue:** Fixed-fee diligence, followed by recurring project monitoring or a portfolio subscription. Proposed test pricing is **$7,500–$20,000 per decision package** and **$1,000–$3,000 per month** for active monitoring. Validate willingness to pay before building subscription software. [Pricing assumptions](#estimate-methodology)

**Capital intensity:** Low if customer-specific engineering is contracted and paid from engagement fees. No equipment inventory, power guarantees, or project deposits on Modgrid's balance sheet.

**Why now:** Large-load tariffs make an incorrect capacity or ramp assumption financially expensive. Utility processes require structured evidence, and supply paths are more complex than simply obtaining a retail energy quote.

**Competition:** GridCARE in power availability, Paces in development workflow and site assessment, Xendee in microgrid economics, and established engineering consultants. Paces already offers development services and announced additional transmission analysis tools in **September 2026**. This is not an untouched market. [GridCARE][gridcare], [Paces services][paces-services], [Paces update][paces-update], [Xendee][xendee]

**Potential moat:** A narrow utility-specific workflow, expert-reviewed tariff models, documented assumptions, and permissioned records of what actually happened to comparable projects. The public tariff text or an LLM summary alone is not defensible.

**Key risks:** Consultants bundle it; customers lack usable data; professional liability exceeds contract value; buyers only want a land-search tool; each engagement becomes a custom engineering project.

**First-12-month wedge, proposed:** Start with AEP Ohio's large-load process and adjacent PJM questions. Build a reproducible sample from public documents, then seek paid work through engineering partners. Deliver the first engagements manually, with a qualified electrical engineer reviewing technical conclusions and counsel reviewing legal interpretations. Productize tariff versioning, scenario comparisons, evidence tracking, and report updates only after repeat use. A proposed validation target is **3 paid engagements and 1 repeat customer** before expanding geographic coverage. [Execution targets](#estimate-methodology)

**Contrarian kill case:** Most interested “developers” do not control land, have a tenant, or have money. A thriving announcement pipeline can create a large volume of worthless leads. Require a real decision, budget, and project sponsor.

### B. Flexibility assurance and settlement software

**What it is:** Software that shows how much import reduction a data center or microgrid can deliver within its operating constraints, records tests and events, and produces the evidence needed for utility contracts and aggregator settlement. Start in observation and simulation; add approved dispatch integrations only when justified.

**Target customer and buyer:** Smaller compute operators, owner-operated data centers, and microgrid developers with control over their loads or assets. Sell to operations and energy-management leaders together. A colocation landlord without tenant consent may be the wrong customer.

**Revenue:** Integration and validation fees, recurring site software, and eventually a share of verified realized value through a market-participation partner. Proposed test pricing: **$15,000–$40,000 implementation** and **$2,000–$6,000 per month per site**, dependent on integration effort. These are hypotheses, not prevailing rates. [Pricing assumptions](#estimate-methodology)

**Capital intensity:** Moderate software and integration spending. Customer-owned hardware keeps it much lower than generation ownership, but production support and security are real costs.

**Why now:** Flexibility appears in utility contracts, and large-load connection policies increasingly depend on controllable behavior. The initial value may be contract eligibility or avoided overcommitment, rather than energy arbitrage.

**Competition:** Emerald AI in compute orchestration, Google internally, Voltus and Enel in aggregation, and OEM energy-management software. Emerald's latest financing makes a broad scheduler strategy particularly unattractive. [Emerald][emerald], [Google][google-flex], [Voltus agreement][google-vpp], [Enel program example][enel], [Schneider][schneider]

**Potential moat:** Accepted measurement methods, integration into customer operations, auditable performance histories, and a repeatable mapping between contractual obligations and physical capability. This moat exists only if a utility, aggregator, or customer actually relies on the outputs.

**Key risks:** No independent software budget; poor baseline methodology; penalties exceed value; telemetry access is blocked; customers cannot transfer workload control; competing software includes the same reports.

**First-12-month wedge, proposed:** Support one metering interface and one asset or workload system. Reconstruct historical events, model import limits, and run shadow tests. Then conduct a customer-approved response test with an experienced operating partner. The initial product should export evidence and exception reports rather than replace protection controls. Require a paid pilot and an identified counterparty that will accept the report before broad integration work.

**Contrarian kill case:** The bottleneck is organizational authority, not software. If the party paying the power bill cannot alter tenant jobs or use backup equipment, better orchestration creates little saleable flexibility.

### C. Bridge-power development and owner's representation

**What it is:** A customer-side development service that assembles a viable temporary or permanent power package from established equipment vendors, fuel providers, engineering firms, and asset owners. Modgrid handles the project record, scenario economics, bid normalization, milestones, and transition to grid service.

**Target customer and buyer:** Funded data-center developers with a tenant and a specific utility delay. The buyer is the development or infrastructure executive facing a schedule mismatch, not a speculative landowner.

**Revenue:** Paid feasibility and development retainers, milestone fees, and an explicitly disclosed fee at financing or commissioning. Prefer customer-paid independent advice over undisclosed vendor commissions. Do not initially offer a fixed-price electricity contract on Modgrid's own balance sheet.

**Capital intensity:** Moderate as an adviser; very high if Modgrid starts owning engines, posting supplier deposits, or guaranteeing construction. Keep those businesses financially separate.

