How the research is built and checked.

Valuation models in plain language

A valuation model is a structured estimate of what a business, project, or asset may be worth. It is not a promise about where a stock price will trade. The result depends on the assumptions used, the quality of the evidence, and whether the company can execute the plan being modeled.

What a discounted cash flow model does

A discounted cash flow model, usually called a DCF, estimates value in four broad steps:

1. Forecast the cash the business or project may generate in future years.
2. Subtract the spending required to operate, build, maintain, and replace the assets that produce that cash.
3. Reduce future cash to a present value because money expected later is worth less and is less certain than money available today.
4. Adjust for cash, debt, other financial claims, and the diluted share count when the model produces a per-share estimate.

The rate used to reduce future cash is commonly called the discount rate or weighted average cost of capital (WACC). A higher rate means investors require more compensation for time and risk, so the estimated value today falls.

A DCF normally includes a terminal value for cash generated after the detailed forecast period. Terminal value can be a large part of the total estimate, so the model should show how sensitive the result is to that assumption.

What makes a model custom

Custom does not mean the formulas are designed to produce a preferred answer. It means the structure is built around how the specific company or asset creates value and where it can fail.

For example:

• an operating data-center and Bitcoin-mining company may require separate forecasts for each site, customer contract, GPU group, construction phase, and hardware-replacement cycle;
• a development-stage data-center campus may be better understood one phase at a time, with explicit lease, power, construction, financing, ownership, and completion assumptions;
• a helium project may be limited by recoverable resources, sustained well flow, plant capacity, selling prices, installed cost, taxes, offtake, and future funding; and
• a stabilized infrastructure asset may be studied through net operating income and cap rates rather than a company-wide cash-flow forecast.

The model should therefore match the asset. Applying the same generic spreadsheet to every company can hide the risks that matter most.

How the models on this site differ

• Company and project DCF: forecasts cash from the operating company while modeling major sites or buildout groups separately. The result is usually an estimated value per share.
• Development-site valuation: estimates what individual project phases may be worth and gives less credit to capacity that is not yet leased, financed, permitted, or built. It may combine net asset value, cap-rate, and probability adjustments rather than use one conventional company DCF.
• Project DCF: forecasts the production, prices, costs, taxes, capital spending, financing, and dilution of a specific development project.
• Valuation teaching tool: demonstrates one valuation relationship, such as how cap rates and interest rates affect asset values. It is not a stock-price estimate unless it explicitly includes the company balance sheet and share count.

How to read a model page

Start with the main question and the base case. Then compare the bear and bull cases to identify what creates the downside and upside. Before focusing on the final number, review the low-confidence assumptions, the largest funding requirements, the diluted share count, and the sensitivity tables. A precise-looking output is not reliable when its key inputs remain uncertain.

Common terms used on the model pages include:

• Free cash flow: cash left after the spending required to operate and maintain the business or project.
• Enterprise value: the value of the operating business before subtracting net debt and other senior claims.
• Equity value: the value left for common shareholders after those balance-sheet adjustments.
• Diluted shares: the expected share count after options, share awards, convertible securities, and modeled financing.
• Net operating income (NOI): property or project income after direct operating expenses, but before financing and corporate costs.
• Cap rate: the annual return a buyer requires from stabilized NOI. A higher cap rate produces a lower asset value.
• Sensitivity table: a grid showing how the result changes when two important assumptions change.

Research process

Each project begins with a falsifiable question, a source map, and a list of facts that must be distinguished from judgment. The working sequence is: collect, verify, normalize, model, challenge, publish, revise, and archive.

A model should not move from a demonstration framework to a current publication merely because a spreadsheet exists. It must have a dated valuation reference point, source-backed assumptions, visible limitations, a version entry, and a workbook or machine-readable export that readers can inspect.

Publication-status taxonomy

Research and models use status language to avoid mixing education, current analysis, and stale work.

• Framework demonstration means the structure is useful, but values remain placeholders or incomplete examples.
• Current preliminary model means the model is tied to a live workbook and source map, but economically material assumptions still require additional evidence.
• Current reviewed model means material assumptions have direct source links, formulas have passed model checks, and unresolved inputs are explicitly disclosed.
• Requires update means new evidence may have changed the conclusion or inputs.
• Archived or superseded means the model remains available for history but should not be relied on as current.

A page can be primary-source supported while still preliminary. Primary-source support means some facts are anchored to filings or company documents; it does not mean every forecast or ownership assumption is verified.

Source hierarchy

1. Regulatory filings
2. Audited financial statements
3. Court records and government records
4. Company contracts and formal disclosures
5. Earnings calls and investor presentations
6. Counterparty disclosures
7. Reputable industry data
8. Journalism
9. Social-media claims and community research

Lower-quality sources can identify useful questions. They do not automatically become verified evidence.

Assumption-level source linking

Every material assumption should identify:

• the model cell, tab, or table where it appears;
• the strongest source that supports it;
• whether the source supports a fact, company guidance, benchmark, estimate, inference, or speculation;
• the date of the evidence;
• what remains unverified; and
• the confidence level assigned to the input.

A source can support the existence of a management plan without proving the economic outcome. For example, a company announcement can support a stated power pathway, while still leaving final ownership, fuel, permits, PPA terms, equipment delivery, and project economics unresolved.

