Dilution Across Funding Rounds and Its Real Impact on Engineer Grants
Equity percentage is only a start; the real value lies in dilution and payout order.
For ML engineers comparing Series A and Series B AI startup offers, the equity percentage is a starting point. New investor shares and option pool expansions shrink that percentage, while liquidation preferences can leave common shareholders with nothing at exit. What a grant is worth depends on the fully diluted ownership, how much funding the company still needs, the refresh terms, and how much exit value is left for employees.
Why your grant percentage and your grant's value are two different things
An engineer gets an offer with a small equity percentage and hears that it could become valuable. But a percentage of what, and for how long? Dilution increases the total number of shares against which the unchanged grant is measured, so the engineer owns less of the company after new shares are issued.
That smaller stake can still produce a larger payout. If the company grows enough, its rising value can outweigh the loss of ownership. The mistake is treating that growth as guaranteed, especially through flat rounds, down rounds, or modest exits. Share count and strike price tell part of the story, but neither shows how the cap table will split the proceeds.
The two dilution forces that hit simultaneously at every round
New investors get newly issued shares when a company raises a priced round. Those who already hold shares keep them but own a smaller share of the new total. That is the obvious dilution. The less obvious hit comes when investors require the company to expand its employee option pool before the round prices, reserving more equity for future grants.
The timing determines who absorbs the pool expansion. With a pre-money expansion, existing shareholders bear that cost before the new investors arrive. With a post-money expansion, the new investors share it. Most term sheets require the pre-money version, so employees can lose ownership to the pool refill and then lose more to the investor stake. Looking only at the money raised misses part of the reduction.
The denominator therefore needs to be fully diluted shares. That count includes common stock, preferred stock on an as-converted basis, granted options whether vested or unvested, unissued pool reserves, warrants, and convertible instruments. Before the financing, the grant sits over one fully diluted total. After the pool expansion and share issuance, the same grant sits over a larger total. Using outstanding shares alone makes ownership look larger than it is.
Convertible notes and SAFEs add a timing wrinkle. They do not dilute at issuance but convert at the next priced round, often at a discount. Their conversion adds shares on top of the round itself, so the headline investor stake may not describe the full ownership change.
What the round-by-round dilution actually looks like on a cap table
Follow a small seed grant through Series A, Series B, and Series C. At Series A, its starting percentage is multiplied by the ownership fraction retained after the pool expansion, then by the fraction retained after the investor issuance. Series B starts with that reduced percentage. Nothing resets to the original grant. The next financing acts on what remains.
At Series B, the engineer takes that smaller percentage and multiplies it by the next pool-retention fraction and then by the next investor-retention fraction. At Series C, the engineer does the same math with the Series B result. Written as one chain, the final stake is the seed percentage multiplied by every retention fraction across all three rounds.
Adding the rounds’ dilution percentages together does not produce the correct cumulative loss, because each round acts on a different base. But looking at any round in isolation understates the journey. A grant that survives the first financing with an apparently modest reduction can become substantially smaller after repeated pool refills and investor issuances.
A senior engineer might receive a larger percentage at Seed, a smaller one at Series A, and a smaller one still at Series B. Those are entry stakes. The Series A offer therefore needs to be compared with the Series B offer after allowing for the subsequent funding path, rather than declaring the larger initial percentage the winner.
When the percentage math translates to dollars — and when it does not
A large exit can make a heavily diluted grant worthwhile. But exit price alone is not enough. Investors hold preferred shares, and liquidation preferences give them a claim on the money before employees who hold common stock get anything.
Take the same diluted grant through two kinds of exit. In the first, the sale produces enough to satisfy the preferred investors’ claims and leaves proceeds for common shareholders. The engineer can then receive a share of the amount available to common under the applicable terms. In the second, the company has raised in three rounds, and its cumulative preferences equal or exceed the sale price. Common gets zero, even for engineers whose options have vested.
The grant’s ownership percentage has not changed between those scenarios. Its place in the payout order has. Multiplying ownership by the headline exit price skips that order and can make a claim worth nothing look like a large payout. That gap can be especially costly when a company has raised a lot of money but sells for less than its last round valuation.
The preference multiple at each round and whether the shares are participating or non-participating belong in that calculation. The legal labels need not take over the talk about the offer. They spell out how the money is split and when common starts to get paid. Vesting establishes an engineer’s entitlement under the grant, but it does not move that grant ahead of investors in the payout order.
How down rounds accelerate dilution and move options underwater
When the new financing prices below the previous round, anti-dilution clauses can adjust the conversion price of earlier preferred shares downward. Those investors then have more common-equivalent shares. Founders and employees who hold common stock bear that dilution but get no such protection.
