JEPA World Models by AMI Labs: status and timeline

JEPA World Models

Predicts world evolution from robot actions. Produces compact representations for planning, trained on internet video and minimal robot data.

ProtoStage 2 of 5

Seed funded March 2026. V-JEPA 2 released May 2026 with zero-shot robotics.

Updated 25 May 2026Checked 10 Oct0 updates this week

Milestones

Next · Early enterprise deployments in robotics and manufacturing

AMI Labs foundedDec 2025Complete.
$1.03B seed round closesMar 2026Complete.
V-JEPA 2 released with zero-shot robotics capabilityMay 2026Complete.
Early enterprise deployments in robotics and manufacturingTarget Q4 2026Current milestone.

Most important updates

AMI Labs publishes V-JEPA 2 with zero-shot robotic deployment25 May 2026
AMI Labs closes $1.03B seed round for JEPA world models10 Mar 2026

Upcoming

Q4 2026Early enterprise deployments in robotics and manufacturing (next)

Current obstacles

Generalizing beyond training environmentsModels trained on one set of objects or conditions may struggle with novel shapes, materials, lighting, or configurations without additional adaptation.
Computational requirements for deploymentReal-time video processing demands significant compute. Deploying on robot edge hardware requires model compression and optimization.

Physics limits

Prediction errors compound over timeErrors in early frame predictions propagate when forecasting many frames ahead. Real-time replanning is needed to correct drift.
Video model scale and latencyProcessing video at scale requires GPUs. Latency-sensitive robotics applications may need lighter models with reduced accuracy.

How it works

3 parts

Video training

Learning from observation

The model is pre-trained on over 1 million hours of internet video, learning visual representations without text or manual labels.

Action prediction

Forecasting consequences

Model predicts frame changes from actions. Accurate prediction demonstrates physical understanding.

Planning and control

Finding effective actions

The robot uses the model to plan: it simulates different action sequences and picks the one most likely to achieve the desired outcome.

Spec sheet

Spec	JEPA World Models
Seed round size	$1.03B^R (reported)
Post-money valuation	$3.5B^R (reported)

^R reported by the company

Update log

2 updates

Mon 25 May

Major: AMI Labs publishes V-JEPA 2 with zero-shot robotic deploymentOtherResearch

Tue 10 Mar

Major: AMI Labs closes $1.03B seed round for JEPA world modelsPressFunding

About AMI Labs

The team behind JEPA World Models

AMI Labs

JEPA world models, robotics, industry automation

Builds world models for robot learning using JEPA. Founded by Turing Award-winner Yann LeCun, the startup develops video-based models for robotics and industrial automation.

JEPA models predict world evolution from robot actions, learning physics from video without text.
Closed $1.03B seed round at $3.5B valuation in March 2026, backed by Bezos, NVIDIA, Samsung.
Targets robot learning, warehouse automation, and industrial reasoning. Offices in four continents.

amilabs.ai (opens amilabs.ai)Company profile →

Founded: 20251 yrs
Headquarters: France
Status: Private
Valuation: $3.5BprivateMar 2026
Raised: $1B1 round
Last round: Seed · $1BMar 2026
Works in: AIVideo & world models
Coverage: 1 program · 2 updateslatest 25 May 2026checked 10 Oct
Lead investors: Cathay InnovationGreycroftHiro CapitalHV Capital
People: Yann LeCun (opens en.wikipedia.org)Executive Chairman, Co-founderAlex LeBrunCEO, Co-founder

Spec

JEPA World Models

Seed round size

$1.03B^R (reported)

Post-money valuation

$3.5B^R (reported)