Five years into a DevOps or SRE career, the salary question stops being "what does this job pay" and becomes "which version of this job am I in." In the US in 2026, a reliability or platform engineer at roughly five years of experience earns somewhere between $135,000 and $430,000 in total compensation. That is not a sloppy range. It is the actual shape of the market, and nearly all of the spread is explained by the type of company writing the check rather than by the depth of your Kubernetes knowledge.
The rest of this guide breaks that range apart: by employer tier, by metro, by the split between base, equity, and bonus, and by the handful of variables that genuinely move the number. It also covers something a salary table cannot: on-call, which is real compensation paid in sleep rather than dollars.
The headline numbers by employer tier
These are US figures for roughly five years of experience, expressed as annualized total compensation.
| Employer tier | Base | Equity (per year) | Annual bonus | Total comp |
|---|---|---|---|---|
| Large public tech / FAANG SRE | $175K–$225K | $110K–$220K | $25K–$45K | $290K–$430K |
| Mid-size public tech | $150K–$185K | $20K–$55K | $12K–$28K | $185K–$265K |
| Venture-backed startup | $150K–$190K | 0.05%–0.25%, illiquid | Rare | $150K–$190K cash |
| Non-tech / enterprise IT | $125K–$165K | ~none | $10K–$20K | $135K–$185K |
Two immediate observations. First, the top tier pays more in equity alone than the bottom tier pays in total. Second, the startup row is deliberately labelled cash, because the equity line there is a lottery ticket denominated in a percentage you should expect to be diluted 50-70% before any liquidity event. That dilution figure is one of the firmer claims in this entire post.
The FAANG band above is consistent with our detailed software engineer salary tables, which put five-year engineers at Google L4 around $270K-$380K and L5 at $380K-$550K, Meta E4 at $260K-$370K, Amazon SDE II at $220K-$340K, and Microsoft levels 61-62 at $210K-$300K. The SRE band sits inside that envelope, which is exactly the point of the next section.
The correction that matters: at big tech, SRE is paid at par
If you take one thing from this page, take this: at big tech, DevOps and SRE are at par with software engineering. Google SRE is not a parallel, lower-paid track — SREs sit on the same engineering ladder, are calibrated against the same level definitions, and are paid on the same scale as SWEs at the equivalent level. The same is broadly true of infrastructure and production engineering roles at Meta, Amazon, Apple, and Microsoft.
So why does every aggregate you read make DevOps look like a discount role? Because the aggregate is not measuring the same population. The title "DevOps engineer" is concentrated among employers that pay no equity at all: banks, insurers, hospital systems, retailers, telecoms, and government contractors. Product-tech companies mostly do not use that title; they hire SREs, production engineers, infrastructure engineers, and platform engineers. When a market-wide median blends those two populations, it is averaging a compensation structure that includes six-figure RSU vesting with one that has no equity line whatsoever. The result is a lower median that says something true about the employer mix and nothing at all about the skill.
That distinction is the most useful thing a five-year SRE can learn about their own market, because it is actionable in a way that "study more Terraform" is not. Moving from a regional insurer's platform team to a public tech company's SRE org, at the same skill level, is worth more than any certification.
A note on two numbers you will see quoted
Job-posting datasets — including Recruiting from Scratch's analysis of roughly 1.9 million postings — put the market-wide median base for DevOps and platform roles near $180K. Separately, self-reported total-compensation datasets put the market-wide median total comp for the same title family around $150K-$175K. Those two numbers are not comparable, and you should never treat one as following from the other. Postings skew toward companies actively hiring at scale and toward the higher end of the market; self-reported total comp draws from a different population entirely. The apparent paradox — median total comp below median base — is a sampling artifact between two datasets, not a real market condition.
Base, equity, and bonus: what the split actually looks like
Salary pages tend to publish base and total, leaving bonus as an unstated residual. It is worth naming explicitly.
- Base is the least variable component and the one you have the least leverage over. Companies run tight bands per level per location, and recruiters genuinely cannot move base far without a level change.
