AI Company Cap Table Modeling and Voting Power Analysis: A Practical Guide

1. Why a Shareholder List Cannot Be Used as a Calculation Input

After a leading AI company closes a new funding round, media outlets, research firms, and corporate data aggregators typically publish a cap table summary: shareholder names, ownership percentages, cumulative invested amounts, and sometimes voting arrangements. This format is useful for getting a quick read on the landscape. But if the list is fed directly into a calculation, three systematic biases usually appear.

The first is mixed valuation bases. Pre-money valuation, post-money valuation, enterprise value, and equity value are not the same concept. If the numerator and denominator of “cumulative invested amount ÷ valuation” are on different bases, the resulting percentage can be off by several points.

The second is an undefined denominator. Whether the employee option pool counts in the numerator or the denominator, whether only granted options or also reserved-but-ungranted ones are included, and whether convertible notes are treated as already converted — each choice produces a different result.

The third is conflating rights with economics. An ownership percentage describes “how much you take”; a voting percentage decides “how much you control.” Under a dual-class structure, the two can diverge sharply.

The more reliable approach is therefore not “look up a table” but to treat financing as a traceable event timeline and build a consistent set of definitions, a data model, and a reproducible calculation pipeline.

For funding information on leading AI companies, the following compliant sources are recommended, and each should be archived:

1. Public disclosure documents: exchange announcements, prospectuses, annual reports, and audited financial statements.
2. Company funding announcements, press releases, and investor relations pages on the official website.
3. Corporate registry and business filing disclosures from the relevant regulatory authority.
4. Reputable corporate and equity databases, used to cross-verify shareholders, capital contributions, and look-through ownership chains.
5. Media coverage and industry databases, with the retrieval timestamp, source link, and the basis used in the original text recorded for each entry.

Any dataset compiled from public information should be labeled as an unaudited basis. Undisclosed deal terms should not be inferred. Personal shareholder details should be redacted as needed. When republishing sensitive information about a real company, preserve source boundaries and avoid dressing up unverified information as a conclusion.

2. Four Core Bases That Must Be Defined Before Modeling

2.1 Valuation basis

Basis	Definition	Typical use	Common error
Pre-money	Company value before the investment is injected	Denominator for a new investor’s stake	Cited as if it were “market cap”
Post-money	Pre-money valuation plus the amount raised	Disclosed round valuation	Sources differ; easily mixed up
Enterprise value	Equity value plus net debt or other debt-like items	Cross-company comparison	Treated as interchangeable with equity value
Fully diluted	Total shares assuming all preferred stock has converted and all options and convertible instruments have been exercised	True ownership and dilution analysis	Ignores reserved options and unconverted notes

A practical rule: the numerator and denominator in a single calculation must come from the same basis. For example, when deriving a new investor’s percentage as “amount raised ÷ post-money valuation,” the denominator should be fully diluted wherever possible, otherwise the new investor’s stake is understated.

Fully diluted basis is easier to grasp with numbers. Suppose there are 1,000,000 common shares, 100,000 granted options, 50,000 reserved options, and convertible notes that would convert into 50,000 shares. Fully diluted share count is 1,200,000. Calculating against 1,000,000 alone understates future dilution. Whether instruments such as SAFEs and convertible notes should be included depends on their terms: if they can convert in the future and dilute common stock, they belong in the sensitivity analysis.

2.2 Who bears the option pool

The option pool serves two functions in a financing: retaining key talent through incentives, and making explicit who absorbs the dilution. Two arrangements are common in practice.

• Borne by pre-money shareholders. This is the most common structure. Before new investors enter, the pool is topped up to an agreed percentage, typically 10% to 20% of total shares. The dilution from the top-up is shared by existing shareholders pro rata, and investors then buy in at the expanded share count.
• Borne by all shareholders pro rata. The top-up applies equally to every shareholder, including the incoming investor. This is less common, because a new investor is generally unwilling to pay for an incentive plan from which it has not yet benefited.

