AI infra hiring trends

Market intelligence observed directly from 83 companies' job boards — no scraped aggregators, no estimates. Counts start at our first crawl on 2026-08-17 and accumulate daily (last snapshot 2026-10-16).

1427
open AI-infra roles
83
companies hiring
+106
new, last 7 days
−98
closed, last 7 days
43d
median posting age

Open roles over time — xAI

Published listings at each UTC day boundary; today is provisional. Source coverage, review, and classifier changes also affect this total. Gaps mean no verified snapshot is available.

Open AI infrastructure roles over time0204060Aug 17Aug 24Aug 31Sep 7Sep 14Sep 21Sep 28Oct 5Oct 12

First published vs source closures, weekly

First publication on this site versus previously active listings confirmed missing from source boards. Excludes reopens and classifier, review, and deduplication removals; these flows do not explain all changes in the published total.

Weekly new vs closed roles — xAI0204060Aug 17 new: 41Aug 17 closed: 0Aug 17Aug 24 new: 1Aug 24 closed: 1Aug 24Aug 31 new: 10Aug 31 closed: 2Aug 31Sep 7 new: 6Sep 7 closed: 1Sep 7Sep 14 new: 4Sep 14 closed: 1Sep 14Sep 21 new: 1Sep 21 closed: 2Sep 21Sep 28 new: 4Sep 28 closed: 1Sep 28Oct 5 new: 6Oct 5 closed: 2Oct 5Oct 12 new: 26Oct 12 closed: 3Oct 12
newclosed

Who is hiring

Live open roles per company, with change over the last 7 and 30 observed days. Click a sparkline to filter the charts above.

companyopenΔ7dΔ30d
Amazon (AWS)hyperscaler231-8—
Nebiusneocloud104-2—
CoreWeaveneocloud72+2—
Nscaleneocloud72+1—
Anthropicfrontier lab69+2—
OpenAIfrontier lab68-2—
Cerebraschip vendor64-3—
Crusoeneocloud58-4—
FluidStackneocloud54+1—
xAIfrontier lab51+7—
Etchedchip vendor42-3—
IRENneocloud41+3—
Together AIneocloud34-3—
Mistral AIfrontier lab33+3—
Lambdaneocloud310—
SambaNovachip vendor29-1—
Tenstorrentchip vendor26+2—
TensorWaveneocloud23+2—
DigitalOceanneocloud22-3—
Baseteninference provider210—
Inferactinference provider200—
Lightning AIai startup190—
Thinking Machines Labfrontier lab17+2—
Coherefrontier lab16-1—
Fireworks AIinference provider130—
Luma AIai startup120—
Modalinference provider12-1—
Radiantneocloud12+2—
Prime Intellectneocloud11-2—
DeepInfrainference provider9-2—
Anyscaleai startup8+1—
NexGen Cloudneocloud7+1—
Perplexityai startup70—
River AIfrontier lab70—
FriendliAIinference provider60—
Parasailinference provider60—
Reactorinference provider60—
Cursorai startup50—
falinference provider5-2—
Reflection AIfrontier lab5+1—
Runpodneocloud5-1—
Harveyai startup40—
Liquid AIfrontier lab40—
Runwareinference provider4-2—
Scale AIai startup40—
Black Forest Labsfrontier lab3-1—
Hedraai startup30—
Inceptionfrontier lab30—
Lila Sciencesai startup30—
San Francisco Compute Companyneocloud30—
DoorDashenterprise20—
ElevenLabsai startup20—
Figureai startup20—
Fish Audio (39 AI)ai startup20—
Generalistai startup20—
Genesis Molecular AIai startup20—
Matter Intelligenceai startup20—
Novita AIinference provider2+1—
Relaceinference provider20—
Xaira Therapeuticsai startup20—
1Xai startup10—
Abridgeai startup10—
Boson AIfrontier lab1-1—
Cartesiaai startup10—
Cognitionai startup10—
Echo Neurotechnologiesai startup10—
Ineffable Intelligencefrontier lab10—
Interfazeinference provider10—
MongoDBai startup10—
Nimble Roboticsai startup10—
Periodic Labsfrontier lab10—
Physical Intelligenceai startup10—
Poolsidefrontier lab10—
Prior Labsai startup10—
Rekafrontier lab10—
Sakana AIfrontier lab10—
Scalewayneocloud10—
Sesameai startup10—
Sierraai startup10—
Skild AIai startup10—
Sunday Roboticsai startup10—
Sunoai startup10—
Waferinference provider10—

Category mix

Share of active listings tagged with each category. A listing can carry several tags in a dimension, so shares can exceed 100% combined.

