The most quoted number in Anthropic's pre-IPO governance announcement is not a number. It is a range.

Take the sentence everyone repeated: the company will commit 5% of its equity to a philanthropic vehicle. Now run that sentence against the three valuation anchors circulating around it. At the $144 billion mark published by New Constructs in October, 5% is $7.2 billion. At the $965 billion figure used in the announcement's own framing, 5% is $48.25 billion. At the $1.5 trillion discussed by some investors, 5% is $75 billion. Same clause. Same percentage. A 10.4x spread between the floor and the ceiling.

I have spent the last eight years reading commitments that can be checked. Transaction hashes. Block numbers. Vesting contracts with a cliff at month twelve and linear release after. When I audited the Aether token in 2017, the gap between the whitepaper's claimed whale activity and the actual mainnet logs was 40% — and that 40% was the entire investment thesis. Here there is no log to pull. The commitment exists as prose, and prose does not have a block number.

That is the finding. Not that the pledge is fake, and not that it is generous. The finding is that a promise worth between $7.2 billion and $75 billion has been published without the three terms that would let anyone price it: the source of the shares, the baseline for the escalator, and the rules for converting equity into money.

In a market where survival matters more than upside, the only promises that count are the enforceable ones. This one is not enforceable yet.

Context: how I read an equity pledge

Anthropic is a Delaware public benefit corporation. It carries a Long-Term Benefit Trust holding a special class of shares with the power to elect a portion of the board. In July it committed $200 million in cash to research on job displacement. Now, weeks before an initial public offering and subject to a shareholder vote, it has committed 5% of its equity to a philanthropic vehicle structured as a limited liability company wholly owned by the company, with an escalator that raises the percentage as valuation rises. The announcement also placed the commitment next to OpenAI's 26% nonprofit stake, inviting a direct comparison.

I read equity commitments the way I read token unlock schedules, because the underlying mechanics are identical. There are exactly three questions that matter. How much, in units that can be settled. Whose units, meaning who absorbs the cost. And when can the units be sold, meaning what hits the market and on what schedule.

Everything else — the mission language, the framing of responsibility, the comparison to a competitor — is presentation. Presentation is not worthless. It just is not priceable.

My 2020 Curve work taught me this in a way that stuck. I built SQL to track impermanent loss across more than 500 wallets during DeFi Summer and found that roughly 15% of the advertised yield was being extracted by bots exploiting front-running. The headline APY was real. The realized APY was different. The gap between them was not a marketing failure; it was a mechanical property of the pool design. You could only see it by querying the wallets.

The same discipline applies here. The headline is 5%. The realized number depends on mechanics nobody has published.

I want to be precise about what this announcement is and is not. It is a governance commitment made in the window immediately before a listing, in a structure the issuer controls, subject to approval by the issuer's own shareholders. It is not a filing, not a contract with an external counterparty, and not a disclosure document. Those distinctions are the entire basis for everything that follows, and anyone pricing the stock should hold them in mind before quoting the $48 billion figure at a dinner.

Core: what the numbers actually say

The arithmetic is internally consistent, which is exactly why it misleads.

96.5 billion times 5% is 48.25 billion, which rounds to the $48 billion everyone quoted. 1.5 trillion times 5% is 75 billion. 144 billion times 5% is 7.2 billion. Every one of those is correct arithmetic on a different input. The maximum-to-minimum ratio is 10.4.

A percentage is a ratio, and a ratio without a denominator is a rumor with a decimal point. The $48 billion is a snapshot taken at one valuation on one day. It is not the intrinsic value of the commitment, because the commitment has no intrinsic value independent of the price at which the shares eventually trade. Anyone quoting $48 billion as though it were a settled figure is quoting a photograph and calling it a contract.

For an IPO buyer this is not academic. You are being asked to price a share. If 5% of the equity is earmarked and the value of that earmark swings by a factor of 10.4 depending on which anchor you use, then your dilution estimate swings with it. The wide range is not a rounding error. It is the risk.