Broad-based weighted average anti-dilution is the standard clause described here. It adjusts the earlier conversion price rather than resetting it entirely to the new lower price. Full ratchet goes further by making that full reset. It is rare, but its ownership cost to common holders is much greater. The new investor stake is therefore only part of the damage a down round can do.
Options face a separate problem at the same time. If their strike price exceeds the new post-down-round 409A valuation, they are underwater. Exercising costs more than the shares’ current fair market value. A board can respond through repricing, an RSU exchange, or a fresh grant at the reset 409A valuation, but none of those remedies is automatic or guaranteed.
This is a planning risk. A meaningful share of venture-backed companies have traded below their last private mark in recent years. Even if a company later recovers in value, that recovery does not reverse the ownership reduction already imposed by the financing.
How refresh grants partially offset ongoing dilution — and where they fall short
Refresh grants are one response to that erosion. After year one or two, most growth-stage companies issue additional equity to keep total compensation at market. A new grant gives the engineer more upside than the first award alone and more unvested equity still to earn. This retention mechanism makes leaving costly because the engineer still has equity to earn.
The quality of that mechanism varies sharply between companies. Strong programs keep renewing the unvested stake, while weak programs can leave engineers below market by year three with little new equity to retain them. Earlier-stage startups commonly offer the less predictable version. A promise that refreshes “happen” is therefore less informative than the timing, size, and approval process behind them.
Some companies instead tie refreshes to technical milestones. This structure ties additional equity to a defined latency target, a user milestone, or a safety benchmark. For an ML engineer, it also makes the terms of the technical outcome part of the compensation discussion.
Anthropic and OpenAI show how stronger refresh practices work, with grants that keep total compensation growing year over year. Anthropic’s shift toward RSUs with double-trigger vesting removes the option strike-price structure, but the value remains illiquid until a qualifying liquidity event. A tender offer, acquisition, or IPO is not on a fixed schedule. At a typical startup, refreshes are smaller, less predictable, and subject to board approval, and every new grant remains exposed to future dilution.
The frontier AI lab exception and what it means for everyone else's math
Does joining OpenAI make this calculation irrelevant? An engineer who held equity before a dramatic valuation increase could capture appreciation that overwhelmed ordinary percentage dilution. That outcome depends on the entry window. Someone joining after the increase starts at a much higher valuation with less room for comparable appreciation.
Large grants do not suspend the underlying mechanics. Their dollar values reflect the valuation at which they are offered, while future ownership still depends on how the company issues equity. A lab offer at a high valuation may have less appreciation runway than a smaller startup grant at a lower valuation.
For engineers considering Series A and Series B AI startups, lab compensation is a reference point alongside cap-table analysis. The realistic comparison remains specific to the offers. Entry valuation, initial ownership, later dilution, refresh practices, and access to liquidity all shape the outcome. An exceptional return from an earlier entry window does not establish what a new hire will receive.
How do you evaluate and negotiate an ML engineering equity offer?
The cap table provides the basis for an engineer to negotiate an offer. Requesting the fully diluted shares outstanding lets you work out what percentage of that total the grant is. Then ask how the option pool was sized at the last round and how much remains unallocated. A nearly depleted pool points toward a pre-money top-up at the next raise, with existing holders bearing the reduction.
Ask next for the liquidation preference stack. The relevant terms include total capital raised, each round’s preference multiple, and participating versus non-participating treatment. Model a range of exits to identify where common receives nothing and where it begins to participate. This turns the question from “What could the company sell for?” into “What would this grant receive if it did?”
To compare Series A and Series B offers, ask each company to walk through the funding rounds still ahead and what the ownership percentage would be after them. A larger Series A grant should not win by inspection. The comparison needs to account for the dilution that may follow it, while a smaller Series B grant still needs scrutiny of the preferences already above it. Willingness to explain these scenarios is useful information, and reluctance to engage is a signal.
The grant terms deserve a separate negotiation. A four-year vest with a one-year cliff is a standard founding-engineer anchor, as is double-trigger acceleration on acquisition. Single-trigger acceleration is rare but worth asking for explicitly. Ask for a refresh at year one, ahead of the standard performance cycle, with the refresh added to the initial grant. Grant bands can also have internal flexibility, and hiring managers can often offer more equity within a band without executive approval.
These questions belong in the discussion when the equity terms first appear. The useful outcome is an offer whose initial ownership, future dilution, payout order, and renewal terms can be explained plainly. That is a firmer basis for choosing a job than the largest number printed in the grant letter.