Active roles by Workload stageinferenceinference: 393 · 28%393 · 28%distributed-inferencedistributed-inference: 216 · 15%216 · 15%pre-trainingpre-training: 129 · 9%129 · 9%evaluationevaluation: 113 · 8%113 · 8%fine-tuningfine-tuning: 102 · 7%102 · 7%reinforcement-learningreinforcement-learning: 87 · 6%87 · 6%post-trainingpost-training: 81 · 6%81 · 6%training-data-infratraining-data-infra: 45 · 3%45 · 3%
Share of Workload stage tagsinference — 393 (34%)distributed-inference — 216 (19%)pre-training — 129 (11%)evaluation — 113 (10%)other — 315 (27%)inference 34%distributed-inference 19%pre-training 11%evaluation 10%other 27%
inferencedistributed-inferencepre-trainingevaluationother

relative share of workload stage tags · snapshot 2026-10-16

Skill demand

Model-extracted skills in published listings. Unknown or stale extraction is not evidence that a skill is absent.

Top skills by share of active postingspython-langpython-lang: 652 · 46%652 · 46%kubernetes-opskubernetes-ops: 453 · 32%453 · 32%observabilityobservability: 356 · 25%356 · 25%cpp-langcpp-lang: 310 · 22%310 · 22%go-langgo-lang: 278 · 19%278 · 19%cudacuda: 212 · 15%212 · 15%infiniband-opsinfiniband-ops: 191 · 13%191 · 13%vllm-enginevllm-engine: 176 · 12%176 · 12%pytorch-distpytorch-dist: 171 · 12%171 · 12%rust-langrust-lang: 153 · 11%153 · 11%roce-netroce-net: 146 · 10%146 · 10%slurm-adminslurm-admin: 138 · 10%138 · 10%terraform-iacterraform-iac: 138 · 10%138 · 10%nccl-libnccl-lib: 120 · 8%120 · 8%sglang-enginesglang-engine: 118 · 8%118 · 8%

Kernel library demand

GPU kernel programming stacks named by active postings — which kernel library the market hires for. A role can name several.

Kernel library demand — currentCUDACUDA: 191 · 13%191 · 13%TritonTriton: 108 · 8%108 · 8%ROCmROCm: 27 · 2%27 · 2%CUTLASSCUTLASS: 24 · 2%24 · 2%CuTe DSLCuTe DSL: 4 · 0%4 · 0%TileLangTileLang: 2 · 0%2 · 0%
Kernel library demand — relative shareCUDA — 191 (54%)Triton — 108 (30%)ROCm — 27 (8%)CUTLASS — 24 (7%)other — 6 (2%)CUDA 54%Triton 30%
CUDATritonROCmCUTLASSother

Deep-learning framework demand

Frameworks active roles work with at a systems level — passing mentions don't count.

Deep-learning framework demand — currentPyTorchPyTorch: 304 · 21%304 · 21%JAXJAX: 118 · 8%118 · 8%TensorFlowTensorFlow: 87 · 6%87 · 6%MLXMLX: 1 · 0%1 · 0%
Deep-learning framework demand — relative sharePyTorch — 304 (60%)JAX — 118 (23%)TensorFlow — 87 (17%)MLX — 1 (0%)PyTorch 60%JAX 23%TensorFlow 17%
PyTorchJAXTensorFlowMLX

Accelerator hardware demand

Accelerator hardware active roles target. “Other” covers Inferentia, Cerebras, Groq, and custom ASICs.

Accelerator hardware demand — currentNVIDIA GPUNVIDIA GPU: 232 · 16%232 · 16%other accelother accel: 165 · 12%165 · 12%AWS TrainiumAWS Trainium: 140 · 10%140 · 10%Google TPUGoogle TPU: 75 · 5%75 · 5%AMD GPUAMD GPU: 27 · 2%27 · 2%Apple siliconApple silicon: 2 · 0%2 · 0%
Accelerator hardware demand — relative shareNVIDIA GPU — 232 (36%)other accel — 165 (26%)AWS Trainium — 140 (22%)Google TPU — 75 (12%)other — 29 (5%)NVIDIA GPU 36%other accel 26%AWS Trainium 22%Google TPU 12%
NVIDIA GPUother accelAWS TrainiumGoogle TPUother

Seniority mix

Listings with a detected seniority level.

Active roles by seniorityseniorsenior: 469 · 33%469 · 33%unknownunknown: 438 · 31%438 · 31%managermanager: 177 · 12%177 · 12%staff plusstaff plus: 174 · 12%174 · 12%midmid: 117 · 8%117 · 8%juniorjunior: 52 · 4%52 · 4%

Remote policy

Listings with a stated remote policy.

Active roles by remote policyonsiteonsite: 525 · 37%525 · 37%hybridhybrid: 168 · 12%168 · 12%remoteremote: 145 · 10%145 · 10%

Posting lifecycle

43d
median age of open listings
22d
median lifetime of closed listings
n=737
433
open 60+ days
long-lived listings — possible evergreen reqs

Methodology

Current-input extraction at the latest complete sample: 1387 / 1427 published jobs. Missing historical extraction remains unknown. 4 extraction versions in that sample.

Published totals are reconstructed from lifecycle events at UTC day boundaries; today is provisional. First publication is not the employer's posting date. Source closures exclude classification, human review, deduplication, and retirement changes. Reopens are not new publications. Adding sources or changing classifier versions can change measured coverage without a change in hiring demand. Skills and facets are model estimates sampled together, not reconstructed historical facts. Coverage: 83 companies — see the full list.