AI Research Scientists Are Winning the Equity War. Here Is What the New Pay Data Means for Engineers.

# AI Research Scientists Are Winning the Equity War. Here Is What the New Pay Data Means for Engineers.

Published: August 26, 2026

The AI compensation market has stopped behaving like one market.

That is the practical conclusion from Pave and Nua Group's new State of AI Talent report, released August 24. The report analyzes compensation data from more than 9,000 companies, including 80% of the Forbes AI 50, and separates three job families that are often collapsed into one label: AI Engineering, ML Engineering, and AI Research Scientist. (Pave and Nua Group report)

The headline is not that every AI job suddenly pays more. It is that the biggest premium is increasingly attached to the point of hire and to equity, especially for research roles. For candidates, that changes how you compare offers. For employers, it changes what a defensible job architecture looks like.

Base salary is only telling half the story

Pave's data shows relatively compressed base pay across the three families at senior levels. At the report's P6 and M6 levels, ML Engineering median base pay is listed at $321,000 and $347,000, respectively. That is already a high market, but it does not explain where the sharpest competition is happening.

The larger separation appears in new-hire equity. Median new-hire equity for AI Research Scientists reaches $4.09 million at P6 and $4.72 million at M6, more than double the comparable AI Engineering and ML Engineering packages at those levels. At the 90th percentile, the report says senior research-scientist grants at private Bay Area companies that have raised $1 billion to $5 billion can reach $45 million to $60 million.

Those are not ordinary salary benchmarks. They are a signal about how companies are pricing scarce frontier-model research capacity: as a strategic asset whose upside is tied to company value, not just annual cash compensation.

That distinction matters when a candidate is comparing a comfortable cash-heavy offer with a lower-base offer carrying meaningful equity. The answer is not “always take equity.” The answer is to model the two components separately, ask what the equity actually represents, and understand the company's financing, vesting, dilution, and liquidity assumptions before treating a headline number as compensation.

AI Engineering is the growth family

The report also points to a different kind of opportunity. AI Engineering's share of all employees in the dataset grew from 0.010% in Q4 2023 to 0.185% in Q2 2026. Its share of new hires rose from 0.11% to 0.35% over the same period.

The important point is not just the size of those percentages. It is the direction and the job-design implication. AI Engineering is becoming a recognizable operating function: the people who turn models into reliable product capabilities, internal systems, evaluation loops, and customer workflows.

That is a different value proposition from training a frontier model. It also means the candidate pool is broader. An engineer who can combine software delivery, model evaluation, data pipelines, security controls, and production ownership can compete for AI Engineering roles without presenting as a research scientist.

The Vibe Coding Ebook covers the developer tooling and workflow layer behind this shift. The AgenticNode material goes deeper on orchestration and the systems that connect models to tools. Both are useful reference points for building a portfolio that shows more than a model demo.

The market is also rewarding recruiting leverage

The Pave report is about compensation architecture, not recruiting operations, but a second August report helps explain why the surrounding hiring work is changing. iCIMS and Lighthouse Research surveyed more than 460 talent-acquisition professionals in high-volume industries and reported that 75% say AI reduces recruiter workload. The areas named include resume screening, candidate sourcing, interview scheduling, and candidate communication. (iCIMS and Lighthouse Research)

That does not mean recruiting disappears. It means the work shifts toward quality of hire, workflow design, candidate judgment, and the governance of automated decisions. The same pattern is visible in engineering: routine production steps become cheaper to accelerate, while the ability to define the problem, inspect the output, and own the consequences becomes more valuable.

For candidates, the lesson is to show the layer above generation. A portfolio project should explain:

• what the system was asked to do and what it was not allowed to do;
• how outputs were tested or evaluated;
• where a human had to review or approve the result;
• what happened when the first output was wrong; and
• how cost, privacy, latency, and reliability affected the design.

That evidence maps to the AI Engineering family much better than a list of model names.

A better way to read an AI offer

Before accepting an offer, classify the role. Is it primarily:

1. Research: new model capability, training, alignment, or fundamental evaluation;

2. ML Engineering: training infrastructure, data systems, serving, and model operations; or

3. AI Engineering: product integration, agent workflows, evaluation, reliability, and customer-facing delivery?

The labels are not universal, but the distinction is useful. The report's central finding is that these families now have different pay curves, seniority mixes, hiring trajectories, and retention dynamics. Treating them as interchangeable makes salary research less accurate and job searches less focused.

Ask the hiring team which outcomes define success in the first six months. Ask how much of the role is research, platform, integration, or operational ownership. Ask whether equity is based on a current 409A valuation, a preferred financing price, or an internal estimate. Ask how the company expects the role to change as models improve.

Those questions turn a vague “AI role” into an opportunity you can actually evaluate.

The career signal

Pave and Nua's report does not say that research is the only winning path. It says the market has become mature enough to price different kinds of AI work differently. Research scientists command the largest equity premiums because their capabilities are scarce and their impact can compound across a company's model platform. AI Engineering is growing because organizations need people who can make those capabilities useful, safe, and dependable in real systems.

The strongest career strategy is therefore not to chase a title. It is to build a defensible combination of skills: technical fundamentals, model literacy, evaluation discipline, security awareness, and the ability to ship inside a real operating environment. That combination is portable across all three families—and it makes your contribution legible when the offer is being priced.

Related: Stanford Finds a 19% AI Hiring Gap for Young Workers · The AI Skills Wage Premium Just Doubled · The MCP Economy

For the workflows behind AI-native engineering teams, see the Vibe Coding Ebook and AgenticNode.