Cap Table Decoded
Cap Table DecodedEquity Structures at AI Labs vs. Product Startups
Equity NormsLong read

Equity Structures at AI Labs vs. Product Startups

Lab RSUs and startup options differ in cash-out terms, not just size.

· 9 min read

For an ML engineer comparing Series A or Series B startup opportunities with a frontier lab offer, equity needs to be judged by its terms. Lab RSUs depend on private valuations and company-run sale windows, while startup options add exercise costs and usually depend on an acquisition or IPO. The better offer depends on what the engineer can keep, when it can become cash, and what must happen for the quoted value to hold.

Why equity at AI labs and product startups are structurally different instruments, not just different sizes

Equity discussions tend to start with dollars or percentages. A lab may put a large dollar value on the grant, while a product startup offers a percentage of fully diluted shares. These are different ways to describe grants, but neither shows who decides when employees can sell.

At a frontier lab, the quoted value is tied to a private funding-round price. At a product startup, options or RSUs sit against a much smaller valuation, with a different dilution trajectory and liquidity horizon. The grant also behaves differently when someone leaves. An options holder may face a payment decision that an RSU holder does not. A total-compensation figure compresses all those differences into one number, hiding much of what makes the offer valuable or fragile.

How frontier labs use RSUs against private valuations, with tender offers as the only exit

OpenAI, Anthropic, and xAI grant RSUs priced against private funding-round valuations. Before an IPO, the route from those RSUs to cash is a company-run tender offer. Employees can sell only when the company opens a sale window. Vesting and selling are separate events. An engineer can therefore have vested equity without an immediate way to turn it into cash.

OpenAI’s equity history shows how much the underlying structure can change. The company converted PPUs to RSUs in January 2026, following its October 2025 restructuring into a for-profit public benefit corporation. In August 2026, it completed a secondary share sale that let current and former employees sell at its then-current valuation. OpenAI paid for that deal itself, with no money from outside investors. An earlier tender had been expanded to accommodate investor demand, but employees sold less than investors wanted to buy.

Anthropic’s most recent tender also fell short of investor demand because most employees held their shares, betting on a higher IPO price. Its preliminary prospectus describes an offering at $70–$78 per share, providing a concrete signal of a possible public exit. Across US private companies, secondary transactions have overtaken IPOs as the primary mechanism for employees to convert equity to cash. Even so, a tender remains a discretionary event.

How the cliff elimination at OpenAI changed the retention calculus across the whole lab market

OpenAI’s removal of its equity cliff changed the cost of leaving before the first anniversary. Before the change, an engineer who left at Month 11 forfeited all equity. After it, that engineer keeps everything vested to that point.

Anthropic retains its one-year cliff, so departure before that date means zero equity, whether the engineer resigns or is laid off. Suppose a company restructures before an employee reaches the cliff. At Anthropic, that employee would leave without equity, while the same scenario at OpenAI would leave months of vested RSUs. Those RSUs would still need a liquidity event.

Across the market, competing offers now differ on a term that directly changes the retention bargain. Headline compensation does not show that difference. For anyone who might not complete a year, the cliff introduces an all-or-nothing condition that can outweigh a modest difference in grant size.

Why equity-as-percentage-of-grant means something very different at a seed startup than at a lab

Product startups usually express equity as a percentage of fully diluted shares. For the earliest engineers, that percentage is the main lever. Entry timing strongly shapes the grant. Carta’s seed-stage data shows a meaningfully higher median grant for the first engineering hire, with the median falling sharply by the fifth hire.

That percentage is not a fixed claim on future proceeds. Later funding rounds can dilute it, so the initial grant shows how much the engineer owns before those rounds. Options also introduce a strike price, an exercise window, and a tax event at exercise that RSUs do not have. Leaving can force a financial decision. An engineer who departs before an exit may need to pay to keep the equity or forfeit it.

The liquidity path compounds that difference because an early-stage startup generally lacks periodic tenders that provide partial cash along the way. Its equity depends on an acquisition or IPO and is effectively worth nothing without an exit. A substantial ownership percentage can therefore remain inaccessible for the entire period of employment. Unlike a lab’s quoted RSU value, it also leaves the exercise cost unresolved.

How AI lab hiring pressure is forcing startup equity floors up — and what that actually signals

At frontier labs, equity makes up most of the total compensation at senior levels, which sets a bar that early-stage startups cannot meet with cash alone. Startups consequently compete through larger ownership percentages. Carta’s data shows that median initial grants for AI and ML engineers have grown faster than for any other early-stage roles, especially at the smallest startups. The pressure reaches beyond research hiring. A startup hiring its first ML engineer now has to win over candidates who compare its offer with both lab compensation and older startup grants.

A rising grant floor may mean a company must issue a larger percentage of a smaller or more speculative business to attract the same candidate. Engineers who join early can get those larger grants before later rounds dilute their ownership. By Series B or later, grants sit on a much larger denominator.

