attention head AI jobs Alerts Saved

AI jobs: pay and hiring

Who is hiring for AI roles, what their postings ask for and what they pay, read from employers' own job boards.

On this page: Demand Roles by work areaWhere AI is hiring Who is hiring Kinds of role Who asks for what Skills and pay Asked together Pay Visa pay

Open roles, week by week

Roles (employer and title, once each) open at the end of each of the last 6 weeks, by work area. Rebuilt from when each posting was first read and when it closed, and drawn only from the weeks after half of these employers' boards had been read: before that, roles are missing because we had not looked yet. A change shows once there are that many weeks. Fewer than 10 open roles show no row.

work areaOpen nowShare4 weeks12 weeksWeeks
Agents6,26035%+7%
Retrieval & recommendations2,03611%+6%
Inference & serving1,91111%+1%
Robotics & autonomy1,5058%−2%
Post-training & RL1,4338%+3%
Multimodal & vision1,3338%−1%
Model evaluation1,2377%+5%
GPU & kernels6564%+1%
Data for AI3002%+7%
Safety & interpretability2852%+5%
Pre-training1541%+3%

Roles by work area

What each open AI role works on, read from its posting; a role can have several. Median pay is the midpoint of stated annual US-dollar ranges, shown where at least ten roles state one, with how many do.

Where AI is hiring

Open AI roles by the country each posting names, and those that say they're remote.

Who is hiring

The 12 employers with the most open AI roles, of 17,736 at 3,659 employers. A role (employer and title) counts once however many places it is posted in. Senior and above is the share whose title says so; remote, the share posted as remote anywhere. Median pay needs 10 roles that state an annual US-dollar range. Job boards and recruiters that post other employers' roles are not in the index.

#EmployerOpen AI roles6 weeksSenior and aboveRemoteMedian stated payAsks for more than most
1Amazonbigtech49%0%too fewC++ 54%AWS 50%Spark 14%
2NVIDIAbigtech68%4%too fewCUDA 38%GPU 79%Triton 13%
3OpenAIfrontier18%63%$350K · 246Distributed Systems 24%
4Bosch Groupapplied18%<1%too fewSpark 18%C++ 32%TensorFlow 14%
5Anthropicfrontier42%8%$362K · 129RL 24%LLMs 76%Evals 43%
6ServiceNowapplied82%12%$242K · 60Go 27%Distributed Systems 31%LLMs 60%
7SpaceXapplied40%0%$146K · 26
8Microsoftbigtech66%0%$177K · 58RL 19%Azure 31%Distributed Systems 21%
9Applied Intuitionrobotics26%0%$200K · 82C++ 48%
10Databricksinfra84%11%$216K · 61Spark 44%GCP 44%Azure 49%
11DoorDash USAapplied41%2%$200K · 72C++ 23%
12Waymoapplied56%0%$270K · 59JAX 56%RL 34%TensorFlow 40%

Kinds of AI role, by level

Each AI role (employer and title, once) by the kind of work its title and team point to. The bar runs left to right in the column order. Levels with fewer than 10 roles are left out.

LevelRolesResearchAI/ML EngineeringSoftware EngineeringInfrastructure & PlatformProductOtherMix
All levels17,74211%32%21%3%5%29%
Intern74315%14%26%2%2%41%
Junior2737%24%12%1%1%55%
Mid7,71614%30%17%2%4%32%
Senior4,5609%37%24%3%6%21%
Staff+3,1168%46%21%2%6%17%
Manager5511%11%48%4%1%35%
Director5332%5%15%5%6%68%
Executive2504%8%14%4%8%62%

Who asks for what

The share of each employer's open AI roles whose posting names a skill, under the first row's share for all of them. The 12 skills are those named most. Read from the postings' text: a posting that names a skill may require it or only mention it. Left out because most of their postings name none of the skills read: SpaceX.

