Reading the Cap Table Before You Sign an ML Engineering Offer at an Early AI Startup
Understand the cap table before equity becomes worthless common stock.
When weighing ML engineer career opportunities at Series A and Series B AI startups, look beyond the equity percentage. Before signing, check who gets paid first, what it costs to exercise your options, and how future funding changes how much you own. Under those terms, the startup’s upside may reach you or stop at its investors.
Why ML engineers are the cohort most exposed to equity that looks real but pays nothing
Early AI startups use equity to compete for ML engineers they cannot win on cash compensation alone. In 2026, frontier labs are in a pay tier of their own, while funded startups lean on ownership and the chance to shape a product. That puts equity at the center of the decision. An engineer accepting that trade needs to know what the promised ownership can actually deliver.
The offer letter leaves out much of what controls the payout. Meanwhile, AI startups raise large rounds at high valuations, so investors already have large claims by the time an engineer joins. Even when a product does well, employee shares can be worth little. Reading the cap table before signing is a learnable skill.
How the liquidation waterfall turns a healthy exit into nothing for common stockholders
Investors usually own preferred stock, while engineers own common stock or options that convert into common. The liquidation waterfall sets the order of payment, with investor preferences ahead of common stock. Your percentage does not override that payment order.
Consider a hypothetical AI startup that sells for a price that sounds healthy. Its engineers have vested equity, but the company raised enough money on protective terms that the sale proceeds do not clear the aggregate liquidation preference. The investors receive the proceeds available under their claims. Nothing remains for the engineers’ common shares. The company found a buyer, but employee equity produced no payout.
The terms determine how hard that threshold is to cross. With a non-participating preference at the base multiple, investors choose between taking their money back and converting to common. They cannot do both. Participating preferred lets investors take their preference and then share in the remainder, reducing the pool available to employees. A multiplied preference makes the hurdle higher still because investors claim a multiple of their investment before common receives anything.
Large AI funding rounds amplify this mechanism. A preference attached to a large investment can create a substantial dollar claim even when the multiple looks ordinary. Later rounds can add more claims that get paid before employee shares. The useful number is therefore the aggregate dollar preference.
What strike price and 409A valuation actually tell you about your cost to own the equity
Most early-stage ML engineering grants are stock options. An option gives you the right to buy shares at a set strike price. Ownership comes with an upfront purchase cost. Multiplying the strike by the number of options you exercise gives the amount you must pay.
The 409A valuation establishes the common stock’s fair market value used to set the strike price. Carta describes it as an independent appraisal required for IRS Section 409A compliance. Common stock is priced below preferred because it has fewer protections and more risk. The useful comparison is between the option’s strike, the current 409A value, and the latest preferred share price.
That comparison reveals the discount built into the option. A wide gap between common and preferred prices signals early-stage risk, while a narrowing gap suggests a more probable exit but less discount. Neither gap guarantees a payout. The liquidation waterfall still governs what reaches common.
The exercise window adds a separate constraint. With a 90-day post-termination window, leaving the company forces you to decide whether to pay the exercise cost within that period or lose the options. At a well-funded AI startup, that bill can reach six figures. An extended window, written into the grant agreement, gives you more time to make that decision and belongs in the offer comparison.
How dilution across future rounds erodes the percentage you signed for
The percentage in your offer is a snapshot. It is calculated against a fully diluted share count, and future financing adds shares to that denominator. Your grant can stay the same while the percentage you own falls. Any startup that raises again introduces this dilution.
For a Series A versus Series B job comparison, funding stage helps frame the questions but does not answer them. A Series B company may already have seed, Series A, and Series B preferences to check. A Series A offer still needs a model of the rounds ahead.
Companies create or enlarge pools at or before new rounds, typically diluting existing holders, including earlier employee grants, while leaving incoming preferred investors undiluted. The timing belongs in your model. A percentage that excludes a planned pool expansion can give a misleading picture of the ownership you are evaluating.
Investors can also have pro-rata rights that let them maintain their percentage by investing in later rounds. Employees generally lack that protection. To model the difference, ask for the fully diluted share count behind your grant. The payout analysis now has three parts, which are investor priority, the cost to exercise, and ownership after dilution.
What should an ML engineer ask for before signing a startup offer?
The practical step is to request four documents before signing. A summary cap table shows shareholder classes, share counts, and the preference terms for each series. A summary version is a normal employee request. Its purpose is to expose the structure behind the offered percentage.
Next, request the term sheet or certificate of incorporation, which contains the liquidation preference multiples and participation rights for preferred series. The certificate is a public filing in Delaware and can be requested directly. Then get the stock option grant agreement in writing before signing the offer. It should state the share count, strike price, vesting schedule, cliff, and post-termination exercise window. Finally, request the latest 409A valuation report summary and ask when it was updated.
