ML Engineer Jobs in San Francisco: The 2026 Hiring Map

Machine Learning Engineer jobs in San Francisco are the densest ML market in the US. LinkedIn lists 9,000+ open roles in the Bay Area and Glassdoor counts 2,386 in San Francisco proper as of May 2026 (Source: LinkedIn Jobs). Total compensation runs from a $151K average on ZipRecruiter (Source: ZipRecruiter) to a $320K median total comp on Levels.fyi (Source: Levels.fyi), with $1M+ packages at frontier labs like OpenAI and Anthropic (Source: Levels.fyi — OpenAI). This piece is what those numbers actually mean for an ML candidate looking at SF in 2026.

The SF ML engineer market at a glance, May 2026

The SF ML market in one paragraph

We built Standout because the application-driven job search is broken for senior tech talent, and nowhere is that breakage more visible than the SF ML market. There are 9,000+ open ML roles in the Bay Area on LinkedIn and 2,341 AI engineer roles in San Francisco proper, but total SF tech listings sit around 4,000 across all categories, roughly half of the late-2019 and early-2022 peaks (Source: SF Standard). SF and San Mateo counties lost 4,400 tech jobs in 2025, a 0.4% contraction (Source: SF Standard). The headline is not "the market is booming." The heat is concentrated. Frontier labs and AI-first companies are hiring aggressively. The broader tech base is still working through a multi-year contraction. If you read "9,000 jobs" and assumed an open field, the denominator is fictional.

What ML engineers actually earn in San Francisco, by company tier

The $151,712 ZipRecruiter "average" (Source: ZipRecruiter) is a base-salary read across a broad listings sample. It under-prices the real market by a wide margin. Levels.fyi total comp data tells a different story: $320,000 median across SF (Source: Levels.fyi), and the spread is bimodal. Below is the band ML candidates should actually negotiate against, by company tier.

Concrete numbers per tier. Anthropic's median software engineer total comp is roughly $600K, with L3 mid-level around $450K and L4 senior around $665K (Source: Levels.fyi — Anthropic). OpenAI L5 software engineers earn approximately $1.15M total ($336K base + $774K stock) (Source: Levels.fyi — OpenAI). On the public-company side, Tesla ML roles list at $124K–$420K and Amazon staff-level ML at $255K–$345K (Source: Indeed). Built In SF comp ranges include General Motors $185K–$335K, Adyen $228K–$401K, Toast $230K–$368K, and BlackRock $270K–$350K (Source: Built In SF).

Hot take: if you are senior (5+ years) and an offer letter from a funded SF company reads below $230K total comp for an ML role in 2026, walk or negotiate. Below $200K is non-negotiable unless the equity is genuinely transformative. Most of the candidates we represent are calibrated against the $300K+ band, not the $151K aggregate.

Top employers hiring ML engineers in SF right now

The four tiers above each hire on a different cycle and through a different channel. Tier-by-tier, here is what the market actually looks like in May 2026.

Frontier labs. OpenAI and Anthropic are ramping headcount aggressively. SF Standard reporting indicates local listings for both companies have swelled significantly in 2026 (Source: SF Standard). These roles compress the comp range upward and the conversion rate downward. Frontier labs read a tiny fraction of cold inbound. Application is almost entirely a sourcing or referral problem.

FAANG ML. Apple, Google, and Meta run structured, slower-moving ML ladders in the Bay Area. Apple alone has a deep ML ladder with a $405K median in SF (Source: Levels.fyi — Apple). These teams hire continuously at a measured pace and reward FAANG-style interview loops. The application path here works better than at the labs because the interview pipeline is wider and more standardized.

AI-native scale-ups. Series B through Series D companies featured on Built In SF (Cohere Health, Deepgram, Tempus AI, CoreWeave) sit in the $250K–$450K band (Source: Built In SF) and are usually the highest risk-adjusted return for a mid-to-senior ML candidate. Roles are scoped narrower than at frontier labs but broader than at FAANG, and the equity grant at this stage can outperform a frontier-lab grant on a 4-year horizon for the right outcome.

