Three of the four dominant US frontier AI labs restructured how they pay their most critical employees between January and June 2026 — not by raising salaries, but by changing the fundamental instrument through which equity is delivered. OpenAI converted from Profit Participation Units to standard RSUs. Anthropic's near-$1 trillion private valuation and a confidential IPO filing on June 1 turned RSU grants issued eighteen months ago into paper wealth that rivals listed-company stock. Google DeepMind bypassed band architecture entirely with a one-time "scale-of-impact" retention payment reaching $900,000 for senior research staff. Taken together, these three moves mark a structural break from the comp mechanics that dominated the 2024 and 2025 talent market.
The compensation race that will define H2 2026 is no longer about who can post the biggest number on a recruiting slide. It is about which equity instrument a top researcher can actually liquidate, and when.
What Happened
OpenAI's PPU-to-RSU conversion is the cleanest structural break of the half. When OpenAI completed its conversion to a Public Benefit Corporation in October 2025 and began operating as a for-profit entity in January 2026, the company retired its Profit Participation Unit model and moved new compensation packages to standard restricted stock units. The shift matters in two ways.
First, the tax treatment changed. PPUs were taxed as capital gains upon tender-offer sale. RSUs are taxed as ordinary income at vest — a meaningful difference for engineers clearing $1M or more in annual total comp. Second, and more important for the recruiting market, the PPU's cap structure disappeared. PPUs were contractually bounded at a 10x return on original value, per publicly analyzed OpenAI PPU tender offer documentation reviewed by Levels.fyi and compensation advisors. RSUs carry no ceiling. For a senior researcher receiving a multi-year equity grant at OpenAI's current implied valuation — the company raised a $122 billion round in late March at an $852 billion post-money valuation — removing the 10x cap is a materially different offer than anything OpenAI could make twelve months ago.
Applications Chief Fidji Simo, who in December 2025 announced the elimination of OpenAI's equity vesting cliff (following a reduction from 12 months to 6 months in April 2025), framed the cliff removal as de-risking the decision to join: candidates no longer need to fear a dismissal or role mismatch before they access their first equity tranche. The combination — no cliff, no return cap, RSU instead of PPU — amounts to the most recruiter-legible comp package OpenAI has ever offered. The company has committed to spending approximately $6 billion on stock-based compensation in 2026, nearly half of its projected annual revenue, according to internal financial documents reviewed by the Wall Street Journal.
Anthropic's valuation trajectory changed the math on every existing employee's equity grant, and the IPO filing filed with the SEC on June 1 crystallized what that means for the next eighteen months. Anthropic's Series H round, closed in late May, raised $65 billion at a $965 billion post-money valuation — a figure that, if sustained, would place Anthropic at the top tier of S&P 500 companies on the day it lists. The valuation moved from $61.5 billion in March 2025 to $965 billion in May 2026: a 15.7x step-up in fourteen months. A senior research scientist at Anthropic who accepted a standard offer in early 2025 — $746,000 median total comp at the time, per Levels.fyi — received RSU grants priced off a $61.5 billion valuation. Those same RSUs, on paper, are now worth roughly 15.7 times the grant-date value. No salary increase of any magnitude produces that arithmetic.
The IPO filing changes the liquidity calculus directly. Private RSUs are illiquid. Post-IPO RSUs vest into public stock that can be sold. An Anthropic research scientist hired in late 2024 and holding a 4-year grant with a 1-year cliff is approaching their first vest window with a company that has filed to go public, potentially as early as October 2026. The expected liquidity event is now a scheduled calendar event rather than a theoretical future outcome. That is a different conversation in a recruiting context than any competing offer can produce.
Anthropic's current compensation floor for research scientists sits at $320,000 in total comp at the low end, with the median at $746,000 and the top of the disclosed band reaching $1.05 million, per Levels.fyi data updated as of June 1, 2026. Software engineers in the San Francisco Bay Area clear a median of $665,000. The 4-year vest with a 1-year cliff structure has not changed. What changed is the underlying asset's worth.
Google DeepMind's one-time retention mechanism is the structural outlier of the three. Where OpenAI and Anthropic reset their comp through instrument change and valuation step-up respectively, Google DeepMind's leadership — under Demis Hassabis — introduced a cash payment for senior research staff with multi-paper publication records or direct model-launch ownership. Reports from talent advisors tracking the program place the payment range at $400,000 to $900,000, paid against a 24-month retention cliff. The mechanism is not a band reset. It does not create permanent cost-base inflation. It does not reset the total comp band that Google carries on its books for the broader research organization.
