Only 13% of AI Startup Jobs Are Remote. Here's Where They Hide
By Fast AI Startup Jobs
TL;DR
- 3,681 of 28,414 open roles at the AI startups we track are advertised as remote-eligible. That's 13.0%. The live set is at fastaijobs.com/jobs/remote.
- Funding stage tells you almost nothing. Seed 13.1%, Series A 13.8%, Series B 12.4%, Series C 13.4%. The line is flat across the entire middle of the market.
- What the company builds tells you a lot. Cybersecurity startups advertise 38.8% of roles as remote. Defense and aerospace startups advertise 0.9%. Same dataset, 43x apart.
- The labs everyone wants to work at are among the least remote. LLMs & Gen AI sits at 6.2%, AI Infrastructure at 8.2%, robotics and hardware near 5%.
- Department matters, but less than sector: GTM & Customer Success 16.8% versus Engineering, Research & Product at 12.3%.
- One caveat before the numbers: we read remote status from the location fields on each posting. A hybrid role listed under "San Francisco" counts as on-site here. This is a floor on flexibility, not a ceiling.
In this article: How we counted · Funding stage · Sector · Why security is different · Most-remote companies · Departments · How to search · FAQ
How we counted a remote job
Our dataset is 28,414 deduplicated open roles across 1,824 tracked AI startups, current as of August 1, 2026. Deduplicated matters: the raw feed lists the same job once per office city, so a single "Backend Engineer" posted to five locations arrives as five rows. We count distinct job IDs.
A role counts as remote when the location fields on the posting say so. That is the whole method, and it has an obvious hole. Companies that run a hybrid policy almost always list the role under the office city, because that is where they want you three days a week. Those roles land in our on-site bucket. So does every job at a company with a generous unwritten work-from-home culture that never made it into the ATS.
What we are measuring, then, is advertised fully-remote eligibility. If you are trying to figure out how many AI startup jobs you could take from another city without a conversation, 13.0% is the honest answer. If you are asking how many AI startup employees work from home two days a week, this data cannot tell you, and anyone quoting a number for that is guessing.
The 13.0% is lower than the cross-industry remote-job share you will see in most 2026 reports. That gap is the actual story here, and it holds up under every cut we ran.
Funding stage barely moves the number
The intuitive model goes something like this: tiny startups are distributed because they cannot afford an office, and big ones drag you back once they have one. Our data does not support either half.
| Funding stage | Remote roles | Total roles | Remote share |
|---|---|---|---|
| Pre-Seed | 6 | 72 | 8.3% |
| Seed | 166 | 1,272 | 13.1% |
| Series A | 634 | 4,597 | 13.8% |
| Series B | 635 | 5,103 | 12.4% |
| Series C | 633 | 4,712 | 13.4% |
| Series D | 487 | 3,176 | 15.3% |
| Growth | 71 | 1,065 | 6.7% |
Seed through Series C spans 15,684 postings and moves within 1.4 percentage points. That is noise. A company that just raised $3M and a company that just raised $150M advertise remote work at effectively the same rate.
Two rows sit outside the band, and only one of them means anything. Pre-Seed at 8.3% is six remote roles out of 72 total, a sample so small that two postings would swing it four points. Ignore it. Growth-stage at 6.7% is built on 1,065 postings and is real: the latest-stage private companies in our set advertise remote at half the rate of everyone else. Series D, meanwhile, is the highest band at 15.3%. So the story is not a clean maturity curve where flexibility decays with headcount. It is a flat line with a late-stage dip that shows up only at the very end.
If you have been filtering your search by stage because you assumed early meant flexible, you have been sorting on a variable that does not do anything.
What the company builds is the real predictor
Here is the same 28,414 roles cut by what the company actually sells. We show categories with more than 500 postings, which covers 18,788 roles.
| Category | Remote roles | Total roles | Remote share |
|---|---|---|---|
| Cybersecurity | 430 | 1,108 | 38.8% |
| Infrastructure | 144 | 657 | 21.9% |
| Developer Tools | 211 | 1,204 | 17.5% |
| Enterprise SaaS | 202 | 1,275 | 15.8% |
| FinTech | 319 | 2,817 | 11.3% |
| AI Infrastructure | 316 | 3,849 | 8.2% |
| Healthcare | 113 | 1,401 | 8.1% |
| LLMs & Gen AI | 70 | 1,127 | 6.2% |
| Vertical AI | 43 | 780 | 5.5% |
| Robotics | 35 | 681 | 5.1% |
| Hardware & Robotics | 31 | 643 | 4.8% |
| Defense Tech | 39 | 1,908 | 2.0% |
| Climate Tech | 11 | 593 | 1.9% |
| Defense & Aerospace | 7 | 745 | 0.9% |
The spread is 38.8% to 0.9%. Nothing else we sliced comes close to that range.