- Equity is where the tier gap lives. At the top tier, four-year RSU grants of $440K-$880K are ordinary at five years of experience; at a mid-size public company the same experience might carry a $80K-$220K grant. Both are real money; they are not the same order of magnitude.
- Bonus at large tech employers runs roughly 10-20% of base at target. Google's L4 numbers are a clean illustration: base plus equity of $235K-$335K against a total of $270K-$380K implies something in the $35K-$45K range, roughly a 15% target.
Amazon is the structural exception. It pays no meaningful annual performance bonus. That value is folded into back-loaded RSU vesting — the well-known 5/15/40/40 shape — plus year-one and year-two signing bonuses that exist specifically to smooth the first two years. If you are comparing an Amazon offer against a Google or Microsoft offer using base plus bonus, you will misread it badly. Compare year-by-year total, not steady-state total.
Metro adjustments
Location bands have loosened since 2020 but have not disappeared. Where they still exist, the spread between top-tier metros and tier-2 cities runs roughly 10-25% on base, with equity often adjusted less aggressively or not at all.
| Metro | Typical adjustment vs national tech band | Notes |
|---|---|---|
| SF Bay Area | +10% to +20% | Highest bands; also the deepest competing-offer market |
| New York City | +8% to +18% | Finance-adjacent infra roles pull base up, equity down |
| Seattle | +5% to +15% | No state income tax changes take-home materially |
| Austin, Denver, Boston | 0% to +8% | Often the same band as national for remote-first firms |
| Tier-2 US metros | −10% to −20% | Where location bands are still enforced |
| Fully remote, national band | Varies | Some firms pay one national band; others geo-adjust annually |
Treat these as directional. Company policy varies more than city cost-of-living does, and several large employers have collapsed multiple tiers into a single national band since 2023.
On-call: compensation that a salary table cannot show
Two offers with identical total comp are not identical jobs if one carries a one-week-in-four primary rotation on a service that pages four times a night. This is the dimension a salary page structurally cannot capture, and it is the one most worth interrogating.
What to ask before signing, in order:
- Rotation frequency. One week in four is common; one in six or one in eight is comfortable; one in two is a staffing problem being handed to you.
- Pages per shift, median and p95. A team that cannot answer this from data is telling you something about its observability maturity.
- Is on-call compensated separately? At most large US tech employers, primary on-call is unpaid and considered priced into base and equity. Separate stipends and time-off-in-lieu are considerably more common at non-tech enterprises and at European employers, where they are sometimes required. Where a stipend exists it is typically a modest per-week amount for the primary rotation — ask for the actual figure rather than trusting any published benchmark, including this one.
- Who owns the runbooks? If the answer is "the on-call figures it out," your effective hourly rate is lower than the offer letter implies.
The title spread
Within the same company, and sometimes within the same org, the title on your offer letter changes your band:
- SRE / production engineer / infrastructure engineer — at product-tech firms, on the SWE ladder, at the SWE scale.
- Platform engineer — usually on or near the SWE ladder at tech companies; frequently a separate, lower IT band at non-tech employers.
- DevOps engineer — the title most concentrated in non-tech employers, and therefore the title most responsible for the depressed market-wide median.
If you are interviewing at a company that offers more than one of these, ask which ladder the req sits on before you accept the screen. It is a cheap question with a five-figure answer.
What moves your number at five years, ranked
1. Level, not years. Five years of experience straddles the mid/senior boundary at nearly every company. Down-levelling is the single most expensive event in a five-year career: you lose the entire band difference today, and because the next employer benchmarks against your current level and comp, it compounds into the offer after that. Argue level first, always, and argue it before a recruiter has written a band into an internal system.
The level argument is won in the loop, not in the email thread. The clearest signal I got about my own calibration came from a PhantomCodeAI mock where the interviewer pushed on blast-radius estimation twice and I gave a mid-level answer both times — scoped to my service, not to the dependency graph around it. That is the exact distinction between mid and senior on most SRE rubrics, and it is cheaper to discover in a mock than in a debrief.