EmployerPythonLLMsEvalsAWSPyTorchRAGAzureKubernetesC++SQLGPUGCP
All AI roles 17,736483922211615141313131212
Amazon 7685839155012811547111
NVIDIA 40359511233311319423792
OpenAI 285391721432364742
Bosch Group 272542998222015732837
Anthropic 14952764379131358106
ServiceNow 11346602215219171744311
Microsoft 10646542942120316249134
Applied Intuition 102567148210513489119
Databricks 965233144714234973201444
DoorDash USA 9532111166208231762

What AI roles ask for, and what it pays

Each dot is a skill. Across: the share of the 17,736 open AI roles whose posting names it, on a log scale so the rarer skills spread out. Up: the median pay of those roles, from the ones stating an annual US-dollar range (at least 20 per skill). The dashed line is the median of all 6,242 that state one, $212K. Select a dot for that skill's roles.

$180K$190K$200K$210K$220K$230K$240K$250K2%5%10%20%50%Median stated payShare of roles naming the skill (log scale) →All roles $212KPython: named by 48% of roles (8,512); median stated pay $208K, from 3,216; most at Amazon (448), NVIDIA (236) and Bosch Group (148). The working language of AI: model code, data pipelines, evals and the services around models.PythonLLMs: named by 39% of roles (6,978); median stated pay $215K, from 2,680; most at Amazon (296), NVIDIA (207) and Anthropic (113). Large language models: building products with them, training them, or making them better.LLMsEvals: named by 22% of roles (3,964); median stated pay $220K, from 1,810; most at Amazon (118), Anthropic (64) and OpenAI (61). Evaluations: measuring what a model can do, and whether a change made it better.EvalsAWS: named by 21% of roles (3,681); median stated pay $203K, from 1,191; most at Amazon (386), Devoteam (46) and Databricks (45). Amazon's cloud, where many AI teams train, keep data and serve models, including through SageMaker and Bedrock.AWSPyTorch: named by 16% of roles (2,916); median stated pay $224K, from 1,098; most at NVIDIA (133), Amazon (94) and Bosch Group (61). The most widely used deep learning framework, for building, training and researching models.PyTorchRAG: named by 15% of roles (2,693); median stated pay $197K, from 876; most at Amazon (61), Bosch Group (55) and NVIDIA (46). Retrieval-augmented generation: giving a model the right documents as it answers, the basis of many AI products.RAGAzure: named by 14% of roles (2,418); median stated pay $195K, from 753; most at Databricks (47), Brillio (44) and Bosch Group (41). Microsoft's cloud, where Azure OpenAI and Azure AI Foundry let companies build on hosted models.AzureKubernetes: named by 13% of roles (2,313); median stated pay $213K, from 857; most at NVIDIA (77), Devoteam (26) and Red Hat K.K. (24). The standard way to run containers across clusters, used to schedule training jobs and serve models.KubernetesC++: named by 13% of roles (2,308); median stated pay $213K, from 820; most at Amazon (418), NVIDIA (168) and Bosch Group (86). The language of performance-critical AI code: framework internals, inference engines, kernels and robotics.C++SQL: named by 13% of roles (2,249); median stated pay $189K, from 819; most at Amazon (52), Brillio (41) and Celonis (30). The language of databases, for pulling, joining and analysing the data models learn from and are judged on.SQLGPU: named by 12% of roles (2,133); median stated pay $230K, from 808; most at NVIDIA (320), Amazon (81) and Cerebras (46). The chips AI models train and run on; roles naming GPUs work close to the hardware.GPUGCP: named by 12% of roles (2,101); median stated pay $205K, from 679; most at Devoteam (49), Databricks (42) and Sia (27). Google's cloud, home to TPUs and Vertex AI, used for training, data and serving models.GCPDistributed Systems: named by 10% of roles (1,818); median stated pay $229K, from 848; most at Amazon (150), OpenAI (67) and NVIDIA (43). Software