If a model ignores that the top-up occurs before the investment, it overstates the founding team’s ownership percentage.

2.3 Separating voting rights from economic rights

A cap table should be split into at least two dimensions:

• Economic rights: determine distributions on dividends and in a liquidation.
• Voting rights: determine decision-making influence at the shareholder meeting and board level.

Two categories of instruments drive the two apart:

• Dual-class shares: specific share classes held by the founding team typically carry more votes per share, for example one vote per common share and ten votes per special-class share. The founding team can therefore hold a majority of voting power with a relatively modest economic stake.
• Voting agreements, proxy arrangements, and protective provisions: the first combine the votes of multiple parties for joint exercise; the second give specific shareholders a veto over a small set of material matters. None of these change ownership percentages, yet all directly affect the control assessment, so they should be modeled separately as “rights terms” rather than stuffed into the ownership table.

2.4 Direct and indirect holdings

Founders frequently hold equity indirectly through vehicles such as limited partnerships or holding companies. Employee share plans typically appear on the register as a single entity, with individual employees as partners of that vehicle. Where the structure is nested across several layers, a look-through calculation is required along the chain:

Indirect ownership = product of the ownership percentages at each layer.

For example, if a founder holds a 60% interest in a holding vehicle that owns 5% of the company, the founder’s indirect interest is 60% × 5% = 3%. Without the look-through, the model both misidentifies the vehicle’s actual beneficial owner and double-counts in any aggregation.

A quick decision checklist:

• If the material says “pre-financing valuation,” it is usually pre-money; if it says “post-financing valuation,” it is usually post-money.
• If the goal is control, dilution, or liquidation analysis, prefer the fully diluted basis.
• If the goal is shareholder voting, board control, or veto analysis, calculate voting rights separately.
• If holding vehicles, fund nesting, nominee holdings, or multi-layer ownership structures exist, perform a look-through calculation.

3. Data Model Design: An Event-Driven Dual-Timeline Model

3.1 Five modeling principles

1. Events are the source of truth. Record financings, option grants, note conversions, secondary transfers, and share buybacks as discrete events. Ownership percentages are a derived result of accumulated events, not a hand-maintained “current state” value.
2. Shares are the unit of account. A percentage is the result of a division and is only valid at a single point in time. Across rounds, everything must be normalized to share counts before it can be correctly aggregated, diluted, or looked through.
3. Economic and voting rights are calculated separately. Shares are stored by class, each class is configured with a votes-per-share weight, and the two are aggregated independently.
4. Terms are versioned. Anti-dilution, liquidation preference, and veto provisions can change with amendments, so they must be stored with effective date ranges rather than overwritten.
5. Use two timelines. Distinguish “business occurrence time” from “data record time.” This both reconstructs the cap table at any historical point and handles later corrections to public information.

3.2 Illustrative core table structure

ENTITY holder (
  holder_id,
  holder_type,
  parent_id
)
ENTITY holder_company (
  holder_id,
  company_id,
  relation,
  role_group
)
ENTITY round (
  company_id,
  round_id,
  close_date,
  pre_money_valuation,
  amount_raised,
  price_per_share,
  fully_diluted_flag
)
ENTITY equity_event (
  event_id,
  company_id,
  event_type,
  instrument,
  holder_id,
  share_class,
  share_delta,
  event_date,
  record_date,
  source_note
)
ENTITY term (
  company_id,
  term_type,
  effective_from,
  effective_to,
  terms_json
)

The equity_event table should be append-only. The term table should be versioned by effective date range. The valuation fields in the round table should be explicitly flagged for whether they are on a fully diluted basis.

3.3 Reconstructing share counts from a round sequence

If each round discloses a price per share, new shares can simply be summed. If only amounts and percentages are available, an identity is needed.