Five questions turn those documents into a usable conversation. “What is the aggregate liquidation preference in dollars across all series?” establishes the total claim above common. “Is the preferred stock participating or non-participating?” establishes whether investors also share in the remainder. Both are normal questions about the pay on offer.
The next question tests the denominator. “What fully diluted share count is my grant calculated against, and does it include a planned new option pool?” Then ask, “What is the post-termination exercise window, and can it be extended in writing?” Close with, “Has there been a tender offer or secondary transaction, and if so, at what price?” That history shows what buyers have actually paid, rather than what an offer presentation suggests the equity might become worth.
How the equity structures at OpenAI and Anthropic differ from a typical early-stage AI startup
OpenAI and Anthropic are not typical early-stage employers. Both publish salary ranges but not equity amounts, leaving equity to be negotiated at the offer stage without a public benchmark. A familiar name is not a substitute for the terms of the grant.
At an early startup, options are the usual starting assumption. At either lab, the diligence begins by establishing exactly what instrument the offer provides, including whether it is RSUs or options and, for options, the strike price. The share or unit count needs to be confirmed in writing. So do the vesting schedule, cliff, valuation used to price the grant, and recent tender-offer history.
Published salary ranges cannot supply missing equity terms, and the terms of one offer cannot stand in for another. The same discipline applies at smaller startups. There, it becomes even more consequential because cap tables are less public and investor preference terms can be more aggressive.
Where an early AI startup’s cap table is most likely to hide engineer-unfriendly terms
Early AI startups need large checks to pay for compute and talent, which gives investors reason to seek stronger downside protection. The first red flag is participating preferred with a high multiple. It combines payment priority with a claim on the remainder. On a large round, that combination can consume most of a moderate exit.
Stacked senior preferences are another red flag. If each new series ranks ahead of the previous one, an engineer joining after seed, Series A, and Series B encounters several layers of claims. Ask for their aggregate dollar amount. Knowing each round’s multiple separately is not enough to judge the hurdle facing common stock.
A high strike price paired with a 90-day exercise window is a red flag for mobility. A stale or unaudited 409A is a valuation warning, especially if it predates the latest round and may not reflect current common stock value. If the company cannot or will not share a summary fully diluted cap table, you cannot reliably evaluate its equity promise.
Green flags point in the opposite direction. They include non-participating preferences at the base multiple, a multi-year exercise window in writing, and a recently updated 409A. A proactively shared cap table supports a clearer discussion. A record of tender-offer history shows there has been liquidity before, though you still need to work through the terms of your own grant.
How to model your equity’s realistic value before the offer deadline closes
This is a five-step sanity check. With the documents in hand, it should fit into under an hour. The model estimates where your equity begins to pay and how plausible that exit is. A rough model is more useful than an unsupported upside claim.
First, add up the aggregate liquidation preference across all preferred series. Second, divide your grant shares by the fully diluted share count and allow for future rounds and pool expansion. Use that ownership figure as an assumption for the scenario, since the share you are entitled to can change.
Third, test exits at the latest post-money valuation, a modest multiple of it, and a higher multiple. For each, apply the preference and participation terms to find what remains for common, then estimate your share. Participation changes the answer after the initial preference is paid. A model that simply subtracts the stack and ignores those rights can overstate employee proceeds.
Fourth, subtract the exercise cost from your estimated gross payout. The cost is the strike price multiplied by the options exercised. Fifth, ask the company to go through the same scenarios. A founder who has raised institutional capital has modeled the cap table, so this is a normal request. Their willingness to explain the result is another useful signal before the deadline.
What to negotiate once you understand the cap table
The negotiation should follow the risks the model exposes. If a large preference stack or narrow 409A discount weakens the offer, ask for a larger grant to reflect that risk. More shares can give you more of a payout, though only if there are proceeds left for common.
Ask for a five- to ten-year post-termination window in the grant agreement so you have time after leaving to decide whether to buy. Also ask whether the plan permits early exercise before vesting at the 409A price. With an 83(b) election filed within the IRS deadline, early exercise can cut the tax cost if the company’s value rises a lot.
Future ownership deserves its own discussion. Ask about refresh grants at performance reviews, since dilution keeps reducing the share you own if your initial grant has no refreshes. Then ask whether the option plan has double-trigger acceleration. That provision accelerates vesting when both an acquisition and termination or a material downgrade occur, protecting the engineer against losing unvested equity in that situation.
These are professional compensation questions. The time to raise them is before signing, while the offer is still being negotiated. With written terms, you can evaluate the benefit being discussed. The percentage on the offer letter begins the conversation, but reading the cap table reveals whether the startup upside can actually reach you.