Seventy-two percent of AI engineering roles posted by San Francisco-headquartered companies in July 2026 carry some form of remote eligibility — hybrid, remote-friendly with required travel, or fully distributed — per ENTRA's July 2026 tracking of career portals across 200-plus SF-based AI companies. But that number conceals a hard fault line. At the two companies that set the gravitational standard for global AI talent, it barely applies. Anthropic lists 8 percent of its open roles as remote-eligible. OpenAI lists 4 percent. The gap between the 72-percent headline and what Anthropic and OpenAI actually offer is the defining tension in SF's AI labor market right now: the labs that pay the most are the ones most insistent you show up.
The Research vs. Applied Split
The mechanics of who must relocate and who can stay distributed run along one axis: research vs. applied.
At Anthropic — approximately 2,500 employees and 391 open roles as of mid-2026 — research scientist and research engineer positions are uniformly listed for San Francisco, with no remote option. The company expects all staff in-office at least 25 percent of the time; for core research teams, the real expectation is higher. Anthropic's published language on select postings uses "remote-friendly (travel-required)" for a narrow set of senior individual contributors in AI safety, policy, and select security engineering functions. The inference is unambiguous: if you are writing training code or running interpretability experiments, you are moving to SF.
OpenAI's picture is tighter still. Only 28 of 681 active roles — 4 percent — carry a fully remote designation, per a June 2026 analysis of the company's Greenhouse board. OpenAI's standard policy places employees in the Mission Bay office Monday through Wednesday; Thursday and Friday are discretionary. New hires within commuting distance are given approximately 45 days to establish proximity. Core research, model development, product engineering, and safety are not remote. The Research Engineer posting on OpenAI's careers page lists San Francisco as the single acceptable location — no exceptions flagged.
Scale AI runs a different profile. With 173 active roles as of July 2026, Scale concentrates engineering and product hiring in San Francisco and New York, but its government, trust-and-safety, and data operations postings carry more geographic flexibility. Scale's model as a data infrastructure layer means its ML platform engineers are more distributed than a pure frontier research shop, even if its SF headquarters still anchors the technical leadership.
The broader SF AI ecosystem — the YC-backed startups, the applied AI companies, the enterprise tooling layer — is where the 72-percent figure originates. Mercor, which placed over 30,000 contractors at OpenAI and Anthropic in 2025 for RLHF and model training work, runs a fully remote hiring model. It competes for the same applied engineering talent the frontier labs want to keep onsite. The applied layer is distributed by design.
The Compensation Arithmetic
If Anthropic and OpenAI require presence, the question is what they pay for it.
Anthropic's software engineer total compensation in the San Francisco Bay Area runs from $555,000 at Senior Software Engineer to $767,000 at Lead, per Levels.fyi data current to mid-2026. That range includes base salary, RSU equity at the company's approximately $380 billion post-Series G valuation, and a performance bonus of 15 to 25 percent of base. Critically, Anthropic does not apply geographic pay adjustments for its US-based roles, meaning that the few remote-eligible positions carry the same band as their SF-anchored equivalents. For candidates in lower-cost metros, that is a meaningful signal.
OpenAI's frontier comp lands in a comparable corridor. Frontier-lab software engineers across Anthropic and OpenAI clear $600,000 to $795,000 median total compensation, per ENTRA's analysis of Levels.fyi data for this cohort. Senior remote AI engineers in the national market — those not co-located with a frontier lab — track a median of approximately $206,600, per MRJ Recruitment's 2026 rate data. The gap between being in the building at Anthropic and being distributed for a mid-tier AI company is approximately $350,000 to $560,000 in total compensation per year. That arithmetic is not subtle. It is the reason SF remains a physical destination rather than an administrative one.
The on-site premium functions differently for different role families. For a research scientist with publications in top-tier venues, the comp is primarily an equity bet on the lab's trajectory — and the SF requirement is the price of access to that equity. For an applied ML engineer who could build RAG pipelines from anywhere, the calculus is harder. The premium is real, but so is the cost of San Francisco housing.
The SF-to-Distributed Pipeline
What looks like a binary — SF or remote — is actually a staged progression for most frontier labs.
Anthropic and OpenAI hire for core research and founding-team-adjacent roles from their SF campuses. Once a product surface matures — inference APIs, enterprise tooling, trust-and-safety pipelines — the applied engineering work around that surface becomes progressively more distributable. This is how scale enters the equation: the SF campus does the frontier work, and the distributed layer operationalizes it.
OpenAI has now expanded to more than one million square feet in San Francisco plus a 439,000-square-foot Mountain View campus, while simultaneously holding offices in New York, London, Dublin, Tokyo, and Singapore. The geographic expansion is not remote-work expansion. It is market-access expansion. The Singapore and Tokyo offices serve sales, policy, and enterprise customer success functions. Core model development stays in Mission Bay. The pattern is hub-and-spoke with a very dense hub.