The core relationship is:

New shares ÷ post-money total shares = amount raised ÷ post-money valuation.

Here the denominator is the post-money valuation, which is also the total value corresponding to the post-money share count. Let pre-round total shares be P0, the amount raised be R, the post-money valuation be Vpost, and new shares be x. Then:

x / (P0 + x) = R / Vpost.

Solving:

x = P0 × R / (Vpost − R).

Since post-money valuation equals pre-money valuation plus the amount raised, Vpost = Vpre + R, this can also be written as:

x = P0 × R / Vpre.

If the post-money total share count Ppost is already known, the calculation is more direct:

x = Ppost × R / Vpost.

This form embeds an important assumption: the pre-money valuation must be on a fully diluted basis, with the option pool and all issued convertible instruments included in the denominator. Otherwise the reconstructed new-investor percentage will be overstated.

4. Key Algorithms

4.1 Allocating option pool top-up dilution

Before closing round i, suppose total shares are S0, the pool already holds P0 shares, and the target percentage is t, for example 15%. The additional pool allocation x satisfies:

(P0 + x) / (S0 + x) = t.

Solving:

x = (t × S0 − P0) / (1 − t).

The new allocation is normally borne by pre-money shareholders pro rata. For example, a shareholder who previously held 30% becomes 30% × S0 / (S0 + x) after the top-up. Once the top-up is complete, the investor’s stake is calculated against the new share count.

4.2 Dilution between rounds

A new financing dilutes every existing shareholder by the same factor, which is equivalent to:

Post-dilution percentage = original percentage × (pre-money shares / post-money shares).

Consequently, analyzing “how much of the founding team remains after several consecutive rounds” requires multiplying all dilution events in chronological order, not subtracting them.

4.3 Voting concentration under dual-class shares

Input:
holders: positions already looked through and consolidated to the
         ultimate beneficial owner
class_votes: mapping from share class to votes per share
total_votes = 0
for h in holders:
    h.votes = sum(h.shares[c] * class_votes[c] for c in h.shares)
    total_votes += h.votes
for h in holders:
    h.voting_power = h.votes / total_votes
hhi = sum(h.voting_power ^ 2 for h in holders)

In antitrust analysis, an HHI below 0.15 is generally treated as unconcentrated and above 0.25 as highly concentrated. When borrowed as a shareholder voting concentration indicator, it should serve only as a reference metric and should not be equated with a legal conclusion. class_votes should be a configurable parameter that reflects different voting arrangements, rather than a rule hard-coded into the logic.

4.4 Anti-dilution adjustment pseudocode

Anti-dilution provisions are typically triggered when a new round is priced below the prior conversion price. The model must first filter provisions by effective date range, then compute the adjusted conversion price under either a full ratchet or a weighted-average method.

Input:
terms: list of anti-dilution provisions
target_date: closing date of the round being modeled
old_conversion_price: original preferred conversion price
new_round_price: price per share in the new round
existing_common_shares: common or as-converted common shares
                        before closing
raised_amount: amount raised in this round
option_pool_flag: whether the option pool is included in the
                  weighted-average calculation
Steps:
1. Filter provisions:
   keep only anti-dilution provisions where
   effective_from <= target_date
   and (effective_to is null or target_date < effective_to).
2. Test the trigger:
   if new_round_price >= old_conversion_price, no adjustment.
   if new_round_price < old_conversion_price, apply adjustment.
3. Full ratchet:
   adjusted_price = new_round_price
   adjusted_conversion_ratio =
       old_conversion_ratio * (old_conversion_price / adjusted_price)
   Meaning: the conversion price is reset to the lowest price,
   prior investors' conversion share count increases, and common
   shareholders are materially diluted.
4. Weighted average:
   A = existing_common_shares
   B = raised_amount / old_conversion_price
   C = raised_amount / new_round_price
   adjusted_price =
       old_conversion_price * (A + B) / (A + C)
   adjusted_conversion_ratio =
       old_conversion_ratio * (old_conversion_price / adjusted_price)
   Under a narrow-based weighted average, A, B, and C may exclude
   the option pool and convertible instruments; under a broad-based
   weighted average, they should include them. This choice must be
   explicitly configured in the terms field.
5. Output:
   write the adjusted conversion price, conversion ratio, trigger
   reason, effective date, and source provision to the event table,
   retaining the historical version.