Post-IPO infra and SaaS. Tesla, Adyen, Toast, BlackRock, Stripe, MongoDB. Total comp lands in the $185K–$420K window (Source: Built In SF / Indeed). These roles are either narrow scope (one production model) or platform engineering with ML adjacencies. Tesla alone has 10+ ML engineer openings across Palo Alto and Fremont (Source: Indeed), a real number but one worth filtering against the actual job description rather than the headline count.

If you mapped your last six months of applications across these four tiers, the response-rate pattern is almost always the same: the tier you spent the most time on is the tier where you got the least response. That is not bad luck. It is structure.

What hiring managers actually screen for in 2026

From the matches we have run with hiring managers at SF AI companies, the 2026 ML screen has consolidated around four signals. None of them is Leetcode-only.

1. 1Production ML at scale. One shipped product running in production beats five Kaggle medals. Hiring managers want to see model lifecycle work: training, evaluation, deployment, monitoring, and retraining. If the resume reads like "trained a model in a notebook," the screen ends there for staff and principal levels.

1. 1Strong Python plus at least one of PyTorch or JAX. TensorFlow alone is increasingly read as a 2018 signal. PyTorch is the modern default for research-flavored work. JAX is the signal at frontier labs and high-end scale-ups.

1. 1LLM and eval pipeline literacy. Even teams that are not building foundation models want candidates who can reason about evaluation rigorously. "Ran a benchmark" on a resume is not enough. "Built the eval harness, designed the rubric, and has an opinion about where standard benchmarks lie to you" is the screen.

1. 1One piece of public work. A paper, a meaningful open-source contribution, a launched product, or a public technical writeup that demonstrates judgment. Anthropic's own careers page makes it explicit that a PhD is not required, and the same is true at every tier below frontier labs. Public work substitutes for credentialing.

The Leetcode-only path is not enough anymore at staff and principal levels in SF. It still works at FAANG mid-level loops where the interview format demands it, but it is not what gets you sourced.

The ghost-job problem, and how to spot one in SF

A meaningful share of those 9,000 LinkedIn roles is not a real role. ResumeBuilder found that 40% of tech companies posted at least one ghost job in the past year, and 79% of those listings remained active when surveyed (Source: Fonzi AI). Independent LinkedIn-data analyses estimate 27.4% of US listings are likely ghost jobs (Source: Fonzi AI). The honest read is that somewhere between 25% and 40% of what you see in the SF ML market is non-actionable.

Strip 30% from the 9,000 LinkedIn figure and you are at 6,300. Filter for SF-only (roughly 25% of Bay Area listings) and you are at 1,575. Filter for your seniority bucket and stack and you are looking at low hundreds of real, conversion-likely roles. The denominator is fictional, and the candidate-side action is to spend zero minutes applying to the 70% bucket.

Three signals a listing is real:

• A specific hiring manager or founder is active on LinkedIn the same week the role was posted
• The job description names a product, a team, or a specific model class, not "build cutting-edge AI solutions"
• The listing is under 14 days old

Three signals it is likely dead:

• The listing has been open more than 60 days
• The copy appears verbatim on multiple postings (template re-use)
• No one internal is visibly connected to the role on LinkedIn

The two-week filter is the single most valuable cut. Roles open more than 60 days are usually either fully funneled, paused, or never real in the first place. Treat the listing date as the most important field on the page.

Three paths that actually work for landing an SF ML role

Three real channels into the SF ML market in 2026. Pick based on the network you actually have.

Path 1. Cold applying through public boards. Indeed, LinkedIn, Built In SF, Glassdoor. The default, the broadest, the slowest. Works at the FAANG mid-level tier and at post-IPO companies with structured intake. Stops working at frontier labs and at most scale-up senior ICs. Conversion rate from cold application to first call is low single digits even at well-fit roles. Useful as a directory, weak as a funnel.

