jim fan

T3 · host / generalist

Jim Fan leads Nvidia's embodied AI/robotics research group. He joined as an intern in 2016 when Jensen Huang delivered the first DGX-1 to OpenAI, and has driven the 'great parallel' strategy applying LLM scaling laws to robotics through world models, simulation, and massive RL.

5 calls·5 names·100% bull·last heard 3 months ago·Sequoia Capital
track record

no scored calls yet — needs a stated position or a categorical verdict, with a matured window vs SPY

top calls

highest conviction · one per company
1sthigh conviction
$NVDANvidiaposition

Nvidia building full-stack robotics platform following LLM playbook with compute flywheel

Nvidia is executing the 'great parallel' to LLM success in robotics: world model pre-training (Dreamer, Ego-Exo), action fine-tuning, neural simulation (Dream Dojo), and massive RL compute — with Jensen Huang's endorsement that 'the more you buy, the more you save' on compute-environment-data equivalence.

2ndhigh conviction
$TSLATesla

Tesla FSD holds the winning data flywheel model for robotics at scale

Tesla's FSD demonstrates the only proven scalable data collection paradigm: ambient, frictionless human driving data uploaded automatically at millions of hours per year — the 'FSD equivalent' that robotics must replicate for dexterous manipulation.

3rdmedium conviction
$JOLESJoles

UMI research spawns unicorn Joles commercializing wearable robot data collection

The Universal Manipulation Interface (UMI) research from Nvidia/Stanford spawned Joles, a unicorn startup improving the wearable gripper design for high-fidelity human demonstration data collection at scale.

5 total
$JOLES
Joles
MEDjim fan·Sequoia Capital·3 months ago
UMI research spawns unicorn Joles commercializing wearable robot data collection
The Universal Manipulation Interface (UMI) research from Nvidia/Stanford spawned Joles, a unicorn startup improving the wearable gripper design for high-fidelity human demonstration data collection at scale.
"Yet, I will say UMI is perhaps one of the greatest papers ever written in robotics data, and it spawned two unicorn startups. On the left hand side is Joles improving this design,…"
9:09
$NVDA
···
Nvidia
HIGHjim fan·Sequoia Capital·3 months ago· position
Nvidia building full-stack robotics platform following LLM playbook with compute flywheel
Nvidia is executing the 'great parallel' to LLM success in robotics: world model pre-training (Dreamer, Ego-Exo), action fine-tuning, neural simulation (Dream Dojo), and massive RL compute — with Jensen Huang's endorsement that 'the more you buy, the more you save' on compute-environment-data equivalence.
"So, the new post-training paradigm for robotics is a massively parallel RL system that runs on a few real robot stations, a bunch of graphics cores running world scans, and heavy…"
16:32
$TSLA
···
Tesla
HIGHjim fan·Sequoia Capital·3 months ago
Tesla FSD holds the winning data flywheel model for robotics at scale
Tesla's FSD demonstrates the only proven scalable data collection paradigm: ambient, frictionless human driving data uploaded automatically at millions of hours per year — the 'FSD equivalent' that robotics must replicate for dexterous manipulation.
"Anyone driving Tesla or Waymo here? Anyone? Right? You know, when you're driving, you're actually contributing to the biggest physical data flywheel. And the beauty is you don't e…"
10:13
$GOOGL
···
Alphabet
MEDjim fan·Sequoia Capital·3 months ago
Waymo shares Tesla's data flywheel advantage in physical AI
Waymo's autonomous driving fleet creates the same category-defining physical data flywheel as Tesla, providing the scale of real-world interaction data needed to train generalist robot policies.
"Anyone driving Tesla or Waymo here? Anyone? Right? You know, when you're driving, you're actually contributing to the biggest physical data flywheel."
10:13
$SUNDAY
Sunday
MEDjim fan·Sequoia Capital·3 months ago
UMI research spawns unicorn Sunday building three-finger data gloves for dexterity
Sunday is the second unicorn from UMI research, developing three-finger data gloves that capture high-dimensional human dexterity for robot policy training without teleoperation bottlenecks.
"On the left hand side is Joles improving this design, so you can wear the gripper here, and on the right hand side, Sunday made these three-finger data gloves."
9:15