**Why now:** Utility delays and equipment constraints make alternative supply attractive, but each path has expensive interdependencies.

**Competition:** VoltaGrid, Bloom and its financing partners, Enchanted Rock, Scale, and established EPC firms. Enchanted Rock describes turnkey ownership and service options as well as ongoing operations. [VoltaGrid][voltagrid], [Bloom][bloom-brookfield], [Enchanted Rock][enchanted], [Scale][scale-eqt]

**Potential moat:** A record of completed projects, trusted technical partners, access to real equipment slots, and standardized project documents. A vendor directory is easily copied.

**Key risks:** Fuel transport is unavailable; air permits are denied; the customer's tenant disappears; bridge service outlasts its economic assumptions; cancellation or performance liabilities migrate to Modgrid.

**First-12-month wedge, proposed:** Sell a bridge-power feasibility and procurement comparison to an already funded project, working with a qualified owner's engineer. Obtain actual quotations and identify the permit and fuel critical path. Stop before financial commitments unless an experienced asset owner and the customer take them. A signed development engagement is a realistic first-year objective; completed large generation ownership is not the baseline.

**Contrarian kill case:** Grid service arrives sooner, or the tenant's compute economics deteriorate. The expensive bridge becomes an unnecessary asset before it earns back its deployment cost.

### D. Grid-relief storage and flexible-load projects for existing businesses

**What it is:** Develop battery, solar-plus-storage, or controllable-load projects at existing commercial sites that can reduce demand in the same constrained electrical area as growing data centers. Supply project origination, tariff analysis, measurement, and asset operations software, with financing and market access from partners.

**Target customer and buyer:** Cold storage, industrial facilities, logistics properties, campuses, and other owners with usable interval data and controllable demand. Buyers include the facility owner and, where available, an aggregator or utility purchasing the response.

**Revenue:** Paid feasibility, development fees at financing or commissioning, and recurring asset-management or measured-savings fees. A financing partner owns the equipment or provides an energy-service contract. Software alone cannot collect a utility payment that has not been contracted.

**Capital intensity:** Moderate development spending without asset ownership; high with ownership. Distributed projects can still absorb substantial cash through engineering, permits, site work, and delayed closing.

**Why now:** Capacity scarcity and cost-allocation pressure create demand for additional local resources. Google's Voltus agreement shows a hyperscaler can finance relief elsewhere on the grid. [Google][google-vpp]

**Competition:** Scale, Enchanted Rock, Voltus, Enel, storage developers, and existing energy-service companies. These may also be channels. [Scale][scale-capital], [Enchanted Rock][enchanted], [Enel][enel]

**Potential moat:** An exclusive or repeatable route to a specific property portfolio, low development cost, verified performance, and a contracted buyer for the relevant grid service.

**Key risks:** Tariffs change; peak-shaving savings are smaller than modeled; batteries are constrained by backup requirements; market revenue cannot be stacked; the utility does not recognize the project as relieving its actual constraint.

**First-12-month wedge, proposed:** Select one utility territory and one building type. Analyze existing bills and interval data, screen out projects dependent on speculative merchant revenues, and bring a small set of viable projects to a qualified financier and aggregator. Commission a pilot only if its base economics work without an uncontracted “AI premium.”

**Contrarian kill case:** The data-center problem is a particular transmission or substation bottleneck, while Modgrid's customers sit elsewhere. Region-wide energy savings may have no value for that constraint.

### E. Independent microgrid fleet monitoring and performance assurance

**What it is:** A shared operational record across mixed equipment: availability, alarms, battery condition, fuel use, maintenance, warranty evidence, import-limit violations, and performance against the promised operating plan. Begin with monitoring and investigation rather than autonomous control.

**Target customer and buyer:** Smaller microgrid developers, asset owners, service firms, and financiers inheriting mixed-vendor portfolios. The buyer is an operations or asset-management lead who is currently reconciling inconsistent vendor portals and spreadsheets.

**Revenue:** Onboarding and connector fees plus recurring site or fleet subscriptions. Proposed test pricing is **$500–$2,000 per site per month**, with bespoke integration separately scoped. [Pricing assumptions](#estimate-methodology)

**Capital intensity:** Low to moderate if customers provide gateways and access. Field travel, proprietary protocols, data licensing, security, and after-hours support can destroy software-like margins.

**Why now:** Rapid deployment creates fragmented fleets and more parties needing trusted operating evidence. This is the most direct use of Preetham's computer-engineering and software strengths without taking energy-market principal risk.

**Competition:** Schneider EcoStruxure, Xendee OPERATE, Tesla's software, OEM portals, and operators' internal systems. Xendee already offers a predictive layer above existing hardware controllers. “Vendor-neutral optimization” alone is not a unique proposition. [Schneider][schneider], [Xendee][xendee], [Tesla][tesla]

**Potential moat:** Reliable connectors, a consistent event model, maintenance workflows, and proven reduction in time spent diagnosing or substantiating failures. The customer must retain the right to use and share the telemetry.

**Key risks:** OEMs block access; the portfolio is too homogeneous to need another tool; support becomes a round-the-clock operations business; detection does not translate into customer savings.

