career · career
Google DeepMind PM salary
Google DeepMind PM total comp sits 15 to 25% below same-level engineers at DeepMind, and meaningfully below frontier-lab counterparts at OpenAI and Anthropic. The base salary range cited in active 2026 hiring posts is $183K to $320K depending on level and team, with a vesting schedule that differs from standard Google: 33% in year one, 33% in year two, 22% in year three, and 12% in year four. The equity is Alphabet GSUs, liquid on vesting, which is a real structural advantage over private-lab equity. The trade is lower headline TC in exchange for liquidity certainty.
PM comp by level: L4 to L7
No public source documents DeepMind PM comp by level specifically. Every existing page conflates PM data with research scientist or software engineer data. The table below is built from offer data, Levels.fyi aggregate Google PM figures, and the standard 15 to 25% PM-to-SWE discount applied to published DeepMind engineer bands:
| Level | Base salary | Total comp (Mountain View) | Total comp (London) |
|---|---|---|---|
| L4 PM | $183K - $215K | $280K - $360K | $190K - $240K |
| L5 PM | $220K - $265K | $340K - $440K | $230K - $295K |
| L6 PM | $260K - $310K | $430K - $570K | $285K - $375K |
| L7 PM | $300K - $360K | $540K - $720K | $355K - $475K |
London compensation runs 30 to 40% below Bay Area in USD terms. L5 research scientists in London earn roughly £210K to £290K total comp versus $475K to $625K in Mountain View; PM bands track the same gap. This is the largest structural disadvantage for candidates based at DeepMind’s London headquarters, where a significant portion of the core research work happens.
The annual performance bonus adds 10 to 20% of base at level. Signing bonuses run $30K to $150K or more for senior hires, and spot bonuses of $1K to $10K are issued for notable contributions. The front-loaded vesting (33/33/22/12) means your first two years capture more than two-thirds of any new grant, which matters for your negotiating position at the renewal mark.
Does DeepMind pay a research-org premium for PMs?
For research scientists, yes: DeepMind equity grants run 5 to 15% above same-level SWE compensation, and Gemini team research scientists earn 10 to 20% above standard Google research rates. For PMs, the picture is more conditional.
A DeepMind core research PM (working alongside PhD researchers on fundamental model capabilities, not on a shipped product) does not automatically inherit the research premium. The premium reflects specialized skills that are genuinely scarce: the ability to evaluate model quality, read an eval, and make the call on whether a capability is production-ready. A PM who can translate research outputs into commercial bets with clear viability arguments, and who has the technical depth to engage directly with research scientists on what they are building, can negotiate for a premium. One who cannot demonstrate that judgment is treated as a product-org PM on a research team.
Gemini team PMs occupy a separate band: if you are hired as a PM on the Gemini App, Deep Research, Intelligence, or Personalization teams (all cited in active 2026 postings), you are closer to the elevated Gemini research bands than to DeepMind core research PM compensation. Expect TC at the upper half of the L5 or L6 ranges above.
GSUs vs. frontier lab equity: the liquidity question
DeepMind equity is Alphabet GSUs, publicly traded. When your grant vests, you can sell it that quarter. There is no lock-up, no double-trigger condition, no IPO timing risk, and no secondary-market haircut.
This is a genuine advantage over Anthropic’s double-trigger RSUs and OpenAI’s profit participation units. But the advantage has compressed in 2026. Alphabet stock has traded sideways while frontier lab private valuations have risen: OpenAI’s most recent round valued the company above $300B, and Anthropic’s Series E placed it at $61.5B. An L5 PM at OpenAI earns approximately $1.15M in total comp. An L5 engineer (not PM) at DeepMind earns $475K to $625K. The PM equivalent is lower still.
The candidate choosing between DeepMind and Anthropic at L5 should model the illiquidity discount carefully. A standard illiquidity discount for pre-IPO equity without a near-term exit runs 25 to 35%. Apply that to Anthropic’s L5 Senior PM equity and the expected annual value drops meaningfully. Whether the remaining gap favors Anthropic depends on your personal liquidity needs and your conviction in Anthropic’s exit multiple. Defaulting to “liquid is safer” without running the numbers is not a full analysis.
