Physical AI: Value should always come first

I feel privileged to have presented at the Google DeepMind Accelerator Demo Day event earlier this week in London 🇬🇧 and that Acumino (www.acumino.ai) is part of such an amazing Physical AI Accelerator cohort. The key messages from my presentation are:

• Reliability needs to come first in every optimization that we run. Reliability unlocks performant robots and guarantees ROI. I would compromise some generalization capability for reliability every single time. You cannot have both.

• Frontier labs remain model-first constrained. Massive capital blinds them. Super cool work (I am a big fan) but a model alone cannot unlock deployments. Control, model selection and orchestration are key and these need customer feedback. The one-model-fits-all approach is in most cases like using a bazooka to kill a mosquito.

• The Physical AI revolution will be triggered by real, scalable value generation for real customers in actual production lines and 100% in factories first. The demand is unprecedented. We need to showcase the substance behind the promise that Physical AI brings. This is best showcased with REAL deployments at a MASSIVE scale, not with the 10th announcement of an even more promising model. Patience is limited when value is minimal.

• This is Space Race #2 and not the bubbly non-physical AI race. This will determine how our planet and society will look like for the decades to come. It can lead to abundance and prosperity.

The strategy for this race is clear. Value first. Minimize friction to market. This is like a giant complex puzzle where all the pieces are more or less already there and we just need execution discipline and the right people to put it together. 🚀