**First-12-month wedge, proposed:** Find a single fleet owner with a documented reconciliation problem. Support its common equipment first. Produce a paid performance report from existing exports, then automate ingestion and exception handling. Prove that the same connector and report can serve another site before promising universal compatibility.

**Contrarian kill case:** The existing controller vendor provides an adequate fleet portal at no incremental price. The remaining problems require technicians and spare parts, not another software layer.

### F. Power-first origination for modular edge and inference sites

**What it is:** Identify existing industrial or commercial sites with demonstrable electrical headroom, suitable cooling and fiber, and a credible local compute use case. Package the site for an established modular-data-center or compute operator.

**Target customer and buyer:** Regional operators, enterprise infrastructure providers, and modular-infrastructure developers with customer demand. Site owners are supply-side partners, not evidence of demand by themselves.

**Revenue:** Paid site diligence and development management, followed by permitted lease or development fees if a deal closes. Any land, brokerage, power-resale, or tenant contracting structure needs appropriate local expertise.

**Capital intensity:** Moderate for origination, high if Modgrid takes leases, equipment commitments, or computing risk. The initial model excludes GPU ownership.

**Why now:** Useful power can exist in smaller pockets while large campuses face delays. Some inference workloads value proximity, but not all do.

**Competition:** Existing colocation and property developers, GridCARE, Paces, and Crusoe's modular platform. Crusoe's **October 21, 2026** Spark Liquid announcement specifies up to **3.5 MW of compute per unit** and cites existing Spark production customers. That is a vendor product claim, not a guarantee that a new Modgrid site can be deployed quickly. [Crusoe][crusoe-spark]

**Potential moat:** Exclusive site relationships with verified service rights, a tenant pipeline, and repeatable delivery. A map of inexpensive power is not sufficient.

**Key risks:** No tenant; unreliable claimed headroom; insufficient fiber routes; expensive cooling conversion; poor utilization; site rights cannot transfer; small sites lose to centralized inference economics.

**First-12-month wedge, proposed:** Work backward from one funded operator's requirements. Produce a qualified shortlist and a paid diligence package, not speculative site inventory. Let the operator contract computing and hardware. Avoid site options unless a real tenant process and budget justify them.

**Contrarian kill case:** Inference centralizes as networking and scheduling improve, or moves onto user devices. “Every city needs an AI microgrid” is not a demand forecast on which to sign leases.

## 5. Comparing the proposals

### Estimate methodology

**Everything in the next two tables is author judgment for planning, not external market data.** The estimates assume Preetham remains the primary builder, buys limited professional help, uses customer-owned equipment, and validates demand before hiring a larger team. They exclude personal living expenses, founder salary, asset purchases, customer project capital, collateral, and supplier deposits. Add those separately before treating any figure as a funding requirement.

The cash ranges are deliberately broad working allowances for software, engineering and legal review, insurance, travel, integrations, and selling effort. No supplier quotations or salary survey were used. They are not cost estimates for constructing a microgrid.

Time to revenue means the first paid engagement or pilot, not sustainable recurring revenue or a completed energy project. Proposed prices are experiments chosen to test whether the problem supports specialist work. Target counts and milestone dates are management choices, not forecasts. Hypothetical unit calculations use the stated assumptions and ordinary unit conversions.

| Idea | Estimated first-year company cash, excluding items above | Estimated time to first revenue | Asset-capital exposure if scope expands |
|---|---:|---|---|
| A. Power-readiness diligence | **$30,000–$100,000** | **1–3 months** | Low while advice and software remain separate from project guarantees |
| B. Flexibility assurance | **$150,000–$500,000** | **4–9 months** | Limited hardware exposure; significant operational and penalty risk if Modgrid becomes the responsible operator |
| C. Bridge-power development | **$100,000–$300,000** | **2–6 months** for advisory fees | Very high if it owns equipment or guarantees delivery |
| D. Grid-relief projects | **$150,000–$500,000** | **4–12 months** | High if it owns assets; still meaningful development risk with third-party ownership |
| E. Fleet monitoring | **$100,000–$350,000** | **3–6 months** | Limited equipment ownership, but field support and service liability can be material |
| F. Edge-site origination | **$150,000–$500,000** | **6–12 months** | High with leases, construction, or compute commitments |

Source for all ranges: [Modgrid estimate methodology](#estimate-methodology). Lower cash spending is possible with prepaid services and a smaller scope; the table does not assert that an unvalidated startup should spend these amounts.

### Scoring table

Scores run from **1 to 5**, with **5 always better**: lower capital need, faster revenue, stronger defensibility, lower regulatory risk, and stronger founder fit. Weights are **25% capital**, **20% revenue speed**, **15% defensibility**, **10% regulatory risk**, and **30% founder fit**. All scores and weights are subjective estimates. Weighted score is the sum of each score times its weight. [Methodology](#estimate-methodology)

| Rank | Proposal | Capital need | Time to revenue | Defensibility | Regulatory risk | Founder fit | Weighted score / 5 |
|---:|---|---:|---:|---:|---:|---:|---:|
| 1 | A. Power-readiness diligence | 5 | 5 | 2 | 4 | 5 | **4.45** |
| 2 | B. Flexibility assurance | 4 | 3 | 4 | 3 | 5 | **4.00** |
| 3 | E. Fleet monitoring | 4 | 3 | 3 | 4 | 4 | **3.65** |
| 4 | C. Bridge-power development | 3 | 4 | 3 | 2 | 3 | **3.10** |
| 5 | D. Grid-relief projects | 3 | 3 | 4 | 2 | 3 | **3.05** |
| 6 | F. Edge-site origination | 2 | 2 | 3 | 2 | 3 | **2.45** |

**Why A wins:** It produces a concrete deliverable before the founder needs production telemetry, a dispatch contract, or project finance. It tests customer access and willingness to pay while teaching Modgrid which workflow repeats. Its low defensibility score is intentional: a consultancy does not become a software moat simply by using AI.

