AI engineer salary negotiation: the offer, the equity, and what to say

You cannot negotiate against a number you do not know. The US market for AI and ML engineers runs roughly $150k–220k base at mid-level and $220k–300k at senior, with New York offers typically 15–25% above national bands. Frontier-adjacent research roles and senior infrastructure engineers price above those ranges, because those companies compete directly with big-tech research and systems teams for the same people.

Before any conversation, fix three numbers in writing for yourself: the market band for your level and city (start with the salary guide and the NYC numbers), the number you would sign today without another word, and the number below which you walk. Every negotiation tactic in the world is a poor substitute for actually knowing these three.

And know what the company can do. A seed-stage startup cannot match a public company’s base — but it can move equity in ways a public company never will. Calibrating the ask to the company’s stage is half the skill.

AI companies raise at extreme valuations, which makes equity both the most valuable and the most misunderstood part of the offer. The right questions depend entirely on what kind of company is across the table. Our NYC AI 100 tracks every company’s latest round and valuation with dates — use it to know exactly which conversation you are walking into.

Public companies (CoreWeave, UiPath, Via, Pagaya, Schrödinger).You get RSUs with a visible price and real liquidity — shares you can sell as they vest. The negotiation is simply about the grant’s size and vesting schedule. This is the cleanest equity there is; value it near face value.

The mega-privates ($10B and up: Ramp at $44B, Cyera at $12B, ElevenLabs at $11B).Real companies with real revenue — and illiquid stock. Three questions matter: What is the refresh policy, since your initial grant is struck against an already-huge valuation? Has the company run tender offers or allowed secondary sales, and how often? And what growth thesis justifies upside from here — Ramp’s valuation nearly tripled in the year to June 2026, which is exactly the argument a recruiter will make, so decide in advance whether you believe it can continue.

The unicorns ($1B–$10B).This is where the 409A gap matters most: your options are struck against the common stock’s appraised value, typically far below the preferred price investors paid, and that gap is your built-in upside — ask for both numbers. But also ask when the valuation was set. A 2026 round is information; a 2021 round is a warning. Attentive’s last priced round said $7B in mid-2021, and secondary-market activity since implies a materially lower value today — engineers who joined on the headline number learned the difference the hard way. Liquidity varies wildly too: Clay ran two employee tender offers in nine months at rising valuations, while most unicorns have never run one.

Growth and emerging companies (roughly Series A–B).Here percentages beat share counts. Ten thousand shares means nothing without the denominator; 0.1% of a company you believe will be worth $2B means something. Ask for the fully diluted share count, the size of the option pool, and the post-money of the last round — a company that won’t share them is telling you how it treats employee shareholders.

None of this is tax advice — exercise strategy, ISO/NSO treatment, and 83(b) elections are worth an hour with an actual tax professional before you sign, not after.

Naming a number first, early, and specifically.The recruiter’s screening question is designed to cap you before the company has decided how much it wants you. Answer with the band.

Valuing private equity at its headline number. A $2M grant at a private company is not $2M — it is a claim on a future liquidity event, filtered through dilution, preferences, and time. Discount accordingly, and negotiate base like the equity might go to zero.

Folding at the first “this is our best offer.” It often is not — and the polite test costs nothing: “I understand. If there’s flexibility anywhere — equity, sign-on, review timing — this is the one I want to sign.”

Panicking at exploding offers. A 48-hour deadline is pressure, not policy. Asking for a week to make a considered decision is professional, and companies grant it far more often than candidates believe.

Treating negotiation as adversarial. Done right, it is the first working session with your future manager: evidence, market data, a clear ask, and a fast yes once terms are met. Companies remember candidates who negotiated well — it reads as exactly the judgment they hired you for.