The design choice reflects a deliberate constraint. Google's compensation architecture spans a company with over 180,000 employees. A structural band lift at the senior research level, if not isolated to a specific unit, creates internal equity pressure across Google's entire technical staff. The one-time bonus is the instrument that allows Google to match frontier-lab retention pricing for a narrow cohort of high-leverage researchers without triggering a company-wide comp recalibration. It is, in the language of compensation design, a precision tool used where a broad-spectrum approach would be too expensive.
The practical consequence for the talent market is that Google DeepMind senior researchers who are in their 24-month retention window are being paid to stay at a rate that was not available to them before Q1 2026. Senior researchers who are outside the retention window, or who are considering offers from other labs, are facing a Google offer without the one-time component. That asymmetry is a known variable in the current recruiting cycle.
Why It Matters
The three mechanisms are complementary in what they signal about the second half of the year. Frontier labs have exhausted the easy levers. Headline base salaries for senior research staff have been at $350,000–$450,000 for over a year, and moving them further creates internal equity pressure and recruiting-narrative complexity. Signing bonuses in the $500,000–$1.5 million range — the xAI model for senior hires — buy a body for 24 months under a clawback agreement. What labs are now competing on is the structural quality of their equity position.
The bifurcation in the senior-IC market is now operating at two levels simultaneously. The first is between frontier labs and everything else. Software engineer median total comp at OpenAI sits at $555,000; at Anthropic, $665,000 for Bay Area engineers; at xAI, $640,000 in the US broadly. Google DeepMind's reported range runs $401,000–$600,000. Against a Big Tech non-AI-designated median that remains well below $300,000, the frontier premium has not compressed — it has widened.
The second bifurcation is within frontier labs, between research scientists and engineers. Research scientists at OpenAI clear a median of $771,000 to $1.47 million, depending on level and publication record. Research engineers at the same lab earn $249,000 to $530,000 for comparable IC seniority. The compensation logic inside frontier labs is increasingly explicit: a researcher who can demonstrate direct contribution to a model breakthrough is being compensated at a rate that reflects the output's commercial value, not the labor market's supply-demand curve.
Equity now represents 55–70 percent of total comp at the senior end of the frontier-lab market, up from 35–45 percent in 2024, per ENTRA Salary Survey H1 2026 and Levels.fyi senior-tier benchmarks. The implication is that the real comp story of H1 2026 is not in base salaries — it is in who holds which equity instrument, at what valuation, vesting on what schedule, with what liquidity path.
What's Next
Three dynamics will define the frontier comp market through December 2026.
First, Anthropic's IPO timing will function as a recruiting catalyst. If the offering proceeds on the October 2026 timeline being discussed, Anthropic will enter the fall recruiting season as a near-public company. That gives its recruiters an argument they have never previously been able to make: we have a listed date. For candidates who have been holding off on Anthropic offers because of private-equity illiquidity concerns, a visible IPO window removes the primary objection. Expect Anthropic's offer acceptance rate to improve materially if a public filing timeline firms in Q3.
Second, OpenAI's RSU transition will clarify its retention story within the next two vesting cycles. The first cohort of employees receiving pure RSU grants (rather than PPU-based offers) will hit their 12-month marks in late Q1 2027. How those grants perform relative to the $852 billion valuation anchor will determine whether the instrument change was a net positive for the employees who took it. Any IPO signal from OpenAI — which has its own public-market speculation running parallel to Anthropic's — will accelerate that resolution.
Third, Google DeepMind's one-time retention program has a 24-month sunset. The researchers who accepted it in Q1–Q2 2026 are locked through Q1–Q2 2028. If Anthropic's IPO and OpenAI's RSU conversion produce significant liquidity events before that retention clock expires, Google will face a wave of senior research departures at a moment when its AI competitive position depends most on the continuity of its research teams. Hassabis's instrument was precise. Its durability depends on whether the frontier market produces a liquidity event before the retention cliff expires.
The H1 2026 comp reset at frontier labs is not a salary story. It is an equity instrument story, and the second half of the year will show whether the instruments that were structured in January can hold the talent they were designed to retain.