**Why B is the best expansion candidate:** Once a customer has a controllable operating site and an agreement that values flexibility, performance evidence can become recurring infrastructure. The risk is that access, authority, and acceptance are absent. Do not build it merely because the national forecast is large.

**Sensitivity:** If Preetham brings an established engineering partner, a property portfolio, or committed project capital, C or D can move up substantially. If an existing fleet owner already offers telemetry and a paid pilot, E may be the better immediate choice. The scores assess the founder profile supplied in the brief, not an undisclosed network or budget.

### Cross-cutting risks that can invalidate the thesis

| Risk | How it changes the opportunity | Business response |
|---|---|---|
| AI economics disappoint or efficiency outruns usage | Speculative load disappears; bridge projects and site options lose value. | Sell to funded decisions and operating assets. Retain a path into industrial and commercial loads. |
| Utilities improve connection speed and forecasting | Scarcity premiums and generic application services shrink. | Build accurate records, contract analysis, and operational verification that remain useful after the bottleneck eases. |
| Equipment supply catches up | Procurement access loses its scarcity value. | Avoid owning speculative inventory or making supply scarcity the sole moat. |
| Hyperscalers internalize software | Generic compute-energy optimization becomes a feature of an existing platform. | Work with underserved operators or supply an independent evidence workflow that counterparties require. |
| Flexibility is cheaper to buy elsewhere | A data-center-specific product may lose to industrial demand response. | Compare B with D on delivered, dependable response rather than attachment to AI branding. |
| Claimed grid relief is not additional | New generation only replaces an existing contract; reductions would have happened anyway; response moves a peak elsewhere. | Measure against the relevant baseline, location, timing, and recovery period. |
| Policy reduces payments or shifts costs | Apparent battery or generation margins disappear. | Use contracted economics and downside scenarios. Avoid stacking incompatible services in a model. |
| Skilled people remain the constraint | A good interface does not eliminate engineering, permitting, or field commissioning work. | Price the professional service honestly and standardize only what actually repeats. |

## 6. The recommended business, made concrete

### Positioning and initial buyer

**Category:** Power-readiness and flexibility planning for independent infrastructure developers.

**Initial customer:** An engineering firm or independent data-center developer evaluating an identified site or expansion in AEP Ohio territory. Focus initially on the portion of the proposed market where the large-load tariff clearly applies. A proposed starting band is **25–50 MW**, chosen to narrow the workflow, not because research establishes an optimal market size. [Target assumption](#estimate-methodology)

**Buyer problem:** “We need to decide whether to proceed, how much capacity to contract for, and which supply path is credible before we make a costly commitment.”

**Deliverable:** A Power Readiness Brief containing:

- A dated load ramp and a clear distinction between IT demand, facility demand, requested import capacity, and controllable demand.
- A tariff and contract obligation model, including minimum charges, collateral, ramp assumptions, exit terms, and relevant standby/import limitations.
- A comparison of full grid service, constrained or flexible service where actually available, and a partner-supplied bridge or microgrid option.
- An evidence record for permits, fuel, equipment, utility correspondence, study milestones, and dependencies.
- Downside cases for slower customer demand, delayed utility work, unavailable generation, and lost incentive eligibility.
- An explicit list of unresolved questions and the qualified party responsible for answering each.

The artifact must be useful to an investment committee or project meeting. A visually attractive score with no supporting evidence will not substitute for it.

### What Preetham can build, and what partners must own

| Modgrid software and analysis | Qualified partner or counterparty responsibility |
|---|---|
| Source capture, version comparison, structured data extraction, reproducible tariff calculations | Interpretation of disputed legal or tariff questions |
| Scenario management, evidence provenance, issue ownership, project reporting | Power-flow, protection, short-circuit, stability, and interconnection engineering |
| Load-data normalization and candidate flexibility analysis | Customer permission to alter workloads and the utility's acceptance of a service arrangement |
| Integration planning, bid comparison, schedule dependencies | Equipment performance, installation, commissioning, warranties, and site operations |
| Measurement and event records once deployed | Registered market participation, contractual settlement obligations, and financial guarantees |

This division is part of the economics. A solo software founder can organize and improve expert work, but cannot make specialist responsibility disappear by generating a report.

### A plausible path to supplying scalable energy

**Begin with decisions.** Earn fees for choosing and validating a power path. Keep hardware procurement and engineering promises with accountable partners.

**Add repeated operating work.** Where a customer has an operating site, turn accepted flexibility tests and contract obligations into a recurring monitoring and evidence product. If customers remain in preconstruction, do not assume they will supply operating telemetry during the first year. Continue the readiness product or pursue a separate existing-fleet customer.