The point of anti-dilution analysis is not memorizing a formula but answering three questions: which shares participate in the calculation, which round’s price triggers the adjustment, and whether the adjustment takes effect before the target date.

5. Worked Example: Three Rounds for an AI Company

All figures below are hypothetical and used only to demonstrate the algorithms. Assume company H brings in a founding team, institutional investors, and a strategic investor across three stages.

Round	Pre-money	Amount raised	Post-money	Key arrangement
Seed	None	None	RMB 50M	Founding team 80%, angel investor 10%, option pool 10%
Series A	RMB 150M	RMB 50M	RMB 200M	Financial investor holds 25%
Series B	RMB 400M	RMB 100M	RMB 500M	Strategic investor holds 20%; pool topped up from 7.5% to 12% before closing

5.1 Reconstructing share counts

Using 10,000,000 seed shares as the base:

Series A new shares = 10 × 50 / 150 = 3.3333 million shares.

Post-money total = 10 + 3.3333 = 13.3333 million shares.

Check: 3.3333 / 13.3333 = 25%, consistent with “amount raised ÷ post-money valuation.”

5.2 Option pool top-up

The pool percentage after Series A is:

1 / 13.3333 = 7.5%.

If the pool must reach 12% before the Series B closing, let the additional allocation be x:

(1 + x) / (13.3333 + x) = 12%.

Solving:

x = (0.12 × 13.3333 − 1) / (1 − 0.12) = 0.6818 million shares.

Total shares after the top-up are 14.0152 million, and the pool holds 1.6818 million.

The dilution borne by existing shareholders is:

1 − 13.3333 / 14.0152 = 4.87%.

The founding team’s reduction is:

8 / 13.3333 − 8 / 14.0152 = 60% − 57.07% ≈ 2.92 percentage points.

5.3 Series B round

Series B new shares = 14.0152 × 100 / 400 = 3.5038 million shares.

Post-money total = 14.0152 + 3.5038 = 17.5190 million shares.

If the option pool carries no votes, the voting denominator includes only the founding team, the angel investor, the Series A investor, and the Series B investor.

Shareholder	Shares (M)	Economic ownership	Class and votes	Voting power
Founding team	8.0000	45.67%	High-vote class, 10 votes per share	91.08%
Angel investor	1.0000	5.71%	Common, 1 vote per share	1.14%
Series A financial investor	3.3333	19.03%	Common, 1 vote per share	3.79%
Series B strategic investor	3.5038	20.00%	Common, 1 vote per share	3.99%
Employee option pool	1.6818	9.60%	No votes	None

Economic percentages in the table are rounded; totals should be based on unrounded share counts.

Founding team voting power:

8,000 / (8,000 + 783.71) = 91.08%.

Here 783.71 = 100 + 333.33 + 350.38, the total common-share vote count, excluding the employee option pool.

The founding team’s economic ownership is 8 / 17.5190 = 45.67% while its voting power is 91.08%, a gap of roughly 45.41 percentage points. Reading ownership percentages alone therefore misjudges control substantially.

A further assessment of “control stability” should also bring in three kinds of information:

• Board seat composition: board voting is usually independent of shareholder voting.
• Protective provisions: whether investors hold a veto over material matters.
• Indirect influence from commercial alignment: if the strategic investor is also the provider of core compute and cloud services, its influence extends beyond its ownership percentage.