that runs across many machines: the basis of large-scale training and serving models.Distributed SystemsTensorFlow: named by 8% of roles (1,504); median stated pay $210K, from 506; most at Amazon (108), Bosch Group (38) and NVIDIA (37). Google's earlier deep learning framework, still running many production, on-device and recommendation models.TensorFlowTypeScript: named by 8% of roles (1,436); median stated pay $208K, from 617; most at OpenAI (31), Databricks (23) and Perplexity (16). The language of web front ends and many AI product backends: the apps people use models through.TypeScriptRL: named by 7% of roles (1,279); median stated pay $250K, from 557; most at Amazon (116), NVIDIA (53) and Helsing (41). Reinforcement learning, including RLHF: training models from rewards, central to post-training today's LLMs.RLSpark: named by 6% of roles (1,115); median stated pay $216K, from 442; most at Amazon (104), Bosch Group (49) and Databricks (42). Apache Spark processes data at scale, for preparing training data and features for models.SparkGo: named by 6% of roles (1,051); median stated pay $221K, from 528; most at ServiceNow (31), Perplexity (26) and Amazon (24). A language for backend services and infrastructure, for the platforms and APIs that serve AI products.GoReact: named by 5% of roles (951); median stated pay $202K, from 384; most at OpenAI (14), monday.com (11) and Amazon (10). The front-end library behind most AI product interfaces, from chat apps to developer consoles.ReactTerraform: named by 5% of roles (915); median stated pay $205K, from 314; most at Devoteam (20), Amazon (13) and OpenAI (12). Infrastructure as code, for setting up and changing the cloud resources AI systems run on.TerraformCUDA: named by 4% of roles (677); median stated pay $236K, from 248; most at NVIDIA (154), Amazon (24) and Together AI (11). NVIDIA's platform for programming GPUs directly, for the fast kernels that training and inference run on.CUDAJAX: named by 4% of roles (661); median stated pay $240K, from 303; most at Waymo (49), Amazon (44) and NVIDIA (36). Google's array library for high-performance training on TPUs and GPUs, common in frontier research.JAXRust: named by 3% of roles (583); median stated pay $234K, from 290; most at Perplexity (30), OpenAI (22) and Helsing (19). A fast, memory-safe systems language, for inference servers, tokenizers, data tools and AI infrastructure.RustTriton: named by 2% of roles (398); median stated pay $242K, from 148; most at NVIDIA (51), Amazon (15) and Cerebras (8). A Python-like language for writing GPU kernels without hand-written CUDA, for fast training and inference.Triton
The figures as a table
SkillOpen rolesShareMedian stated payStating pay
Python8,51248%$208K3,216
LLMs6,97839%$215K2,680
Evals3,96422%$220K1,810
AWS3,68121%$203K1,191
PyTorch2,91616%$224K1,098
RAG2,69315%$197K876
Azure2,41814%$195K753
Kubernetes2,31313%$213K857
C++2,30813%$213K820
SQL2,24913%$189K819
GPU2,13312%$230K808
GCP2,10112%$205K679
Distributed Systems1,81810%$229K848
TensorFlow1,5048%$210K506
TypeScript1,4368%$208K617
RL1,2797%$250K557
Spark1,1156%$216K442
Go1,0516%$221K528
React9515%$202K384
Terraform9155%$205K314
CUDA6774%$236K248
JAX6614%$240K303
Rust5833%$234K290
Triton3982%$242K148

Often asked for together

For the most-named skills, the skills their roles name far more often than AI roles do as a whole: the share of that skill's roles naming each one, beside its share of all of them. Each pair is named together by at least 15 roles at five or more employers.

Pay by work area and level

From the 6,242 open roles (35% of 17,736) whose posting states an annual range in US dollars. Each role (employer and title) counts once, at the midpoint of its range, averaged over the places it is posted. Low and high are the 25th and 75th percentiles. Fewer than 10 roles show no figures.

