OpenAI Posts 177 SF Jobs a Month. The Bottleneck Is Relocation.

If you are hiring for an SF AI startup in 2026, your "Location: San Francisco Bay Area" filter is the reason your pipeline is empty. Per the SF Examiner's late-April 2026 reporting, 10 top SF AI companies have 1,384 open roles, 646 of them SF-only, and OpenAI alone is now posting 177 new SF roles a month, up from 66/month in 2024. The local pool was tapped out two boom cycles ago. The job to be done is no longer geo-targeted sourcing. It is relocation persuasion.

The math broke. Here's where.

Frontier labs are hiring faster than the SF Bay can produce engineers. OpenAI's posting velocity has roughly tripled in two years (66 → 126 → 177 per month, per Lightcast data compiled by SF chief economist Ted Egan). Of OpenAI's 651 current openings, 412 are SF-only and another 60 are SF-preferred. Thinking Machines Lab, Mira Murati's shop, has 29 of 30 listings in SF and has 4x'd headcount in two years to ~130 people inside a three-story Mission building. Physical Intelligence, the Bezos-backed robotics company, intends to fill all 20 of its current listings in SF.

177

Now stack that against the supply side. SignalFire's data puts 49% of US big-tech engineers and 27% of US startup engineers in the Bay Area. That's the headline number every SF founder cites when they say "the talent's already here." Read it the other way: 51% of big-tech engineers and 73% of startup engineers are not in the Bay Area. That's your actual hiring pool now, because the local one is mathematically saturated. OpenAI and Anthropic combined employ fewer than 10,000 people, and OpenAI is trying to add roughly 2,000 new SF-based hires this year by itself.

The recruiters still typing "Location: San Francisco" into LinkedIn Recruiter are fighting for the same shrinking subset of names that every other AI startup has already cold-emailed twice this quarter.

In-person is the product, not the apology

The cultural defaults shifted with the math. Sierra (Bret Taylor and Clay Bavor) publishes its policy directly on the careers page: "Since we work in person, interviews take place at one of our offices," and describes itself as "primarily an in-person company based in San Francisco." Cognition runs a 25,000 sq ft HQ in South Park with private chefs and "family dinners," and the SF Standard's March 2026 profile reports engineers literally living in apartments around the corner from the office, with a culture the piece characterized as past 996. Wordware requires Bay Area presence before starting.

These companies are not embarrassed about it. They are advertising it.

That is the recruiting reframe most founders haven't internalized yet. The relocation pitch in 2026 is not "sorry, we know SF is expensive, but…" It is "we are deliberately filtering for people who want this." Sierra flies finalists out and pays for hotels. Cognition feeds candidates by private chef during onsites. These are deliberate top-of-funnel persuasion mechanics, and they only work on candidates who self-selected in. Apologize for in-person and you lose the candidates who actually want it.

The recruiters still typing Location: San Francisco are fighting for a subset every other AI startup has cold-emailed twice this quarter.

The in-person persona is real and small

Treat "wants to move to SF for an in-person AI startup" as a persona, not a logistics question. They tend to be 26-34, single or partnered without school-age kids, have already done at least one stint in a high-intensity environment (HFT, YC, a frontier lab, military, top-tier consulting), and read SF moves as career capital. They do not need to be persuaded that 60-hour weeks are fine. They need to be persuaded that your team is the one worth doing them with.

Stop filtering on location. Start filtering on willingness.

Here's the operational shift. On inbound and outbound, drop location as a hard filter. Replace it with a willingness-to-relocate qualifier early in your screen. Then build separate funnels for three pools, because they convert differently:

1. Already in SF. Your fastest pipeline, but smallest. Treat it as opportunistic, not the strategy.
2. Willing to relocate to SF. The bulk of your pipeline now. Most of these candidates are in NYC, Seattle, Mountain View, LA, Denver, Austin. Refolk's index of ~9,564 US-based engineers currently holding "AI Engineer" or "Machine Learning Engineer" titles shows only about 9 of the top 25 by signal sit in SF proper. The rest are scattered across exactly the cities a relocation pitch needs to reach.
3. Already in SF but not labeled "AI engineer." This is the sleeper pool. SF and San Mateo counties lost 4,400 jobs in 2025, and the information sector alone lost 4,500 (-4%), per SF Standard's January 2026 reporting. Pinterest cut 15%. Meta Reality Labs cut 1,000+. Many of those displaced engineers are still in SF, still want to work in tech, and are invisible to anyone running "title contains ML" searches. They are the highest-conversion outbound pool of the three.

The standard sourcing stack is bad at all three of these. LinkedIn Recruiter's location filter is binary and based on declared location, which lags reality by 6-18 months. Boolean strings on title don't catch the displaced backend engineer who shipped a RAG side project last month. And nothing in the standard stack encodes "open to moving."

This is exactly the friction we built Refolk for. You describe the candidate in plain English, including soft criteria like "willing to move to SF" or "previously did an SF stint and left," and you get a ranked shortlist across GitHub, LinkedIn, and the open web. The geo filter stops being a wall and becomes a tag.