6. Common Pitfalls and Frequently Asked Questions

Common pitfalls include:

1. Mixing pre-money and post-money: both valuations alternate within one document with no basis labeled.
2. Ignoring the option pool: excluding reserved but ungranted allocations from the denominator systematically overstates existing shareholders’ percentages.
3. Mixing amount-based and share-based percentages: treating “cumulative invested amount ÷ current valuation” as an ownership percentage ignores that valuation changes across rounds and that convertible instruments exist.
4. Ignoring secondary transfers: a transfer changes the register but not total shares, and must be handled differently from a primary issuance.
5. Ignoring buybacks and option forfeitures: unexercised options from departing employees return to the pool, and total shares may decrease.
6. Treating economic rights as the basis for cash distributions: shares carrying liquidation preferences follow a distribution order in a liquidation event that differs from their ownership percentage.
7. Failing to look through multi-layer vehicles: treating a holding platform as a “real shareholder” distorts beneficial ownership analysis.

Common reader questions reduce to the following:

• How should a shareholder list be calculated? Recompute it from events and share counts; do not simply divide an amount by the current valuation.
• Who bears the option pool? Usually pre-money shareholders, but the definitive answer is the investment agreement.
• Why do voting percentage and ownership percentage differ? Dual-class shares, voting agreements, or protective provisions.
• How does anti-dilution affect existing shareholders? A full ratchet or weighted-average mechanism increases the conversion share count for investors in a down round, diluting common shareholders.

7. Data Quality and Validation

The largest risk in cap table analysis lies in the data, not the algorithms. Three validation layers are recommended.

7.1 Conservation checks

At any given point in time, the sum of all shareholder share counts equals total shares; the sum of votes across classes equals total votes; and the sum of looked-through indirect interests equals the upper-tier holding.

Example assertions:

assert sum(holder.shares for holder in holders_at(t)) == total_shares(t)
assert sum(holder.votes for holder in holders_at(t)) == total_votes(t)
assert sum(event.share_delta for event in events_before(t))
       == total_shares(t) - initial_shares

Look-through check:

for holder in holders:
    chain = get_chain(holder)
    indirect_equity = 1
    for link in chain:
        indirect_equity = indirect_equity * link.shareholding_ratio
    assert indirect_equity == expected_indirect_equity

7.2 Consistency checks

A new investor’s percentage should be derivable by two independent paths — “new shares ÷ post-money shares” and “amount raised ÷ post-money valuation” — and the two results should agree within tolerance.

ratio_from_shares = new_shares / post_shares
ratio_from_valuation = amount_raised / post_money_valuation
assert abs(ratio_from_shares - ratio_from_valuation) < tolerance

7.3 Plausibility checks

Set thresholds on metrics such as the founding team’s percentage, any single institution’s percentage, and the option pool percentage; a breach should trigger manual review. For example, if the founding team falls below 20%, a single institution exceeds 50%, or the option pool exceeds 25%, check for anomalous events or a basis error.

Recommended engineering practices:

• Replayable snapshots: assign each financing event a unique version number and source note. When data is corrected, append a new version so that historical conclusions remain reproducible.
• Full recomputation as a fallback: periodically recompute cap table snapshots from the event table in full and compare against incremental results, preventing accumulated drift in incremental logic.
• Explicit basis fields: attach a basis tag to every valuation, percentage, and share count field to prevent mixing at the source.
• Traceable sources: retain the link, retrieval time, archival method, and basis description for every piece of external data.

8. Conclusion

A large-model company’s cap table looks like a static list, but it is fundamentally a dynamic sequence driven by financing events. Analyzing this kind of data comes down to three points:

1. Fix the basis first, and align the definitions of valuation and denominator.
2. Reconstruct every dilution from events and share counts.
3. Calculate economic and voting rights separately.

Implementing these three points in a dual-timeline data model, anti-dilution logic, and automated validation is what moves an analysis from “a plausible-looking pie chart” to “a conclusion that survives follow-up questions.”