Pre-training

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels68$250K$283K$343K
Mid38$250K$275K$412K
Senior8too few to say
Staff+21$280K$290K$315K
Manager1too few to say

Post-training & RL

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels626$209K$250K$298K
Intern4too few to say
Junior5too few to say
Mid270$200K$240K$310K
Senior102$200K$225K$252K
Staff+208$228K$265K$297K
Manager15$264K$285K$318K
Director14$253K$298K$337K
Executive8too few to say

Inference & serving

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels825$200K$236K$275K
Intern3too few to say
Junior5too few to say
Mid282$180K$225K$275K
Senior197$188K$213K$238K
Staff+276$220K$251K$291K
Manager26$260K$276K$316K
Director25$231K$266K$285K
Executive11$249K$280K$317K

Model evaluation

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels598$190K$229K$272K
Junior1too few to say
Mid205$170K$222K$265K
Senior157$180K$208K$231K
Staff+190$215K$245K$280K
Manager27$241K$278K$312K
Director10$241K$262K$287K
Executive8too few to say

Agents

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels2,499$171K$212K$254K
Intern4too few to say
Junior18$100K$133K$147K
Mid851$154K$194K$240K
Senior641$168K$196K$220K
Staff+695$210K$240K$275K
Manager104$198K$242K$282K
Director132$209K$243K$280K
Executive54$213K$269K$304K

Safety & interpretability

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels157$175K$234K$305K
Mid70$171K$237K$346K
Senior32$151K$182K$288K
Staff+41$209K$249K$280K
Manager6too few to say
Director6too few to say
Executive2too few to say

Data for AI

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels155$214K$250K$298K
Mid64$205K$250K$340K
Senior39$202K$225K$252K
Staff+41$240K$275K$309K
Manager4too few to say
Director6too few to say
Executive1too few to say

GPU & kernels

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels249$208K$242K$281K
Junior3too few to say
Mid99$198K$238K$278K
Senior45$195K$220K$237K
Staff+96$230K$266K$300K
Manager6too few to say

Multimodal & vision

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels543$183K$220K$265K
Intern7too few to say
Junior4too few to say
Mid201$160K$200K$262K
Senior159$189K$214K$238K
Staff+136$220K$256K$296K
Manager14$231K$274K$306K
Director18$239K$271K$289K
Executive4too few to say

Retrieval & recommendations

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels675$163K$199K$235K
Intern1too few to say
Junior6too few to say
Mid180$139K$168K$203K
Senior208$161K$185K$210K
Staff+211$198K$226K$262K
Manager23$180K$202K$259K
Director32$208K$229K$248K
Executive14$198K$249K$295K

Robotics & autonomy

LevelRolesLowMedianHigh25th to 75th percentile, median marked
All levels667$182K$214K$250K
Intern5too few to say
Junior10$121K$129K$142K
Mid259$165K$200K$227K
Senior193$191K$210K$235K
Staff+158$225K$253K$287K
Manager27$256K$275K$286K
Director9too few to say
Executive6too few to say

Pay in US visa filings

From 13,983 certified H-1B, H-1B1 and E-3 applications for these roles, FY2025–FY2026: the base wage offered, yearly, without bonus or equity. Employers must offer at least the prevailing wage, and many that don't state pay in postings file here. Applications from H-1B-dependent employers, and for workers placed at clients, are left out (mostly IT staffing: they pay their own rate, not the role's). Low and high are the 25th and 75th percentiles; fewer than 10 applications show no figures. How they're read.

AI roles

LevelApplicationsLowMedianHigh25th to 75th percentile, median marked
All levels13,983$145K$175K$206K
Intern1too few to say
Junior148$84K$105K$141K
Mid8,818$139K$170K$197K
Senior2,925$150K$180K$209K
Staff+1,245$173K$214K$250K
Manager278$160K$198K$249K
Director289$172K$213K$250K
Executive279$162K$190K$226K

How roles are chosen: methodology.