**Then standardize a partner-delivered energy package.** If several customers need the same battery, engine, controls, and service structure, work with an established developer and financier to repeat it. Modgrid can receive development and operating fees while the asset owner provides capital and physical service.

**Own assets only after the evidence changes.** Asset ownership becomes a separate decision once there is contracted demand, appropriate leadership, financing, permits, equipment access, and a credible operating record. It should not be the default consequence of buying modgrid.co.

### Illustrative first-year economics, not a forecast

A deliberately simple planning case is **8 engagements × $12,500 = $100,000**, plus **4 customers × $1,500/month × 6 billed months = $36,000**, for **$136,000 of gross revenue**. Every input is an author-selected assumption within the proposed pricing experiments. [Estimate methodology](#estimate-methodology)

This is not profit, cash collected, recurring software revenue, or evidence of demand. Engineering review, acquisition effort, rework, professional costs, taxes, and founder compensation still need to be paid. If most revenue requires bespoke expert labor, treat Modgrid as a specialist services business until the facts support a product company.

There is no defensible bottom-up market-size estimate yet. The missing facts are the number of reachable funded buyers, purchase frequency, accepted price, delivery effort, and retention. Multiplying all forecast data-center MW by an invented software fee would hide those unknowns.

## 7. Concrete next steps and stop conditions

These are proposed future actions. This research session did not contact anyone, enroll in any program, make commitments, or operate equipment. All counts and schedules below are chosen validation targets. [Planning methodology](#estimate-methodology)

### Days 1–30: make a saleable decision package

1. **Choose one utility workflow.** Use AEP Ohio's current tariff, application requirements, public contracts/process descriptions, and PJM dependencies. Record source dates and effective dates separately.
2. **Create a public-data sample.** Clearly label any fictional load profile as synthetic. Show how a slow load ramp, minimum charges, a generator outage, and uncertain delivery dates change the recommendation. Never imply the sample has a utility-approved service date.
3. **Identify professional review coverage.** Scope a limited paid review with an electrical engineer experienced in large-load interconnection and tariff counsel where needed. The first deliverable should state exactly what is and is not technically established.
4. **Design the research interviews for a later authorized phase.** Proposed target: **12 conversations**, split among developers, engineering firms, and operators/aggregators. Ask about the most recent delayed decision, the artifacts used, who could authorize spending, and what they actually paid for. [Target assumption](#estimate-methodology)
5. **Define a paid offer.** Fixed scope, named inputs, review process, delivery conditions, and no guaranteed power date. Use the proposed A pricing range as a test, not an advertised discount from a fictitious market rate.

**Gate:** Proceed only if there is a named buyer with a current decision, usable data, and a willingness to pay. Interest in a newsletter or an impressive AI demonstration is not enough.

### Days 31–90: prove repeatability

Target **3 paid decision packages**, with at least **1 repeat or extension engagement** as the strongest early signal. These are go/no-go targets, not forecast sales. [Target assumption](#estimate-methodology)

Track time spent on data collection, calculation, specialist review, and rework. Record which findings actually change the customer's decision. Build the smallest tool that removes repeated work: a versioned tariff model, evidence table, scenario comparison, and report generator.

**Stop or narrow the offer if:** Buyers will not pay without a guaranteed energization date; every project requires unrelated engineering; or customers repeatedly lack site control, funding, and a credible tenant.

### Months 4–6: choose the recurring workflow

If tariff updates and project evidence recur, extend A into portfolio monitoring. If an operating customer has telemetry, controllable assets, and a utility or aggregator ready to use the evidence, test B. If the demand is mostly cross-vendor troubleshooting, test E instead.

Do not pursue all three at once. Obtain a paid engagement for the selected recurring need, define the accepted output, and confirm data rights before building integrations.

### Months 7–12: make one expansion decision

Choose among:

- **A focused services business:** Repeatable, profitable readiness work with limited software automation.
- **A software product:** A recurring evidence or fleet workflow used across customers without major custom work.
- **A development partnership:** Repeat customer demand for a standard energy package, delivered and financed by experienced partners.

The proposed minimum evidence for moving into operating software is **one customer-approved response or performance test** and **one paying counterparty that accepts its output**. For developing energy projects, require a qualified buyer, finance path, site rights, permits strategy, and real equipment/fuel quotations before taking irreversible commitments. [Proposed validation threshold](#estimate-methodology)

### The questions that could overturn the recommendation

- Will a developer pay independently for readiness analysis, or is it already included in its engineer's scope?
- Which utility territories have the most reachable, funded buyers for Preetham, rather than merely the largest forecasts?
- Can the customer lawfully and practically share its tariff, contracts, interval data, and equipment telemetry?
- Who can authorize load changes, and who bears the penalty if they fail?
- Will an aggregator or utility accept Modgrid's measurement method without its own replacement analysis?
- Does an operating fleet have a repeated unmet need that makes E faster to sell than A?
- Can local commercial demand response relieve the same constraint as the data-center growth, making D more useful than direct data-center software?

