Why ML Compensation Is Different
ML engineering compensation is higher than general software engineering — and more variable. The variance comes from three sources: company type, specialization, and how well you negotiate.
Understanding all three is the difference between being in the top quartile of your level and the bottom quartile. The gap is often $50–150k in total compensation at senior levels, and it rarely closes on its own.
The Compensation Stack
Most tech companies pay ML engineers in four components:
Base salary: The predictable part. Paid in regular payroll cycles. Ranges from ~$150k at mid-size companies to ~$250k at top-tier big tech for senior roles.
Annual bonus: Usually 10–20% of base for target performance. Variable based on individual and company performance. Less common at startups.
Equity (RSUs or options): The largest component at public companies. At big tech, a senior ML engineer's annual equity refreshes can dwarf their base salary. At startups, options with uncertain value.
Signing bonus: One-time payment at joining. Often used to offset unvested equity being left behind. Ranges from $20k at mid-size to $100k+ at top companies.
Total Compensation (TC) is what matters. A $200k base at a company with no equity can pay less than a $170k base at a company with $200k/year in RSUs.
Compensation by Level
The numbers below are rough medians for top-tier US tech companies (FAANG+, Stripe, Snowflake, etc.) as of 2025. Actual ranges are wide — these are useful for calibration, not precision.
L3 / Junior ML Engineer (0–2 years experience)
- Base: $150–180k
- Equity (annual): $50–100k
- Bonus: $15–25k
- Total: $215–305k
New grads at top-tier companies tend to land at the bottom of this range. Strong candidates with prior internships or research publications land higher.
L4 / ML Engineer (2–5 years experience)
- Base: $175–220k
- Equity (annual): $100–200k
- Bonus: $20–40k
- Total: $295–460k
This is where the variance starts to matter. L4s at Google and Meta earn meaningfully more than L4s at mid-size companies in the same city.
L5 / Senior ML Engineer (5–10 years experience)
- Base: $210–260k
- Equity (annual): $200–400k
- Bonus: $30–60k
- Total: $440–720k
The senior-to-staff gap is large at some companies and small at others. Google's L5→L6 gap is significant; at smaller companies, the Staff title often doesn't come with a proportional compensation jump.
L6 / Staff ML Engineer (8+ years)
- Base: $240–300k
- Equity (annual): $400–800k
- Bonus: $50–90k
- Total: $690k–1.19M
Staff is where equity dominates compensation. At public companies with strong stock performance, top staff engineers routinely exceed $1M TC.
How Company Type Affects Compensation
Big Tech (Google, Meta, Amazon, Apple, Microsoft)
- Highest TC at all levels
- Strong equity component (vested RSUs in liquid stock)
- Structured leveling with consistent bands
- Annual refreshes keep equity competitive after initial grants
- Downside: slower equity growth at mega-cap companies vs. high-growth companies
High-Growth Public (Stripe, Snowflake, Databricks, OpenAI)
- Competitive with big tech at senior levels
- Higher equity upside if stock continues to grow
- Less predictable because stock price variance is higher
- Some (especially pre-IPO) pay below-market base to compensate for option upside
Mid-Size Tech (series B–D startups, companies with 500–5000 employees)
- Base salaries often 80–90% of big tech
- Equity is a meaningful component but less liquid
- Total compensation is typically 60–80% of big tech equivalent level
- Tradeoff: more ownership, faster career growth, but lower expected TC
Early-Stage Startups (pre-series B)
- Base often at or below mid-size tech
- Large option grants (0.1–1%+ of the company)
- Expected value of options is highly uncertain
- Most appropriate when you believe in the company specifically, not as a compensation optimization
Research Labs (DeepMind, FAIR, MSR, Anthropic)
- Compensation varies widely
- Top researchers at FAIR or Anthropic earn big-tech-comparable TC
- Many research roles are below-market on base but include research publication time
- The career optionality from a research publication record can be worth more than near-term TC
How Specialization Affects Pay
Not all ML engineering specializations pay equally. In practice, the market in 2025 prices certain skills significantly higher.
Highest-Demand Specializations (premium over median)
LLM / GenAI Infrastructure: Training and serving large models at scale is a specialized skill that commands a 10–20% premium over general ML engineering. Very few people have experience operating 70B+ parameter model training runs or low-latency inference infrastructure.