Path 2. Warm introductions through your network. The highest conversion path by a wide margin. Time from intro to first call is often under a week. Constraint: you need a network that overlaps with hiring managers at the tier you target. If you are new to SF or transitioning from research into industry, this path is real but bottlenecked on relationships you have to build.

Path 3. Get represented by a talent agent. Standout matches tech professionals with hiring companies in the US and introduces matched candidates directly to the founder when both sides say yes (Source: standout.work). First matches arrive within a few hours of profile completion (Source: standout.work). Free for candidates; placement-fee model on the company side (Source: standout.work). Standout covers all tech roles across the US, with ML and AI engineering as one of many categories (Source: standout.work). Best fit for mid-level through staff/director candidates with one piece of public work or 5+ years of experience.

The three paths layer. Most of the senior ML candidates we represent run all three in parallel for the first two weeks of a search, then concentrate on whichever channel surfaced the highest-quality matches.

When to skip the job board entirely

Public job boards are designed for hiring managers to collect inbound, not for candidates to find roles. If you have 5+ years of ML experience, one piece of public work, and a clear seniority target, applying is the wrong unit of effort. Getting introduced is.

The clearest case to skip the board entirely: you are targeting a frontier lab or a Series B–D AI scale-up. At those companies, every senior hire we have placed went through a warm-intro or sourcing channel. The listing exists; the hire does not happen through it.

The case where the board still works: FAANG mid-level rotations, structured intake at public companies, or government-adjacent roles where the application is the official intake. The board is the right tool for those buckets. For everything else in SF ML in 2026, it is a directory you scan, not a funnel you apply through.

FAQ

What is the average ML engineer salary in San Francisco in 2026?

The ZipRecruiter base-salary average is $151,712, with the 25th–75th percentile band at $119,600–$182,600 (Source: ZipRecruiter). Levels.fyi shows $320,000 median total comp when stock and bonus are included (Source: Levels.fyi). Treat anything in the $150K range as a base-only quote, not the full package.

Which companies pay ML engineers the most in San Francisco?

Frontier AI labs lead. OpenAI L5 software engineers earn roughly $1.15M total comp ($336K base plus $774K stock) (Source: Levels.fyi — OpenAI). Anthropic's median software engineer total comp is approximately $600K, with senior engineers at $316K base plus $247K in stock (Source: Levels.fyi — Anthropic). Apple's ML engineer ladder in the SF Bay Area runs $190K (ICT2) to $560K (ICT5) with a $405K median (Source: Levels.fyi — Apple).

How many ML engineer jobs are open in San Francisco right now?

LinkedIn lists 9,000+ open ML engineer jobs in the SF Bay Area; Glassdoor counts 2,386 in San Francisco proper as of May 2026 (Source: LinkedIn). SF Standard counts 2,341 open AI engineer roles specifically in SF in early 2026 (Source: SF Standard). Strip the 27–40% ghost rate and the bucket filter, and the real number of conversion-likely roles for any given stack and seniority is in the low hundreds.

Are many SF ML job listings fake or "ghost jobs"?

Yes. ResumeBuilder found 40% of tech companies posted at least one ghost job in the past year, with 79% of those listings still active at survey time (Source: Fonzi AI). Independent LinkedIn analysis pegs ghost listings at 27.4% of US tech listings (Source: Fonzi AI). Filter on listing date and on whether anyone internal is visibly connected to the role.

Is it better to apply directly or get an introduction for an ML role in SF?

Get introduced. Direct intros convert dramatically better than cold applications, especially at frontier labs and Series B–D AI scale-ups. If you have 5+ years of experience or one piece of public work, get represented by Standout (free for candidates), where matches are made and direct founder intros run when both sides say yes (Source: standout.work).

---

Stop applying. Get introduced.

Standout is the AI talent agent for tech professionals in the US. Build your profile in 10 minutes and get matched with hiring companies in SF within hours. No apply-and-pray, no recruiter spam. Free for candidates.

Build your Standout profile →