Equity Refresh Grants at Growth-Stage Startups
How refresh policy, not just the offer, shapes an engineer's stake.
For an ML engineer weighing a Series A or Series B AI startup offer, equity refresh grants decide how much of a stake remains as the initial grant runs down. The offer is the joining grant along with the company’s policy for adding equity, the vesting terms attached to it, and the pool available to fund that promise.
Why a four-year grant stops working long before the company exits
A thirteen-year median wait for an exit does not fit a four-year grant. The grant can finish vesting long before the employee has a chance to turn it into cash. An engineer who joined at Series A is roughly halfway vested after two years, just as competing employers can offer a fresh four-year grant at a higher valuation. Staying means watching the remaining award shrink, unless the company adds more.
Engineers in years three through five feel this pressure most. These are often the people who built the product and understand its customers, so their institutional knowledge has compounded while their unvested equity has declined. Losing them is expensive and disruptive. The AI talent market has sharpened that mismatch because rising new-hire packages for AI and ML engineers make older grants look stale even when the employee’s role has not changed.
What a refresh grant actually is and what it is not
A refresh grant is an additional equity award to an existing employee, on top of the original new-hire grant. Companies also call it a refresher grant, top-up grant, or annual equity grant. A refresh grant keeps unvested equity ahead of the employee as the original award vests down, with each refresh following its own schedule, typically four years, with or without a cliff.
A promotion grant serves a different purpose because it rewards increased scope rather than maintaining compensation parity at the same level. A sign-on bonus rewards joining, while a one-time retention award responds to a particular risk of departure. A refresh program is a recurring discipline. Those distinctions prevent an offer letter from making a conditional promotion award or a reactive payment look like a dependable source of future equity.
The five structures companies use and the trade-off each one makes
The five approaches to distinguish are promotion-triggered, tenure-based, performance-driven, back-weighted, and boxcar grants. They are not perfectly separate categories because a company can combine an eligibility trigger with a particular vesting schedule. That overlap can obscure the real terms. The useful distinction is between what earns the award and when the award starts becoming yours.
Promotion-triggered grants give engineers more equity when they take on more responsibility. This gives engineers a clear link between leveling and compensation, but it does not solve the problem for someone doing strong work at the same level. Tenure-based grants instead use time at the company as the trigger. Who gets them depends less on how their work is rated, but whether they stop unvested equity from dropping later still depends on the vesting terms.
In Carta’s performance-driven model, performance reviews set the size of the refresh, which starts vesting right away on a traditional four-year schedule. Carta’s tenure-based “boxcar” grants work differently, starting vesting two years after the grant date. These terms should not be confused. The delay is intended to prevent the year-five equity cliff that can undermine tenure-based programs, but it adds administrative work and makes forward equity harder for an engineer to predict.
Amazon’s back-weighted model concentrates most equity in later years and offsets the early years with large cash sign-on bonuses for senior engineers. The cash is part of the design. A package can look generous in total while delivering relatively little equity during years one and two, so the offset changes how the offer works in practice. Startups burning through runway cannot just copy that mix of cash now and equity later.
A sequential boxcar grant is given in year two or three and starts vesting only when the original four-year award ends. The schedules run one after the other rather than overlapping. That makes the sequence predictable, but it creates a transition window in which the new award provides no current vesting. An engineer therefore needs the actual dates for the company’s model.
How refresh grants are sized at different company stages
Pave’s equity program data shows that how refreshes are shared out varies with stage and level, with no single formula for all companies. Carta’s data places refresh grants meaningfully below a new-hire award for the same role or level. A refresh is not another full joining package. Earlier-stage companies reserve part of their equity pool for refreshes, while later-stage companies devote more to them as retention becomes a larger concern.
Pave’s 2026 equity program data shows that more than four in five director-level employees get refresh grants, while about a quarter of P1-level employees get them. Level changes the odds substantially. Promotion is also the strongest trigger, with a strong majority of promoted employees receiving an ongoing grant, compared with fewer than half of high performers who are not promoted.
The data also shows that roughly one in five employees receives a refresh by year one, rising to nearly half by year two. That does not make a grant automatic. An engineer who reaches year two without a talk about a refresh is already behind the norm in the data, so asking at that point is hardly early. For engineers seeking more equity, the practical lesson is to raise leveling and refresh questions together without treating the two awards as interchangeable.
For a Series A versus Series B comparison, company stage provides context for grant size. The more mature company has a stronger reason to fund repeatable retention, but the engineer’s level and the company’s actual policy still determine the offer. AI-native startups add another distinction. At the upper compensation ranges and above a valuation threshold, their equity packages are materially higher than those of non-AI-native startups at the same valuation. A literal first engineering hire at seed stage is a separate case, with a meaningfully larger grant than a senior engineer joining an already-staffed team.