Whose shares, and therefore whose tax.

The announcement does not specify whether the 5% comes from newly issued shares or from existing holders transferring stock. That single omission determines who pays.

If the shares are newly issued, the cost is borne pro rata by everyone holding after the IPO, including the new buyer. The buyer is effectively taxed 5% at the moment of purchase, paid in dilution rather than cash. If the shares are transferred from pre-IPO holders, the cost falls only on founders, employees, and early investors, and the public buyer is untouched.

In token terms this is the difference between inflation and an unlock. Inflation dilutes every holder continuously. An unlock moves existing supply from a locked wallet to a liquid one and dilutes nobody — it reprices everyone through market impact. Conflating the two is the most common error I see in tokenomics reviews, and that same conflation sits inside this announcement, unaddressed.

The vehicle is a multisig with one key holder.

The philanthropic vehicle is an LLC wholly owned by Anthropic. Read that sentence the way an auditor reads it. There is no independent trustee. There is no mandatory annual distribution. There is no public disclosure obligation. The company is the sole member.

Compare that to what a private foundation would require. A 501(c)(3) foundation must distribute at least 5% of assets annually. It must file Form 990, which is public. It is barred from self-dealing transactions with its founders. It faces excise tax on excess business holdings — the rule designed precisely to prevent a foundation from becoming a controlling shareholder of an operating company. An LLC faces none of these constraints. Five exemptions, stacked.

I do not raise this to accuse anyone of bad faith. I raise it because an entity with no independent trustee, no mandatory payout, and no public filing has the governance profile of a multisig where the same person holds every key. That is not a comment on intent. It is a description of the enforcement surface, and the enforcement surface is close to zero.

The 2025 labeling project I led for a major asset manager is the closest professional analogue I have. We spent six months mapping more than 50,000 wallet addresses to regulatory-compliant entity labels and cut data ambiguity by roughly 90%, which is what allowed a $100 million institutional inflow to clear compliance review. The lesson was not that labels are paperwork. The lesson was that an unlabeled address is an unpriced risk, and the market prices what it can name. This fund is unlabeled in exactly that sense.

The flexibility is real, and it cuts both ways. An LLC can hold shares as collateral, invest them, vote them, or pledge them toward a future buyback. Every one of those options is a legitimate use of a flexible vehicle and every one of them is also a way for the commitment to be reshaped without the beneficiaries being consulted. Flexibility without disclosure is not a feature. It is an undisclosed variable.

The escalator compounds, and the baseline is the master dial.

The announcement includes an escalation clause: for every 50% increase above a baseline valuation, the commitment rises by two percentage points. Read against the numbers that circulated, the steps land at 7% around $1.45 trillion, 9% around $2.17 trillion, and 11% around $3.26 trillion.

Notice the shape. $1.45 trillion is 1.5 times $965 billion. $2.17 trillion is 1.5 times $1.45 trillion. $3.26 trillion is 1.5 times $2.17 trillion. The thresholds are geometric, not linear — each step compounds on the previous one rather than adding a fixed increment to the baseline. Almost nobody reading the headline will catch that, and it changes the schedule's character: the series has no terminal value, and the percentage creeps upward in perpetuity as long as the valuation keeps climbing.

Which brings the real variable into view. The baseline is not a technicality. It is the dial that determines whether the escalator ever fires. Set the baseline low and the commitment ratchets upward almost immediately. Set it high and it never triggers. The baseline has not been disclosed. And the party best positioned to know where the IPO will price — the issuer, working alongside its underwriters — is the party that will propose it.

That is a structural information asymmetry, and it sits at the center of the most-quoted governance commitment of the year.

The accounting substance is smaller than the headline, and that is the point.

Because the LLC is wholly owned, it will most likely remain consolidated. The shares move from one part of the group to another. No cash leaves. No expense is recognized. On a discounted cash flow model the effect is a smaller float rather than a smaller enterprise value.