Who Is Competing for Remote AI Engineers
Four categories of employer are pulling remote-eligible AI talent away from SF lab orbit in 2026.
xAI, Elon Musk's lab which now employs roughly 5,000 people, pays $180,000 to $440,000 base in the Bay Area, below Anthropic and OpenAI bands but above the enterprise market. xAI's aggressive recruiting strategy includes direct approaches to current Anthropic and OpenAI engineers, with offers front-loaded on equity at a valuation that has moved sharply upward since the Grok-3 release cycle.
Mercor runs a fully remote contractor model — AI interview-to-offer, paid weekly via Stripe — and has become the primary vehicle for applied AI engineers who want frontier-lab adjacency without the SF mandate. Its customers are the frontier labs themselves, which creates an irony: OpenAI and Anthropic are funding the very platform that gives their potential full-time hires a remote alternative.
Google DeepMind operates a hybrid model with more latitude than OpenAI or Anthropic for senior researchers. DeepMind's distributed research culture — rooted in its pre-acquisition London structure — has given it a competitive advantage in recruiting senior AI researchers who are unwilling to commit to a three-day-a-week SF office schedule.
Globally, Mistral in Paris has demonstrated that a credible frontier lab can be built outside the Bay Area by concentrating world-class researchers in a single non-SF city rather than competing head-on for SF salaries. The talent pool for which SF labs must actively compete is now global, not domestic.
H2 2026: Three Things to Watch
Anthropic's remote-eligible expansion. Anthropic's $3.5 billion Series G (closed May 2025 at a $380 billion post-money valuation, bringing cumulative funding to approximately $30 billion) was paired with stated plans to reach approximately $70 billion in revenue by 2028. That trajectory requires hiring far beyond what the SF research campus can absorb. Watch whether Anthropic begins reclassifying senior applied engineering and safety roles to remote-eligible through Q3 2026. A shift in that direction would signal that the company is entering the distributed-scaling phase.
OpenAI's 8,000-employee target. The Financial Times reported in March 2026 that OpenAI is targeting 8,000 employees by year-end, adding roughly 3,500 people to a base of approximately 4,500. Doubling headcount at that pace while maintaining a Mon-Wed-in-office mandate will require either a significant San Francisco housing subsidy program or a quiet expansion of remote eligibility for non-research functions. The company's real estate footprint already runs over one million square feet; the constraint is talent availability, not floor space.
The compensation ceiling for remote-only offers. The 1.6 million open AI roles globally against 518,000 qualified candidates — a 3.2-to-1 demand-to-supply ratio, per Second Talent's 2026 research — means that distributed employers cannot simply underprice SF labs and expect to win. The remote-work premium documented in AI hiring runs roughly 22 percent above office-based equivalents nationally, per Dice's 2025 analysis. That premium will compress if frontier labs begin selectively extending remote eligibility for their highest-demand applied roles. When Anthropic or OpenAI decides a function can be done anywhere, the market rate for that function globally reprices overnight.
San Francisco is not losing its grip on frontier AI. It is holding that grip more deliberately, concentrating it on the research core while allowing the applied layer to distribute. The labs know the difference between the two. The candidates who can credibly claim residence in either category need to know it too.
The 72% remote-eligible figure reflects ENTRA's July 2026 analysis of job postings from 200-plus SF-headquartered AI companies, classified as remote-eligible where listings stated hybrid, remote-friendly with travel, or fully remote as a primary or co-primary work arrangement. Anthropic remote-eligible share (8%, 35 of ~391 roles) and OpenAI remote-eligible share (4%, 28 of 681 roles) reflect job-board analyses published in mid-2026 by JobsByCulture and corroborated by ENTRA review of Greenhouse listings; figures are point-in-time and subject to change. Anthropic total compensation ranges ($555K–$767K, SF Bay Area) sourced from Levels.fyi, mid-2026. Frontier-lab median total compensation ($600K–$795K) is ENTRA's analysis of Levels.fyi data for Anthropic and OpenAI software engineering cohorts. National senior remote AI engineer median ($206,600) sourced from MRJ Recruitment 2026 data as cited in Noon.ai compensation benchmarks report. OpenAI headcount (7,850 target: 8,000 by year-end) sourced from Financial Times reporting, March 21, 2026, and MakerStations OpenAI employee statistics 2026. Anthropic headcount (~2,500), Series G round size ($3.5B, closed May 2025), Series G post-money valuation ($380B), cumulative total funding (~$30B), and revenue projection ($70B by 2028) sourced from company announcements and JobsByCulture 2026 tracking. xAI employee count (~5,000) and compensation band ($180K–$440K) sourced from LinkedIn workforce data and AI compensation benchmarks (Pin.com, 2026). Demand-to-supply ratio (1.6M roles, 518K candidates, 3.2:1) sourced from Second Talent's 2026 Global AI Talent Shortage research. Remote wage premium (~22%) sourced from Dice 2025 analysis as cited in Noon.ai 2026 benchmarks. Scale AI active role count (173) sourced from Recruiting From Scratch, July 2026. OpenAI office footprint (1M+ sq ft SF, 439K sq ft Mountain View) sourced from CoStar reporting and company announcements, 2026.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.