How the research-org structure affects leveling
This is absent from every other source on DeepMind PM comp, and it matters.
A PM at DeepMind does not own a roadmap of incrementally shipped features. They work alongside research scientists who are optimizing for capability advancement, not shipping velocity. Your scope is defined by: which research outputs are mature enough to turn into products, whether the compute cost is justifiable against the expected commercial return, and whether the capability clears the ethical risk bar.
That scope is large in influence and narrow in output volume. Google’s standard PM leveling criteria weight cross-functional leadership, shipping track record, and stakeholder management. At DeepMind, the criteria that actually move your leveling conversation are: research partnership depth (can you engage on model eval design, not just requirements gathering), viability judgment (did you correctly call what research was worth productizing), and influence on research direction. A PM who treats the role as a prestige posting without building those competencies will level behind their peers in the product org.
If you are coming from a standard Google PM role and interviewing internally for DeepMind, expect to be evaluated on your ability to assess research quality, not just your product execution track record. The interview includes an AI deep-dive round that has no equivalent in the standard Google PM loop.
The DeepMind PM interview is a separate pipeline
DeepMind runs its PM hiring independently of the standard Google PM pipeline. The process is less standardized, takes four to ten weeks, and includes four final-loop rounds. The distinctive addition is an AI deep-dive round: expect to discuss how you would evaluate a model capability, what constitutes production-readiness for a frontier system, and how you would handle ethical risk identification. Roles require seven to ten or more years of PM experience; a technical background is preferred, not optional.
Experience with ethical AI risk identification is explicitly cited in role requirements. This is not a checkbox. Interviewers test whether you can identify second-order risks in model deployments (not just the obvious failure modes) and whether you have a framework for how to weigh those risks against commercial timelines.
Negotiation: what actually moves at DeepMind
Standard Google negotiation tactics mostly apply, with one important exception: DeepMind’s partially separate pipeline means the comp committee has more discretion than a standard Google offer. You are not constrained by the same rigid band enforcement that applies to, say, a Maps PM offer.
What moves:
Competing offers from frontier labs. A documented OpenAI or Anthropic offer at your target level is the most effective lever. It shifts the conversation from “what band are you in” to “what does it take to close this candidate.” Given the TC gap between DeepMind and those orgs, a competing offer that documents the difference forces a specific response rather than a generic refresh.
Research-org premium argument. If you have AI-native work (eval design, model quality judgment, capability scoping for AI products), document it specifically. Make the case that your skills are closer to the research-scientist premium than the standard PM band. This argument works at DeepMind in a way it would not work in the product org, because the comp committee understands the scarcity of that skill set.
Level. The gap between L5 and L6 is $100K or more in annual TC. Level is nominally set in the loop, but documented scope from prior roles can move it. Evidence that you have operated at research-org PM scope (translated research outputs into shipped products, influenced research direction, made capability-readiness calls) is more relevant than volume of features shipped.
Signing bonus to bridge unvested equity. Bring your current vesting schedule with dollar values. DeepMind will bridge verified forfeiture. Size the ask to what you are actually leaving behind, with documentation.
Base compression toward the top of band is harder to move than equity grant size or signing bonus. Negotiate grant size and level first.
The 2026 framing
In 2026, the defining question for a DeepMind PM is viability. You are not shipping incremental features. You are deciding which frontier research outputs are worth the organizational cost to turn into products, and whether the market is large enough to justify the compute required to serve them at scale.
The research-org premium, where it exists for PMs, compensates for exactly that judgment. A PM who can read a model eval, estimate inference cost per query, and make the call on whether a capability is production-ready earns it. The comp gap with OpenAI and Anthropic is real and large. But DeepMind offers something the frontier labs cannot yet: liquid equity in a company that already generates substantial revenue, a research environment with access to some of the best scientists in the field, and a career path that sits at the intersection of Alphabet’s AI monetization strategy and frontier research. Whether that trade is worth it depends on where you think the most important AI products are going to be built in the next three years, and whether you need your equity to be liquid while you are there.
For the full equity comparison across frontier labs, see frontier lab comp decoded. For the negotiation framework, see negotiate equity, not base. For the Anthropic side of the comparison, see Anthropic PM salary.