Forecasting framework

Forecasts start at the physical and contractual level: capacity, timing, utilization, unit revenue, operating cost, capital cost, ownership, and financing. Company-level outputs are aggregated only after site and phase definitions are normalized.

For development-stage infrastructure, the forecast should separate:

• nameplate power;
• critical or usable IT load;
• contracted load;
• revenue-producing load;
• tenant-funded versus sponsor-funded capex;
• project debt versus parent equity;
• direct ownership, fee, and promote economics;
• power-side value versus data-center value; and
• diluted share count.

Valuation methodology

The project can use DCF, NAV, cap-rate, comparable-company, and probability-adjusted frameworks. The method must match the asset and must avoid double counting. Enterprise value is reconciled to common equity through cash, debt, leases, minority claims, non-operating assets, and diluted shares.

The required output for a current model is not only a per-share value. It should include the calculation chain: revenue, margins, capex, funding mix, debt draw, interest, equity contribution, ownership share, terminal or residual value, net-debt bridge, dilution, and model checks.

Treatment of management guidance

Management guidance is labelled as company guidance. It may be the best available input, but it remains distinct from an audited result, executed contract, or independent record.

Guidance can anchor a plan, capacity target, development milestone, or strategic relationship. It should not be upgraded to a verified cash-flow assumption unless the relevant contract, financing, permit, counterparty, or construction evidence supports that upgrade.

Probability adjustment

Probability adjustment captures whether a conditional outcome occurs. Discounting captures when cash flow occurs and the return required for bearing risk. The two should not be treated as substitutes.

A probability should move only when evidence changes a specific gate. A price move, market enthusiasm, or repeated management target does not by itself prove higher completion probability.

Project-stage risk adjustment

Development phases are evaluated through site control, power, permitting, customer, financing, and construction gates. Completion probabilities rise or fall only when evidence changes.

For large data-center and power projects, the following gates should be visible where material:

1. site and property control;
2. power, generation, grid, fuel, or interconnection rights;
3. environmental, emissions, zoning, utility, and construction permits;
4. tenant obligations, lease economics, pass-throughs, credit support, and remedies;
5. lender commitments, draw conditions, guarantees, collateral, reserves, and covenants; and
6. equipment procurement, notice to proceed, construction progress, energization, and revenue commencement.

Financing and dilution assumptions

Every model reconciles sources and uses. Debt availability, interest, fees, covenants, guarantees, equity requirements, and resulting share count are explicit.

Per-share estimates should not be shown without the denominator. A model must disclose whether the share count is basic, diluted, treasury-stock-method adjusted, pro forma for financing, or a scenario assumption.

Scenario construction

Bear, base, and bull scenarios must be internally consistent. A scenario cannot combine a delayed project with unchanged financing carry or a higher capex budget with an unchanged share count unless another funding source is identified.

When a project is development-stage, scenarios can also be organized by evidence state: current platform, signed first phase, and full buildout. The label must make clear whether the scenario is a probability-weighted state, a conditional signed-lease case, or a full-execution case.

Treatment of terminal value

Terminal value is separated from explicit forecast value and disclosed as a percentage of total value. Exit assumptions must be consistent with stabilized asset quality, maintenance capital, growth, and market duration.

If a model uses a cap-rate or terminal-multiple shortcut, it must disclose what is being capitalized, when stabilization occurs, and whether the value has already been captured elsewhere in the DCF.

Treatment of non-operating assets

Cash, investments, tax assets, land, equipment, and optional capacity are valued separately when they are not already included in operating cash flow.

Optional capacity should not receive full stabilized value unless the model separately supports power, customer, financing, construction, and ownership economics.

Model checks

Every production model should include checks appropriate to its structure. At minimum, checks should confirm that phases sum to the stated capacity, formulas do not silently break, scenario probabilities or weights are internally consistent, per-share values use the intended share count, and promote/upside cases do not appear below no-promote cases unless a clear economic reason is identified.

Model archives and revision logs

Before a material model change, copy the full workbook to an archive folder with a dated title. The live workbook should include a revision tab identifying the snapshot, date, change type, source basis, validation status, and website-sync status.

Historical model copies exist to preserve context, not to create multiple current conclusions. The live public page should identify which version is current and which versions are historical.

Error correction

Material errors receive a dated public change-log entry showing the previous value, revised value, reason, source, and expected effect. Corrections are not silently overwritten.

Version control

Research articles and models use semantic-style version identifiers. Git history preserves file-level changes; the public update log explains analytical changes in reader-facing language.

Conflicts of interest

Positions, compensation, personal relationships, and other material conflicts should be disclosed where relevant. Ephesus Research does not accept issuer- or company-specific paid coverage, and reader support does not purchase editorial control.

Limitations

Independent research can be incomplete, delayed, or wrong. Public filings can omit commercially sensitive information, estimates can fail, and models can create false precision. Readers should challenge assumptions and perform independent due diligence.

Evidence labels

Every major conclusion should distinguish among verified fact, plausible interpretation, unverified claim, model estimate, and speculative scenario. The interface uses six operational labels: fact, company guidance, analyst estimate, model assumption, inference, and speculation.