As Sifted's analysis suggests, the core tension here is between a scaling playbook borrowed from digital AI and a safety-first imperative demanded by the physical world. The data-first bet assumes that more data will eventually solve problems like reliability and 'out-of-distribution' operation, but this remains an unproven leap for systems that can cause real harm. The architecture-first approach, while arguably more prudent, may struggle to match the rapid capability expansion promised by its rival.
In our view, the ultimate commercial winners will likely need to synthesize both philosophies, finding a way to scale while maintaining the rigorous, verifiable safety that real-world operations require.