## 8. Proposed modgrid.co homepage copy

The following is draft positioning for an initial readiness offering. Publish claims only when the corresponding service and review process exist. No customer logos, operating MW, savings percentages, or delivery guarantees are implied.

**Eyebrow**

Power planning for data centers and microgrids

**Headline**

Find a credible path to power.

**Subheadline**

Modgrid helps infrastructure developers compare utility service, on-site generation, storage, and flexible demand. Understand the commitments, dependencies, and unanswered questions before you commit capital.

**Primary action**

Discuss a power-readiness review

**Secondary action**

View a sample brief

**What you get**

- **Compare your options.** Evaluate power strategies against your site's load, schedule, and operating needs.
- **Understand the commitments.** Make minimum bills, collateral, ramp assumptions, and service conditions visible.
- **Know what needs proving.** Keep utility requirements, equipment availability, permits, and technical review in one decision record.

**Who it is for**

Independent data-center developers, engineering teams, and infrastructure investors evaluating a new site or expansion.

**How we work**

We start with your project documents and load plan. We organize the evidence, compare scenarios, and work with qualified specialists where technical or regulatory review is needed. You receive a practical brief for your next investment or project decision.

**Longer-term mission**

Help more computing run on dependable power while making better use of the grid.

**Editorial note:** Do not lead with “AI-powered,” “instant grid capacity,” “zero-carbon gas,” or an unqualified promise of faster energization. The initial claim should be a better, documented decision. Add performance claims only after measured customer results support them.

## 9. Evidence quality and remaining uncertainty

### Sources most important to revisit before launch

| Source | Evidence type | Refresh trigger |
|---|---|---|
| [LBNL electricity report][lbnl-energy], [EPRI][epri-forecast], [IEA][iea], [BNEF][bnef] | Modeled demand and scenario assumptions | New forecast editions; changes in server deployment and utilization |
| [LBNL queue dataset][queues] | Compiled historical project data | New annual release; regional process changes |
| [PJM load forecast][pjm-load] and [ERCOT preliminary outlook][ercot-forecast] | Operator planning, with different treatments of requested load | Annual revisions, batch results, and updated project vetting |
| [FERC status page][ferc-status], [PJM orders][pjm-orders], [large-load implementation page][pjm-large-load] | Regulatory process and implementation status | Orders, compliance filings, effective tariff sheets, and extensions |
| [Ohio tariff process][aep-tariff] and [Virginia SCC][virginia] | Utility-specific obligations | New tariffs, rate orders, and executed service agreements |
| [GE results][ge-backlog], [Wood Mackenzie][woodmac], [Rystad][rystad] | Manufacturer backlog and survey evidence | Actual supplier quotations for the proposed equipment |
| [Google flexibility][google-flex] and [Voltus partnership][google-vpp] | Buyer-reported contracts | Actual deployments, program terms, and demonstrated performance |
| [Fervo operating milestone][fervo-cod] and [NRC construction permit][nrc-natrium] | Company-reported operation versus regulator-issued construction authorization | Operating results and subsequent licenses |

### Limits of what this research establishes

- It establishes demand pressure, substantial commercial activity, and changing rules. It does not establish willingness to buy Modgrid's proposed service.
- It does not verify every supplier's operating MW, every project's financing close, or every announced delivery date. The report keeps those statuses separate.
- Public grid data cannot establish site-specific firm power availability. Some decisive network models, queue details, contracts, and commercial quotations are not public.
- Corporate funding, project financing, framework ceilings, procurement intentions, and operating revenue are not interchangeable. No values in this report should be summed into a total investment market.
- The cited equipment surveys have dates and category limitations. A live quotation remains necessary for project scheduling.
- The policy review is current to the stated cutoff, but several regional proceedings remain open. The recommended business should track changes as part of the product.
- Proposed company economics remain assumptions until paid work, delivery costs, and repeat purchases are observed.

**Decision:** Start with A, preserve B as the most attractive software expansion, and keep D as the clearest route to relieving grid strain through partner-financed physical projects. Make the first sale a documented project decision. Earn the right to promise an operating outcome later.

## Source links

Inline citations identify the evidence for numerical claims. The following reference definitions preserve direct links for the report and its charts. All sources were accessed during the November 4, 2026 research session. Research dates refer to the source publication or announcement where given, rather than a search engine's crawl date.