The top of the table is software sold to technical buyers who are themselves scattered: security teams, platform teams, developer-tools customers. The bottom is anything that touches a physical object or a secure facility. Defense and aerospace at 0.9% means seven remote roles out of 745, and that is not a policy preference. You cannot debug a satellite bus from your kitchen, and cleared work happens inside a cleared building. Robotics and hardware are the same constraint in a friendlier package.
What surprised us was the middle. AI Infrastructure at 8.2% and LLMs & Gen AI at 6.2% are the two categories most people picture when they picture an AI job, and they are down near the hardware companies. There is no physics forcing that. A researcher training models on a cluster is not standing next to the cluster. This one is a choice about how these companies want to work, and they have chosen the office.
Why cybersecurity is the outlier
430 remote roles out of 1,108 is not a rounding artifact. Security startups advertise remote at three times the rate of the dataset average and more than forty times the rate of defense.
Some of it is who they sell to. A security buyer is a CISO with a team spread across whatever offices and time zones the parent company has. The product is delivered over the network, the proof-of-concept runs in the customer's cloud, and nobody has ever asked a vendor to fly out to install an agent. When your entire sales motion works remotely, an in-office mandate is solving a problem you do not have.
Some of it is the talent market. Security engineers who can do detection engineering or offensive research are scarce, they know it, and a large share of that community was already distributed before 2020 through open-source projects and conference circuits rather than through employers. A security company that insists on Palo Alto is competing for a fraction of the people it needs.
And some of it is that incident response never respected office hours anyway. If your on-call rotation already spans three continents so somebody is awake when an alert fires, you have functionally accepted distributed work. Writing it into the job posting costs nothing.
The counter-pressure is real and getting louder. The South China Morning Post reported this year that several US AI startups have adopted China-style extended in-office schedules, the "996" pattern of 9am to 9pm, six days a week, explicitly as a way to move faster than competitors. That norm and remote hiring cannot both win. Our data suggests the frontier-model labs have already picked a side, and it is not the flexible one.
The companies where remote is the default
Aggregate percentages hide the companies actually doing the hiring. These are the twelve companies in our set with the most open remote roles right now.
| Company | Stage | Category | Remote roles | Total roles |
|---|---|---|---|---|
| Leland | Series A | EdTech | 188 | 194 |
| Grafana Labs | Series E | Data & Analytics | 137 | 137 |
| Chainguard | Debt-funded | Cybersecurity | 69 | 69 |
| Cribl | Series E | AI Infrastructure | 67 | 67 |
| Perle | Seed | Data & Analytics | 62 | 63 |
| Abnormal Security | Series E | Cybersecurity | 61 | 71 |
| Vanta | Series D | Cybersecurity | 61 | 97 |
| Horizon3AI | Series D | Cybersecurity | 58 | 96 |
| Vultr | Debt-funded | Cloud Computing | 55 | 63 |
| AlphaSense | Series G+ | FinTech | 54 | 194 |
| Mercury | Series D | FinTech | 53 | 56 |
| Databricks | Series G+ | AI Infrastructure | 51 | 771 |
Four of the twelve are cybersecurity companies, which is what the category table predicted. Grafana Labs, Chainguard, and Cribl post every single open role as remote, with no on-site listings at all. That is a company built remote-first from the start, not one that added a policy.
Leland is the volume leader at 188 of 194 roles, and it is worth knowing what those roles are before you get excited: an education marketplace at that ratio is hiring coaches and instructors, which is genuinely remote work but a different proposition than a salaried engineering job.
Databricks is the trap on this list. 51 remote roles is a large absolute number and puts it in the top twelve, but those 51 sit inside 771 total openings, a remote rate of 6.6%. Sorting any job board by "most remote jobs" flatters the biggest employers. Sort by rate instead and Databricks disappears while Mercury, at 53 of 56, moves near the top.
