A closer look
Brett Adcock’s compute bet: what Figure still has to prove
Figure’s September compute agreement extends its data strategy. The next question is whether that spending translates into more capable, useful robots.
The big picture · Jensen Huang
Physical AI has a bigger toolkit. What would prove it works?
Jensen Huang’s vision has moved from a broad promise to a larger set of models and tools. We trace what changed—and what still needs evidence.
Read the analysis →From statement to evidence
A robot’s promise has three different tests.
01Can it do
the task?
the task?
02Can it repeat
the result?
the result?
03Does it make
economic sense?
economic sense?
RoboAngle’s framework for reading physical AI announcements.
Beyond the headline
All analysis →Promises & progress
Elon Musk’s Optimus timeline: production is only the first milestone
Read Musk’s Davos forecast alongside Tesla’s 2026 factory updates. Production capacity, useful work and public sales are different milestones.
The evidence test
What counts as a working robot? Reading the Figure–BMW evidence
A factory task, a training platform and a consumer promise answer different questions. A practical way to compare physical AI claims.
Follow the people. Question the claims.
All people →Brett Adcock
Training data, general-purpose capability and evidence from deployment.
View the timeline →Elon Musk
Production plans, useful work and the path to customer deployment.
View the timeline →Jensen Huang
Robot learning, simulation and the infrastructure behind physical AI.
View the timeline →Context matters. We connect public statements with earlier promises, technical evidence and what happens next. How we research and check our articles.