ML Platform / Infrastructure: Building the tools that ML teams use (feature stores, training platforms, serving infrastructure) is increasingly recognized as a distinct, highly-leveraged specialization. Platform engineers at companies like Airbnb, Uber, and Lyft earn at the top of their bands.
Recommendation Systems at Scale: Deep experience with production recommendation systems (retrieval, ranking, real-time personalization) at large-scale companies is consistently one of the highest-paid ML specializations outside research.
Well-Compensated Specializations (at-median to slight premium)
Fraud / Trust & Safety ML: High-stakes ML with adversarial dynamics. Specialized knowledge with a smaller candidate pool.
Computer Vision (production): Applied CV for production (not research) is well-compensated, though the pool of companies is smaller than NLP/recommendation.
NLP / Search: Strong demand. Large pool of candidates keeps compensation at median for most practitioners.
Below-Median Specializations
General Data Science: Roles focused on analysis and reporting rather than production ML tend to pay 10–20% below ML engineering at equivalent experience levels.
Classical ML / Tabular Models: Important work, but the candidate pool is large and the technical premium over general software engineering is lower.
Negotiation Strategy
The Most Important Rule: Always Negotiate
The single most impactful thing you can do. Fewer than half of candidates negotiate their initial offer. Of those who do, nearly all get an improvement. Companies expect negotiation — the first offer is rarely the best offer.
The fear: "They'll rescind the offer if I negotiate." This essentially never happens for good-faith negotiations. The risk is vastly smaller than the upside.
Know Your Leverage
Your leverage is highest when:
- You have competing offers (the strongest possible leverage)
- You're currently employed (you're not desperate)
- The company has been trying to hire for this role for months
- Your skills are specialized and hard to find
Your leverage is lower when:
- You have no competing offers and the company knows it
- You've already communicated enthusiasm without anchoring on compensation
- The role has been open for less than two weeks
The Competing Offer Approach
The most effective negotiation: get competing offers. Not to accept, but to create leverage at your target company.
Practical approach:
- Apply to 3–5 companies you'd genuinely consider
- Try to get offers on similar timelines
- When your target company makes an offer, honestly share that you have competing offers
- Ask: "Is there flexibility to improve the offer? I'm very interested in this role, but I'm trying to make the right decision."
This works because companies respond to market data. "I have a competing offer at X" tells them the market price for your skills. "I feel I deserve more" does not.
What to Negotiate
Equity is the highest-leverage item. A $20k increase in base is $20k/year. An extra 10% in your equity grant could be worth $30k–100k per year depending on the company's growth. Ask for more equity before asking for more base.
Signing bonus is the easiest to get. Companies often have more flexibility on one-time payments than on base or equity, because bonuses don't set a precedent for the band.
Level is the highest-value negotiation. Coming in as L5 instead of L4 is worth far more than any adjustment within a band. If you believe you're being leveled one step below where you should be, make the case explicitly. This requires evidence — not just seniority, but scope of prior impact.
What Not to Do
Don't anchor low first: If asked for your target range before an offer is made, deflect. "I'm focused on finding the right fit — I'd want to understand the full scope and expectations before naming a number."
Don't accept on the spot: Ask for time to consider even if you're excited. Standard practice is 24–72 hours, sometimes a week for senior roles.
Don't reveal your current compensation: In most US states, employers are prohibited from requiring this information. Whether or not it's required, your current comp is rarely relevant to your market value.
Don't negotiate just the base: Total compensation is what matters. A company that can't move on base may have significant flexibility on equity or signing bonus.
Timing: When to Optimize for Compensation
Early in your career (L3–L4), the highest-leverage thing is learning velocity, not starting salary. A role that teaches you production ML at scale is worth more than a $20k premium at a company where you'll maintain a single model.
At senior levels (L5+), compensation gaps between companies widen significantly and compound through refreshes. Once you have demonstrated senior scope, optimizing for company and compensation is rational.
The inflection point for most engineers: once you can make a credible case for L5 performance, the market pays significantly more than your current company will promote you to. That's when it makes sense to go to market.
For the career progression side of this picture, read our guide on How to Become a Senior ML Engineer. For understanding which roles and companies to target, see our ML career transition guide.