In crypto I have watched this pattern play out dozens of times. A protocol treasury transfers tokens to a foundation wallet and publishes a release about ecosystem funding. Circulating supply does not change. Price does not change. Nothing changes until the tokens are sold, and then the impact is a function of the disposition schedule, not the announcement. The market impact of a commitment lives in its unlock schedule, and this commitment has no published unlock schedule.

The notional illusion: 240 to 1, and almost none of it spendable.

Set the equity pledge against the July cash commitment. $48 billion against $200 million is a ratio of 240 to 1. The equity commitment is nominally 240 times larger.

But the $200 million is cash. It can be spent, it is deductible, and the beneficiaries can be identified. The $48 billion is paper in an entity the company controls, with no independent trustee, no disposition rules, and no named beneficiaries. Nominal value is not settlement value.

I have run this comparison before. In 2021 I mapped the transfer history of 1,200 tokens in a collection called CryptoClones on OpenSea and found that 85% of secondary sales were occurring between wallets controlled by a single entity. The reported volume was enormous. The settled economic activity was a rounding error. When I published the graph, the floor fell 60% — not because I revealed anything hidden, but because I revealed the difference between notional and settled. A promise that cannot be enforced has a settlement value near zero until the terms are written, regardless of the number attached to it.

Three layers of special equity, and a discount nobody has priced.

Anthropic is a public benefit corporation. It has a Long-Term Benefit Trust holding special shares with board election rights. It will now have a philanthropic LLC holding 5% and potentially more. That is three layers of non-standard equity stacked on top of an ordinary share class.

The announcement does not mention the trust at all. That is the largest omission in the coverage I have read. Index providers apply governance screens, and multi-class structures with unusual voting arrangements routinely receive reduced weights or exclusion from passive benchmarks. For a company at this scale, passive inclusion is one of the few reliable sources of demand in a bear market. A governance discount is not an abstraction. It is a measurable reduction in the bid.

Who votes the 5%, and why both answers are bad.

If the company exercises the voting rights attached to the fund's shares, the company is voting its own stock — a self-dealing loop that governance screens exist to catch. If an independent trustee exercises them, a philanthropic entity becomes a 5% to 11% voting bloc in a public company, with board influence and no shareholder accountability. Neither answer has been published. Both are priceable. Neither is priced.

There is a third possibility that has gone entirely undiscussed. A wholly owned LLC holding 5% or more of the equity is a stable voting block available to management in any future control contest. That is an anti-takeover shield built from charitable language, and it may be the most consequential — and least acknowledged — function of the structure.

The 20% against 26% comparison is apples to oranges.

The announcement pairs Anthropic's roughly 20% against OpenAI's 26%. The percentages are not measuring the same thing. OpenAI's 26% is a nonprofit entity's retained stake from a 2025 restructuring — passive, historical, carrying specific mission constraints and a defined legal structure. Anthropic's figure blends a public benefit corporation's charter obligations, a trust's special shares, $200 million in cash, and this new 5% equity pledge with an escalator attached.

Adding those together and comparing the sum to a single number elsewhere is a category error dressed as competitive positioning. What is genuinely interesting is not the percentage. It is that the two leading labs have both decided that governance architecture is now a competitive dimension, and both decisions landed immediately before a major capital event. That is convergence under shared regulatory pressure, not independent moral reasoning.

The $144 billion number is a short thesis, not a consensus.

New Constructs, which produced the low anchor, runs a cash-flow-and-return-on-invested-capital framework and has a long record of systematic bearishness on high-multiple technology names. Their $144 billion is a legitimate extreme scenario. It is not a median expectation, and treating it as one overstates the downside in exactly the way the $1.5 trillion anchor overstates the upside.

The real anomaly is the number that is not there.

