Frontier lab comp decoded: PPUs, RSUs, and the liquidity thesis

Frontier lab PM comp is not a salary negotiation. It is a liquidity thesis negotiation. The base is the floor you live on; the equity is a bet on whether and when a private company creates a window for you to convert paper into cash. A PM who understands PPU caps, double-trigger RSU risk, and tender offer mechanics can negotiate $200K or more in expected value than one anchoring on the headline number. Sign before you understand the instrument and you may hold paper for years.

What the numbers actually look like

OpenAI PM total comp at L5 (San Francisco) sits at a median of $860K per Levels.fyi as of June 2026: roughly $310K base and $550K annualized equity. The full range runs $300K to $950K+. Anthropic does not publish PM bands directly. Engineering TC runs $300K to $759K with a median around $443K, and PM comp tends to land 20 to 40% below that, putting a senior Anthropic PM at approximately $260K to $550K in total comp.

Those headline numbers mean nothing without understanding what the equity instrument is, when it becomes taxable, and what could prevent it from becoming cash at all.

OpenAI PPUs: no cliff, but there is a cap

OpenAI has historically issued Profit Participation Units rather than RSUs. The key mechanics for a PM evaluating an offer:

• No cliff. PPUs vest at 25% per year from day one. You start earning immediately, unlike the standard cliff-plus-monthly structure at most tech companies.
• 10x profit cap. A $2M grant maxes out at $20M regardless of how high OpenAI’s valuation goes. If the company is worth 50x your grant value, you do not capture that upside.
• Liquidity via tender offers. OpenAI runs periodic, discretionary tender offers. New hires face a roughly two-year lockup before they can participate. There is no guaranteed window.

As of January 2026, OpenAI is reportedly transitioning new offers from PPUs to RSUs as part of its public-benefit corporation restructuring. If you have a current offer in hand, confirm which instrument you are receiving and ask whether the 10x cap still applies. The answer changes the expected-value math substantially.

Anthropic RSUs: double-trigger and IPO dependency

Anthropic issues RSUs with a double-trigger structure. An RSU does not settle (and is therefore not taxable) until two conditions are both met: time vesting (four years, 25% per year) and a qualifying liquidity event, either an IPO, an acquisition, or an approved tender offer.

The IPO risk is real. Anthropic is targeting a public offering as early as October 2026 with Goldman Sachs, JPMorgan, and Morgan Stanley in early discussions. If that slips, RSUs that are fully time-vested do not automatically become cash. They sit as paper until the second trigger fires.

In April 2026, Anthropic ran a tender offer priced at a $350B valuation with $5 to $6B available. Most employees declined to sell, betting on a higher IPO valuation. That is useful signal for a candidate: the people closest to the information chose to hold. But it also means the only recent liquidity window passed, and the next one depends on the IPO landing on schedule. Tender eligibility at Anthropic requires 12 or more months of tenure, a 20-business-day window, and uniform pricing across all participants. If you join after a tender closes, you are waiting for the next one.

xAI follows a similar private-lab tender model. Equity structure documentation is less public, so ask the same questions: tender cadence, lockup period, and the conditions required for any liquidity event.

Modeling a real offer: $300K base, $2M grant

A $2M grant at 25% per year means $500K in equity per year at grant value. At OpenAI, the actual per-share price at each tender (not the grant date value) determines realized value. At Anthropic, under the double-trigger structure, that $500K/year tranche only becomes cash when the liquidity event fires. If you join in Q3 2026 and the IPO lands in early 2027, your first year’s tranche becomes taxable at IPO pricing. If the IPO slips to 2028, you have two years of time-vested equity you cannot touch.

The tax trap most candidates miss. When equity settles at a tender or IPO, the gain is ordinary income, not capital gains. For a $600K settlement event, standard withholding runs at 22 to 37%. Your actual marginal liability is closer to 48% (federal plus California state). That gap is roughly $96K you owe at tax filing with no automatic withholding to cover it. Set aside cash for the bill before you spend the tender proceeds.

Which levers actually move in negotiation

Base salaries at frontier labs are band-constrained. A $10K to $20K base adjustment is possible; a $50K base move is not. Equity is where the negotiation lives.

A 20% increase on a $2M PPU or RSU grant adds $400K to total grant value, roughly $100K per year across four years. That is the number worth fighting for. The primary lever to get there is a competing offer from another frontier lab or a senior role at Meta FAIR, Google DeepMind, or a well-funded AI startup.

Sequencing that works: get offers from at least two labs before entering final negotiation with your preferred option. Disclose the competing offer number specifically, with instrument type, grant value, base, and vesting terms. Vague competing offer claims move nothing. Then ask about tender offer history before signing: how many tenders has the company run, what was the cadence, and what was the pricing relative to the last primary round. For Anthropic specifically, ask where the IPO process stands and what the double-trigger fallback is if the IPO does not close within 18 months of your start date.

The one check before signing

Confirm the instrument in writing. Ask whether your grant is PPUs, RSUs, or stock options, and ask for the vesting schedule, any cap on upside, and the conditions required for settlement. At OpenAI, with the reported transition to RSUs underway, verbal descriptions of “equity” from a recruiter may not match the actual grant documents. Read the grant agreement before you sign the offer letter.

In 2026, near-anything is feasible to build with AI. The scarce resource is viable economics, for the company and for you personally. The base is how you pay rent. The equity is a thesis about whether this company creates a liquidity window before you leave. Evaluate both on their own terms.

For salary context by level across the industry, see OpenAI PM salaries and Anthropic PM salaries. For context on negotiating equity over base, see negotiate equity, not base. For the broader viable/lovable lens that shapes how great AI PMs think, see proving viability and feasibility is free.