GTM is the most remote-friendly department
| Department | Remote roles | Total roles | Remote share |
|---|---|---|---|
| GTM & Customer Success | 1,063 | 6,333 | 16.8% |
| Engineering, Research & Product | 1,754 | 14,278 | 12.3% |
| Operations | 354 | 3,185 | 11.1% |
The gap between the most and least remote department is 5.7 points. The gap between the most and least remote sector is 37.9. Sector wins, and it is not close, so if you only have energy to filter on one thing, filter on what the company builds.
GTM coming out on top is not a surprise once you look at what those roles do. An account executive covering the Midwest was never going to sit in a San Francisco office, and a customer success manager assigned to accounts in three time zones is remote whether or not the posting says so. This connects to something we found in the AI agent startup hiring mix: agent companies are tilting hard toward GTM headcount as they move from demo to deployment. The most commercially urgent hiring at these companies happens to also be the most location-flexible.
Engineering at 12.3% is the number that will disappoint most readers of this site, since engineering is 14,278 of the 28,414 roles we track. Half the market, hired mostly on-site.
How to actually find the remote roles
Three things fall out of this data that should change how you search.
Start from the sector, not the stage. If remote is non-negotiable for you, the entire defense, aerospace, climate, robotics, and hardware set is roughly 4,570 postings offering 123 remote roles between them. You can skip all of it and lose almost nothing. Cybersecurity, infrastructure, and developer tools are 2,969 postings with 785 remote roles. That is where your time goes.
Filter by rate, not by count. A company with 800 openings and 40 remote ones is not a remote-friendly company, it is a big company. The twelve-company table above sorts by count because that is what job boards do; the ratio column next to it is the number that tells you what you would actually be walking into.
Read the location field literally, then ask anyway. Our floor-not-ceiling caveat cuts both ways for you personally. Plenty of roles listed under a city will be negotiable for the right candidate, and a hybrid arrangement never shows up in this dataset. The 13.0% is what companies advertise, not the boundary of what they will agree to.
The live remote board is at fastaijobs.com/jobs/remote, filtered on the same location logic described here. If you want to see the full set first and cut it yourself, all open roles are here, and the startup hiring atlas breaks the same dataset down by company and stage.
FAQ
What percentage of AI startup jobs are remote in 2026?
13.0%. Across 28,414 deduplicated open roles at 1,824 tracked AI startups as of August 1, 2026, 3,681 are advertised as remote-eligible. That figure counts only roles whose posting location says remote, so hybrid positions listed under an office city are excluded. The real share of jobs with some location flexibility is higher, but 13.0% is the share you can apply to from anywhere without negotiating.
Which AI startup categories hire the most remote workers?
Cybersecurity leads by a wide margin at 38.8% remote, followed by Infrastructure at 21.9%, Developer Tools at 17.5%, and Enterprise SaaS at 15.8%. At the other end, Defense & Aerospace posts 0.9% remote, Climate Tech 1.9%, and Defense Tech 2.0%. The pattern tracks whether the product touches physical hardware or a secure facility.
Do early-stage startups offer more remote jobs than late-stage ones?
Barely, and not in the direction most people expect. Seed companies post 13.1% remote, Series A 13.8%, Series B 12.4%, Series C 13.4%, and Series D 15.3%. The one real outlier is Growth stage at 6.7%. Funding stage is a poor predictor of remote availability compared to what the company builds.
Why are AI research labs so rarely remote?
Companies in our LLMs & Gen AI category advertise only 6.2% of roles as remote, and AI Infrastructure sits at 8.2%. There is no technical reason for it, since model training happens on remote clusters regardless of where the researcher sits. It is a cultural stance about co-located intensity. The South China Morning Post reported in 2026 that some US AI startups have gone further and adopted China-style extended in-office schedules to compress iteration cycles.
Which AI companies have the most remote job openings?
By absolute count: Leland (188), Grafana Labs (137), Chainguard (69), Cribl (67), and Perle (62). By share of total openings, Grafana Labs, Chainguard, and Cribl all post 100% of their roles as remote. Mercury (53 of 56) and Vultr (55 of 63) are close behind. Databricks appears in the top twelve on volume with 51 remote roles, but that is only 6.6% of its 771 openings.
Data sourced from fastaijobs.com as of August 1, 2026. We analyzed 28,414 deduplicated open roles across 1,824 tracked AI startups; funding data covers 97% of companies.
Related: The AI Agent Startup Hiring Mix 2026. Agent companies are shifting headcount toward GTM as they move from demos to deployments, which is the same department that leads on remote availability here.