Here is what the announcement does not contain: annualized revenue, gross margin, customer count, API volume, enterprise concentration, or use of proceeds. A company discussed at a $965 billion valuation has published a governance commitment and no revenue line.

I have done pre-mortems on protocols that failed, and the pattern is consistent. The balance sheet tells you before the price does. In 2022 I audited three lending protocols during the Terra collapse and found one carrying $30 million in undercollateralized positions created by oracle manipulation. The alert went out privately and prevented a $5 million loss. Nothing about that finding required a prediction. It required reading a number that had been published and noticing what it implied.

The missing revenue line is that kind of number. A nine-figure governance pledge is not the interesting disclosure. The interesting disclosure is the one that is absent.

Contrarian: the cynical read is right and also lazy

The prevailing interpretation is that this is a public relations maneuver timed to an IPO window. That reading is directionally correct and analytically incomplete, and the difference matters for anyone trying to price the stock.

The cynical version says the pledge is meaningless because it is unenforceable. The sharper version says the unenforceability is the product. Anthropic is buying something concrete — reduced regulatory friction, a responsible-AI narrative at the moment of pricing, and a defensible answer to the question every institutional allocator now asks about labor displacement. It is paying with a contingent claim on future paper wealth that costs nothing if the stock declines and dilutes only nominal value if it rises. That is not deception. It is an option premium, and from the issuer's side it is well priced.

The second correction is about causation. The timing correlates with the IPO. That does not make the IPO the cause. The cause is a regulatory window: the EU AI Act's transparency obligations for general-purpose models, a growing patchwork of state-level AI legislation, and SEC review at listing. Governance announcements cluster around regulatory deadlines for the same reason token projects announce audits before exchange listings. The correlation is real. The causal story is about compliance, not sincerity.

The third correction concerns the number itself. A $48 billion headline is not a lie. It is correct arithmetic at a specific valuation on a specific day. The honest criticism is not that the figure was inflated. It is that the figure was published without a denominator, and a reader with no denominator anchors on the largest version available. I made that mistake in my own early writing, quoting notional figures without their base, and I stopped after an investment committee asked me a question I could not answer.

There is also a blind spot in the coverage itself. Everyone is arguing about whether the commitment is sincere. Sincerity is not auditable. Terms are. And the terms are missing. Meanwhile the absence of a revenue line goes unremarked, because a governance fight is easier to write about than a blank column in a spreadsheet.

Finally, consider what a credible version would look like, because the fix is not exotic. A published baseline. A named independent trustee. A quarterly attestation with the fund's holdings labeled by entity. A disposition schedule with a stated lockup. A public dashboard anyone could query. That is not a high bar — it is the standard we already apply to a mid-sized DeFi protocol with $50 million in a treasury. The fact that a company at this scale is being held to a lower standard than a Curve pool is the part that should bother people.

Takeaway: four lines in the prospectus will settle this

When the filing lands, four disclosures determine whether this pledge is a governance innovation or a footnote with a marketing budget.

The first is share source: primary issuance or secondary transfer. If primary, dilution risk moves from a reasonable inference to a documented fact, and every IPO buyer should re-run their model before the book closes.

The second is the baseline definition for the escalator, and whether it is fixed at the IPO price or set by the board afterward.

The third is the voting treatment of the fund's shares, and whether an independent trustee exists with real authority over disposition.

The fourth is the lockup and disposition schedule — because a 5% to 11% block with no published disposition rules is an overhang with no expiry date, and overhangs without expiry dates get priced by the most pessimistic holder in the book.

Silence is just data waiting for the right query. Right now the query returns nothing, because the fields do not exist yet.

Which raises the question I keep returning to. The most scrutinized governance commitment of the year will be adjudicated not by a transaction hash but by a footnote in a registration statement — and every reader will have to decide how much weight to give a promise whose enforcement depends on the same board that benefits from weakening it.

Truth is found in the hash, not the headline. There is no hash here. That is the whole story.