[lbnl-energy]: https://www.energy.gov/documents/united-states-data-center-energy-usage-report-2025-update "LBNL, U.S. Data Center Energy Usage Report: 2025 Update, published June 18, 2026"
[epri-forecast]: https://powering-intelligence.epri.com/annual-peak-use.html "EPRI, Powering Intelligence 2026, annual and peak use"
[epri-faq]: https://restservice.epri.com/publicattachment/96784 "EPRI, Powering Intelligence 2026 FAQ, February 2026"
[iea]: https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary "IEA, Key Questions on Energy and AI, 2026"
[bnef]: https://about.bnef.com/insights/data-centers/six-things-to-know-about-bnefs-new-us-data-center-capacity-outlook/ "BloombergNEF public outlook summary, July 29, 2026"
[pjm-load]: https://insidelines.pjm.com/pjms-updated-20-year-forecast-continues-to-see-significant-long-term-load-growth/ "PJM, January 14, 2026 load forecast summary"
[ercot-forecast]: https://www.ercot.com/news/release/04152026-ercot-releases-preliminary "ERCOT, April 15, 2026 preliminary load forecast"
[aep-contracts]: https://www.aepohio.com/company/news/view?releaseID=10753 "AEP Ohio, February 2026 contracted-load update"
[queues]: https://emp.lbl.gov/queues "LBNL and GridTracker, Queued Up 2026 edition, data through 2025"
[queues-scope]: https://eta.lbl.gov/publications/queued-2026-edition-characteristics "LBNL, scope of the 2026 interconnection dataset"
[aep-tariff]: https://www.aepohio.com/company/about/rates/data-center-tariff/ "AEP Ohio current data-center tariff and application explanation"
[ge-backlog]: https://www.gevernova.com/news/ge-vernova-reports-strong-third-quarter-2026-financial-performance "GE Vernova Q3 results, October 28, 2026"
[woodmac]: https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/ "Wood Mackenzie, October 2025 discussion of Q2 2025 equipment survey"
[rystad]: https://www.rystadenergy.com/insights/grid-equipment-market-outlook "Rystad, Grid Equipment Market Outlook, May 28, 2026"
[ge-case]: https://www.gevernova.com/gas-power/resources/case-studies/crusoe-ai-data-centers-lm2500xpress "GE Vernova Crusoe case study"
[pjm-auction]: https://insidelines.pjm.com/pjm-capacity-auction-procures-138318-mw-of-generation-resources-as-work-continues-to-address-growing-electricity-demand/ "PJM 2028/2029 auction results, July 14, 2026"
[pjm-prices]: https://www.monitoringanalytics.com/reports/Market_Messages/Messages/2025-som-pjm-press-release.pdf "Monitoring Analytics, 2025 State of the Market findings, March 12, 2026"
[pjm-counterfactual]: https://www.monitoringanalytics.com/reports/PJM_State_of_the_Market/2026/2026q1-som-pjm.pdf "Monitoring Analytics, first-quarter 2026 report"
[ferc-2024]: https://www.ferc.gov/sites/default/files/2024-11/20241101-3061_er24-2172-000.pdf "FERC Susquehanna amendment order, November 1, 2024"
[ferc-2025]: https://www.ferc.gov/news-events/news/fact-sheet-ferc-directs-nations-largest-grid-operator-create-new-rules-embrace "FERC co-location fact sheet, December 18, 2025"
[pjm-orders]: https://www.pjm.com/library/filing-order.aspx "PJM regulatory filings and orders register"
[ferc-status]: https://www.ferc.gov/cost-transparency-customer-protection "FERC, current large-load show-cause process status"
[pjm-large-load]: https://www.pjm.com/markets-and-operations/large-load "PJM large-load implementation and pending-approval notices"
[talen]: https://ir.talenenergy.com/node/9206/html "Talen 2025 annual filing, filed in 2026, AWS PPA discussion"
[virginia]: https://www.scc.virginia.gov/media/sccvirginiagov-home/about-the-scc/fact-sheets/scc-data-center-initiatives-02-2026.pdf "Virginia SCC data-center initiatives, February 24, 2026"
[texas-sb6]: https://capitol.texas.gov/tlodocs/89R/billtext/html/SB00006F.htm "Texas SB 6, enrolled text, 2025"
[ercot-batch]: https://www.ercot.com/news/release/06182026-puct-approves-ercots "ERCOT, PUCT approval of Batch Zero, June 18, 2026"
[ercot-rule]: https://www.ercot.com/mktrules/issues/PGRR145 "ERCOT PGRR145 approval and effective dates"
[oregon]: https://www.oregon.gov/puc/Documents/2025-Legislative-Report.pdf "Oregon PUC, 2025 legislative report and POWER Act"
[xcel]: https://newsroom.xcelenergy.com/news/xcel-energy-to-power-new-google-data-center-in-minnesota "Xcel resource agreement announcement, February 24, 2026"
[epa-data]: https://www.epa.gov/stationary-sources-air-pollution/clean-air-act-resources-data-centers "EPA, current Clean Air Act resources for data centers"
[epa-engines]: https://www.epa.gov/stationary-engines/understanding-stationary-engines-rules "EPA, stationary engine operating rules"
[irs-timing]: https://www.irs.gov/instructions/i7211 "IRS, Form 7211 instructions and wind/solar timing rules"
[irs-pfe]: https://www.irs.gov/newsroom/treasury-irs-provide-guidance-for-certain-energy-tax-credits-regarding-material-assistance-provided-by-prohibited-foreign-entities-under-the-one-big-beautiful-bill "IRS, 2026 material-assistance guidance"
[nerc]: https://www.nerc.com/globalassets/our-work/webinars/large-loads-technical-conference-presenations-public.pdf "NERC large-load technical conference, March 6, 2026"