Rebuild the funnel around the onsite visit

If relocation persuasion is the bottleneck, then the onsite visit is your most important funnel stage, not a final formality. Sierra and Cognition have already figured this out. The rest of the SF AI ecosystem hasn't.

Here's what changes operationally when you accept that:

Track onsite-visit-to-offer-acceptance as its own metric

Most ATSes still report phone-screen → onsite → offer → accept. Split the onsite into "remote technical" and "in-person SF visit," and track conversion through each. If your in-person visit isn't pushing acceptance materially higher than a remote final round, you are wasting the flight. Anthropic's reported 95% offer-acceptance and 80% retention (per SignalFire) is the upper bound. If you're below 70%, your visit is mid.

Pay for the trip. Always.

Sierra flies candidates to SF and books hotels by default. If you are a Series A AI startup competing for the same engineers, this is now table stakes, not a perk. A $1,200 flight-and-hotel spend is rounding error against a $50K+ recruiter fee or a six-month vacancy.

Build the SF visit itself

Don't run a single 5-hour panel and send them home. Add dinner with two engineers (not the founders, the engineers they'd actually work with), a walk through the neighborhood they'd live in, and unstructured time. Cognition has the private-chef "family dinner" because they understood early that the visit is a sales motion. The candidate is interviewing the city as much as the company.

Address the housing "no" head-on

The real "no" for relocation candidates isn't culture, it's $1.35M median home price and $3,100-$3,800/mo for a 1BR in Hayes Valley or Noe Valley. The pitch has to acknowledge this, ideally with a relocation package that's specific (e.g., $25K signing earmarked for first/last/deposit, six weeks of corporate housing, a real-estate contact list) rather than a vague "we'll help with relocation."

412

49%

What this means for your sourcing stack

If you are hiring for OpenAI, Anthropic, or any in-person AI startup at SF rates, three changes are non-negotiable in 2026:

1. Drop "Location: San Francisco" as a hard filter on outbound. It cuts your pool by half before you've evaluated a single resume. Use it as a tag and a downstream qualification, not a gate.
2. Build a "displaced SF engineer" pipeline as a separate workflow. Different titles, different messaging, different urgency. They are already here. They don't need a relocation pitch. They need a "your skills transfer faster than you think" pitch. This is the most underpriced pool in the SF market right now.
3. Treat the onsite visit as a paid acquisition channel. Budget it. Measure it. Iterate on it. Your competition (Sierra, Cognition, Thinking Machines) already does.

The SF AI hiring problem flipped without most recruiters noticing. The cities where your candidates currently live (Seattle, NYC, Austin, Cleveland, Denver) are the same cities your competitors are also waking up to. The window on a clean relocation pitch, before every Series B AI startup is sending the same "fly out for a Sierra-style onsite" email, is shorter than you think. Tools like Refolk exist precisely so you can describe what you actually want ("senior backend engineer, ex-frontier-lab adjacent, currently in NYC, has tweeted about moving to SF in the last six months") and get a list, instead of running 14 boolean strings against a saturated geo filter.

OpenAI is going to post another 177 jobs next month. Anthropic, Sierra, Cognition, Thinking Machines, and Physical Intelligence are going to keep their in-person mandates. Your move is to stop pretending the local pool will fill them.

FAQ

How do I tell which non-SF engineers are actually open to relocation?

Look for behavioral signals, not stated preferences. People who recently moved cities (visible on LinkedIn or GitHub bio updates), people who've previously done an SF stint and left, engineers who follow or interact with SF AI accounts heavily on X, and people who've publicly tweeted about visiting SF or attending SF events. Stated "open to relocate" toggles on LinkedIn are weak signals because most engineers leave them on by default. The strong signals are recent travel, recent friend-graph shifts, and public posts. A natural-language sourcing tool surfaces these alongside the resume; a boolean string can't.

Isn't the displaced-SF-engineer pool already picked over?

Less than you'd think. Most AI recruiters search on current title ("ML Engineer," "AI Engineer," "Research Engineer"), which excludes the laid-off Pinterest backend engineer who's been shipping RAG side projects on weekends. The 4,500 information-sector jobs SF lost in 2025 are still circulating in the market, and the candidates who haven't landed yet are typically the ones whose titles don't match the AI-engineer template. Source on what they've built in the last 12 months, not their last job title.

Should I just go remote-first instead?

If you can, sometimes yes. But if you're hiring against OpenAI, Sierra, or Cognition, you're competing for a candidate persona that explicitly wants in-person. Going remote doesn't broaden your pool against those competitors, it shifts you into a different (also crowded) market against Vercel, Replicate, and every Series B remote-first AI startup. Pick your lane and commit. The worst outcome is hybrid-with-an-asterisk, which loses both audiences.

What's a realistic relocation package for a Series A AI startup in 2026?

For a senior IC: $25-50K signing bonus earmarked for relocation, 4-6 weeks of corporate housing, paid flight and hotel for the onsite visit, and a real broker contact for both rentals and home purchases. For staff and above, add a one-time housing assistance grant ($75-150K) or a forgivable loan tied to a 3-year tenure. The number that closes is rarely the headline comp; it's the specificity. "We'll cover relocation" loses to "here is a $35K wire on your start date and the realtor's number."