The round label alone cannot tell an ML engineer which equity offer is better when looking at roles at Series A and Series B AI startups. The comparison still has to account for the percentage, the valuation, and how much dilution is expected. Startups also compete through faster shipping cycles and broader scope, which are distinct noncash benefits. The pressure to hire explains why an offer has grown. It does not prove that the resulting equity will deliver the value used to recruit the engineer.

What engineers get wrong when comparing a lab RSU offer to a startup options grant

The first mistake is treating a lab’s stated RSU value as cash. It is a private valuation number whose realization depends on a tender or IPO, neither guaranteed at the grant price. The second mistake follows from the first. Engineers may treat the latest funding-round valuation as a floor, even though a downward re-rating would proportionally reduce the grant’s realizable value. An equity-heavy package is a concentrated bet on that valuation holding or growing.

Engineers focus on the ownership percentage while giving too little weight to the strike price, exercise window, and chance of a sufficiently valuable exit. A percentage does not pay the exercise bill. Nor does it say whether an employee who leaves can afford to preserve the potential upside. Ignoring those terms can make an attractive ownership figure look more usable than it is.

A further error is assuming equal headline totals mean equal vesting protection. Anthropic’s cliff and OpenAI’s lack of one make that assumption false for someone leaving before a year. The comparison needs to account for instrument type, vesting including the cliff, liquidity mechanism and tender history, and the valuation thesis. Each explains a different failure point. Without them, the comparison puts a conditional dollar estimate beside a conditional ownership estimate and pretends both are pay.

How the researcher premium works at labs, and why it matters for the equity evaluation

At frontier labs, most of the extra pay for research comes in equity. At Google DeepMind, base salaries are essentially equivalent across research and engineering tracks at the same level, while research scientists receive modestly larger equity grants. The distinction changes what “higher pay” means. Moving tracks can increase exposure to equity without producing a comparable increase in dependable salary.

At Anthropic, 2026 compensation data puts the research-role premium over engineering at 15–30% at the same level, larger than at DeepMind. The gap widens at senior tiers, with the difference concentrated in senior and principal researcher packages. Headline offers are a separate category. Nine-figure deals for individual researchers are negotiated for a handful of specialists in pretraining, reinforcement learning, and reasoning. The qualified pool is small enough for labs to track researchers at peer institutions individually.

For an engineer comparing tracks, those outliers are therefore a poor benchmark. Base pay will be similar, and the equity difference is real but smaller than exceptional public offers suggest. A larger grant also increases the dollars exposed to the same valuation and liquidity conditions. It does not remove those conditions. If the private valuation falls, a larger grant loses more stated value even though the proportional change is the same.

How xAI’s equity differs from both OpenAI and Anthropic — and what the liquidity gap means in practice

xAI has used high headline compensation to attract engineers from OpenAI and DeepMind. Its Colossus compute cluster also helps draw researchers who want large training runs with fewer people inside the company competing for resources. Those are distinct parts of the offer. Access to compute shapes the work a researcher can do, while the RSU terms set out how the equity might turn into cash. One does not resolve the other.

The liquidity distinction is that xAI has not built the tender history of OpenAI and Anthropic. Anthropic has run multiple tenders at rising valuations, while xAI’s realized equity value depends on a future tender or IPO that is not currently scheduled. A shared instrument name obscures that gap. An RSU at a company with past employee sale windows has a different liquidity profile from one without that record.

An established history of cash realization distinguishes these offers from those with a less developed record, though future tenders remain uncertain. The xAI offer may present concentrated upside potential, but the timing of access remains unresolved.

How should an ML engineer evaluate an early-stage startup offer against a lab role?

The founding engineer role makes this trade-off especially sharp because the first engineering hire at a seed-stage company captures a meaningfully higher percentage than later hires. That percentage is set before substantial dilution from later rounds. The opportunity is genuine, but so are the conditions. A lab grant and a founding engineer grant cannot be judged fairly by converting both into optimistic dollar estimates.

The practical evaluation starts with the instrument, including the cost and window for exercising options. After early departure, employees keep equity vested under the grant’s terms, while cash can arrive through periodic tenders or only an exit, depending on the liquidity mechanism. These are separate gates. The valuation anchor also shows how much of the promised reward rests on a speculative price. Finally, the probability-weighted outcome weighs the chance that a startup will reach a valuable exit against the chance that a lab will hold or raise its private valuation.

For Series A and Series B roles, the same reasoning keeps stage from becoming a shortcut for value. The actual grant and its terms remain the basis of comparison. The chance to take on more, build products that reach users, and get feedback faster can still be good reasons to choose a startup over a lab. That is a legitimate career trade. It should be made explicitly, with a clear account of what the engineer will hold, when it can be sold, and what the company must achieve before the offer’s equity becomes wealth.

Filed underEquity Norms

More in Equity Norms