[voltagrid]: https://voltagrid.com/voltagrid-collaborates-with-oracle-to-power-next-gen-ai-data-centers "VoltaGrid and Oracle, October 15, 2025"
[ge-crusoe]: https://www.gevernova.com/news/press-releases/ge-vernova-crusoe-announce-major-29-unit-aeroderivative-gas-turbine-deliver-ai-data-centers "GE Vernova and Crusoe equipment orders, July 22, 2025"
[bloom-oracle]: https://www.bloomenergy.com/news/bloom-energy-and-oracle-expand-strategic-partnership-to-deploy-up-to-2.8-gw-to-accelerate-ai-infrastructure-build-out/ "Bloom and Oracle expanded agreement, April 13, 2026"
[bloom-brookfield]: https://www.bloomenergy.com/news/brookfield-and-bloom-energy-expand-ai-infrastructure-partnership/ "Bloom and Brookfield expanded financing framework, June 30, 2026"
[google-flex]: https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/demand-response-data-center-milestone/ "Google contracted data-center demand response, March 19, 2026"
[google-vpp]: https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/virtual-power-plant-1-gigawatt/ "Google and Voltus expanded VPP agreement, October 22, 2026"
[emerald]: https://www.emeraldai.co/blog/emerald-ai-raises-150-million-series-a "Emerald AI Series A, August 25, 2026"
[form-funding]: https://formenergy.com/form-energy-secures-750m-in-series-g-financing/ "Form Energy Series G, August 12, 2026"
[fervo-cod]: https://fervoenergy.com/fervo-energy-declares-commercial-operation-at-cape-station-ahead-of-schedule-leading-the-race-for-next-generation-geothermal-energy/ "Fervo Cape Station commercial operation announcement, October 1, 2026"
[fervo-finance]: https://fervoenergy.com/fervo-energy-reports-first-quarter-2026-results/ "Fervo first-quarter 2026 results, project finance and Google framework"
[scale-eqt]: https://eqtgroup.com/news/eqt-to-acquire-distributed-energy-company-scale-microgrids "EQT acquisition announcement, January 9, 2025"
[scale-capital]: https://www.scalemicrogrids.com/capital-solutions "Scale Microgrids, capital solutions and June 2025 financing"
[linevision]: https://www.linevisioninc.com/news/linevision-secures-capital-to-accelerate-global-deployment "LineVision Series D announcement, September 17, 2026"
[heimdall]: https://www.heimdallpower.com/news/heimdall-power-and-entergy-deploy-multi-state-dynamic-line-rating-dlr-project "Heimdall and Entergy deployment, February 4, 2026"
[gridcare]: https://www.gridcare.ai/post/silicon-valley-ai-and-energy-pioneers-launch-gridcare "GridCARE launch and seed financing, May 27, 2025"
[crusoe-funding]: https://www.crusoe.ai/resources/newsroom/crusoe-announces-series-f-funding "Crusoe initial Series F closing announcement, September 17, 2026"
[nrc-natrium]: https://www.nrc.gov/node/2156776 "NRC Kemmerer construction permit, March 9, 2026"
[kairos]: https://www.kairospower.com/google "Kairos, Google fleet agreement and Hermes 2 target"
[kairos-construction]: https://www.kairospower.com/updates/the-time-is-now "Kairos construction update, May 28, 2026"
[xenergy]: https://www.sec.gov/Archives/edgar/data/2088896/000119312526347752/xe-20260630.htm "X-energy, quarter ended June 30, 2026 filing"
[meta-nuclear]: https://about.fb.com/news/2026/01/meta-nuclear-energy-projects-power-american-ai-leadership/ "Meta nuclear agreements, January 9, 2026"
[oklo]: https://oklo.com/newsroom/oklo-meta-announce-agreement-in-support-of-1-2-gw-nuclear-energy-development-in-southern-ohio "Oklo and Meta development agreement, January 2026"
[duke]: https://nicholasinstitute.duke.edu/sites/default/files/publications/rethinking-load-growth.pdf "Duke, Rethinking Load Growth, February 2025"
[paces-services]: https://www.paces.com/products/services "Paces development services"
[paces-update]: https://www.paces.com/post/paces-introduces-transmission-line-reports-and-injection-studies-for-faster-more-informed-site-decisions "Paces product update, September 22, 2026"
[schneider]: https://www.se.com/us/en/product-subcategory/88993-microgrid-management/ "Schneider Electric microgrid product portfolio"
[xendee]: https://xendee.com/operate "Xendee OPERATE product and integration model"
[tesla]: https://ir.tesla.com/_flysystem/s3/sec/000162828026003952/tsla-20251231-gen.pdf "Tesla 2025 annual filing, published 2026, storage products and software"
[enchanted]: https://enchantedrock.com/wp-content/uploads/White-Paper-Microgrids-for-Military-External-Use-Oct-2024.pdf "Enchanted Rock, microgrid ownership and service options"
[enel]: https://www.enelnorthamerica.com/content/dam/enel-na/resources/flyers/pdf/CenterPoint-Indiana-FAQ_Flyer_MISO.pdf "Enel and CenterPoint commercial demand-response program"
[crusoe-spark]: https://www.crusoe.ai/resources/newsroom/crusoe-introduces-crusoe-spark-liquid "Crusoe Spark Liquid announcement, October 21, 2026"
[pjm-get]: https://www.pjm.com/markets-and-operations/grid-optimization-solutions "PJM grid-enhancing technology overview and technical references"
[ferc-october]: https://www.pjm.com/pjmfiles/directory/etariff/FercOrders/9207/20261009-er26-3515-000.pdf "FERC order on large-load registry and IRAS, October 9, 2026"