When refresh timing follows company maturity rather than a calendar
Promotion-based awards come first, once the company has a leveling framework, and performance-based awards follow once reviews are calibrated. Tenure-based awards come later. They become more useful when employees reaching the end of their original vesting schedules create a recurring need to retain them.
At seed stage, the initial grants have only recently started vesting, and the company may be too small to need structured retention. Once there are 30 or more employees and the first hires are approaching one to two years of service, refresh becomes a live concern. The calendar alone does not explain the shift. Headcount and tenure together make it harder to manage equity through individual conversations.
The early-stage objection to a formal program is reasonable because founders can still know each engineer’s vesting position and act as needed. That approach has a limit. As the organization grows, visibility weakens, and a vesting cliff can arrive before anyone has noticed it. An engineer comparing Series A and Series B opportunities should therefore distinguish a small company’s deliberate flexibility from a larger company’s failure to build a process.
What OpenAI's January 2026 equity restructure reveals about how companies signal commitment
OpenAI’s January 2026 equity restructure made its packages simpler and more liquid in engineers’ eyes. The cap on upside had been a structural disadvantage relative to standard equity at competing labs, while removing the cliff signaled a willingness to compete on vesting terms. The headline award was not the whole offer. Limits on upside and delays before vesting shaped how real that award felt to the person receiving it.
The case gives engineers a reason to scrutinize the instrument alongside negotiating the number. Even a well-known AI lab had to change terms that made its equity less appealing. That does not make every startup’s package directly comparable.
For a growth-stage startup, the same distinction separates a compensation promise from terms an engineer can evaluate. A company signals commitment through when vesting begins, what restrictions apply, and how clearly it explains the award. Those details deserve attention before an offer is signed. Otherwise, a negotiation can improve the headline while leaving the structural disadvantage untouched.
The dilution reality engineers rarely model and founders often underestimate
Refresh equity has a cost beyond the award itself. New equity dilutes existing shareholders, including employees who already hold equity, so a refresh can give an engineer more years of vesting while shrinking the share of the company held through the original grant. That does not make refreshes a bad bargain. Engineers should weigh the new grant and the old stake together because one affects the other.
A 15% option pool sized for initial grants and two years of hiring can run short much sooner when refreshes consume a material share of annual grants. Founders who have not budgeted for that demand can exhaust the pool before the next financing. The promise then meets a funding constraint. Engineers can expose that constraint by asking about the current pool, the share already allocated, and when it was last refreshed at a financing round.
The time to liquidity also matters because a fresh four-year award may still be vesting when an exit comes. Engineers near that point cannot treat the full grant as already earned. Its value depends heavily on the likely time to liquidity and what happens to the unvested portion. A larger refresh with a long remaining schedule is therefore not interchangeable with equity that will vest sooner.
Milestone-based refreshes are a newer way to address this by giving more equity for specific technical results. Those results can include cutting model latency by a set amount, reaching a user milestone, or hitting a set benchmark. The trigger is a result. This can give high performers who deliver measurable outcomes meaningful additional compensation faster than a fixed schedule would.
How should an ML engineer evaluate a startup’s refresh policy before signing?
First, engineers need to establish whether the company has a documented refresh policy or grants equity “as needed. A written policy is a commitment, while an informal promise depends on whoever controls compensation when the review arrives. The distinction deserves a direct answer. For remote engineers, it is especially revealing because distributed-first companies tend to codify policies earlier when managers cannot rely on face time to judge retention risk.
Next, engineers need to ask whether awards are annual, biennial, promotion-triggered, or tied to performance. That answer needs to be paired with the vesting schedule. Does the refresh have a cliff, and does vesting begin at the grant date or later? A two-year delay like Carta’s tenure-based design is materially different from immediate vesting, and either schedule needs to be understood relative to the end of the original award.
Refresh grant size needs to be evaluated at the engineer’s level. Carta’s benchmark is a refresh worth three tenths of the original grant, so an offer materially below that at Series B or later deserves probing. The benchmark gives an engineer a concrete basis for asking how the company sets awards and what it means by “competitive equity”.
Engineers also need to establish whether the company has enough pool capacity to support the policy it describes. A nearly exhausted pool means further commitments depend on board approval of a new pool, which may not arrive. Together, the written policy, cadence, grant size, pool capacity, and vesting terms reveal whether refresh equity is a functioning program or an aspiration.
For engineers with verifiable LLM-product shipping experience, the AI talent market has pushed refresh numbers higher, but that leverage is time-limited. To use that leverage, engineers need to know the benchmark before talks begin. When comparing Series A and Series B roles, the stronger equity proposition is the offer whose future grants, funding, and vesting terms make the promised stake believable through year three and beyond.