2. Employer tier. Our own tables show roughly $100K of spread between Microsoft 61-62 and Meta E4 at nominally equal seniority. For DevOps the tier gap is wider still, because the population spans tech and non-tech employers rather than sitting entirely inside tech.
3. Equity refreshers and the year-four cliff. Refreshers typically run 25-50% of the initial grant, each on its own multi-year schedule. The arithmetic consequence is that total comp frequently falls in year five even with flat, solid performance, because the initial grant has finished vesting and refreshers have not stacked high enough to replace it. Microsoft-style refreshers vesting over five years make the dip through years two to four more pronounced. Sequoia's 2025 work on equity refresh trends is a useful primer here. Also note that dollar-denominated grants paid in shares cut both ways: a falling stock means more shares next cycle, a rising stock means fewer.
4. Location. 10-25%, as above, and only where bands are still enforced.
5. Competing offers. They are the only real negotiation lever. Everything else is framing.
Negotiating with actual leverage
The mechanics differ from a standard SWE negotiation in one respect worth naming: your leverage is disproportionately about level and scope of production ownership, because those are what SRE bands key off.
- Sequence matters. Level, then number. Once a level is set, the band is a fact and you are haggling inside someone else's constraint.
- Where the money moves. A competing offer typically produces a 10-30% uplift, and it lands on the sign-on bonus and initial equity grant, not on base. Base is band-locked; equity and sign-on are the discretionary levers a recruiter actually controls. That range is a rule of thumb from practitioner reports rather than a measured statistic — treat it as an expectation, not a promise.
- Bring evidence, not feelings. levels.fyi for level-specific bands, Ravio's Compensation Trends 2026 for European and cross-market benchmarks, Glassdoor for the non-tech employers the tech-focused datasets undersample.
- Name the on-call. "I'm carrying a one-in-three primary rotation" is a legitimate scope argument at the level discussion, where it belongs, not a dollar argument at the offer stage.
For the full mechanics of running competing offers, deadlines, and exploding-offer pressure, see our dedicated guide on how to negotiate software engineer salary offers.
None of this leverage exists if you do not clear the loop at the level you are arguing for, and the SRE loop is where most of these arguments are actually won or lost. When I was preparing for a platform-team loop last year, the thing that surprised me was not the Linux depth — it was how much of the incident round is a communication exercise. I ran three mock loops on PhantomCodeAI in the week before, and the pattern I kept fumbling was jumping to root cause while the interviewer was still waiting to hear a mitigation plan. Seeing that on a transcript twice was more useful than a fourth pass through a networking cheat sheet.
How firm are these numbers?
Stated flatly, with confidence:
- FAANG level bands and the fact that SRE sits on the SWE ladder at big tech.
- Refresher mechanics and the year-four cliff.
- Startup dilution of 50-70% before liquidity.
Stated with moderate confidence:
- Mid-size public and startup bands. Fewer public data points, wider real variance.
- The at-par claim. Well supported, but phrased precisely: at big tech, at par. It does not generalize to every employer.
- The 10-30% competing-offer uplift.
Stated with explicit hedging:
- Non-tech enterprise bands. These are the least visible segment in every public dataset, and the ranges above are estimates assembled from postings and self-reports rather than measurements.
- Any 2026 trend direction. A single year of movement is not a trend, and we are not going to pretend otherwise.
Where to go next
If your read on the tier table is that you are underpaid because of employer mix rather than skill, the fix is an interview loop, not a raise conversation. The DevOps and SRE interview guide covers the loop structure end to end — Linux troubleshooting, networking, incident drills, IaC, and the on-call maturity signals that decide senior-versus-mid calibration. The DevOps engineer question bank is the drilling companion to it.
What I did after mapping my own tier gap was simple: booked the loops first, then ran mocks against the incident and design rounds until the mitigation-before-diagnosis habit stuck. The comp conversation was the easy part once there were two offers on the table.