The Physical Pivot: Why Intelligence is Moving Beyond the Screen

The current AI narrative is stuck in a digital loop, focused on tokens and chatbot efficiency. However, the real economic shift is happening in the physical world, what Morgan Stanley analyst Adam Jonas calls physical AI. This move from software intelligence to robotics represents an 8x to 10x multiplier on global GDP, yet it remains tied to a fragile, China-centric supply chain. The takeaway is that the next decade of competitive advantage will not go to the company with the most sophisticated large language model, but to those that master the onshoring of manufacturing and energy density. Readers who recognize this shift from digital tokens to physical atoms will be better positioned to identify the next generation of industrial infrastructure winners.

The Hidden Cost of the Digital-Only Trap

While the market obsesses over the cost of tokens, such as the 40% cost reduction Anthropic targeted with its Opus 5.5 release to make AI viable for enterprise, this is just optimizing the status quo. The real systemic risk is the assumption that AI infrastructure can stay separate from physical reality.

As Adam Jonas notes, the future of AI is not just cloud inference; it is inference at the edge, embedded in robots that require rare earth minerals and manufacturing ecosystems currently dominated by geopolitical rivals. The system is responding to this dependency by pushing for re-architected supply chains. This creates a paradox: to build the robot economy, the U.S. must diversify away from China while remaining entangled with it, because a total decoupling would be economically catastrophic.

> We think historians of the future are gonna write about this time now where were you when the robots came? Where were you when we really started to industrialize and make use of infrastructure in space?
>
> -- Adam Jonas, Morgan Stanley

The 18-Month Payoff: Why Screenless is the New Frontier

The shift toward physical AI is mirrored in consumer technology. Apple’s reported move toward a screenless fitness tracker is a direct response to the limitations of current wearables. By removing the screen, manufacturers can optimize for battery life, sensor accuracy, and physical form factor, shifting the value from digital interaction to passive health monitoring.

This mirrors a broader industry trend identified by Bessemer’s Sameer Dholakia: intelligence is becoming the new oil. Just as oil’s initial use in kerosene lamps masked its eventual role in powering the entire global economy, current AI use cases are just the lighting phase. The long-term payoff belongs to those building the infrastructure, the last mile vertical applications and the physical agents that will eventually handle the majority of the world’s inference work.

> I think that intelligence as a market would be bigger than all of, it is like oil. Like oil powers all the machines in the world, intelligence will power all the humans and every organization in the world.
>
> -- Sameer Dholakia, Bessemer Venture Partners

Where Immediate Pain Creates Lasting Moats

The cybersecurity sector provides the clearest example of how fast the system is adapting to these new risks. Enterprises that previously estimated they had 300 AI agents discovered they had 18,000. This is not a bug to be patched; it is a fundamental expansion of the attack surface.

The companies winning here are not the ones fighting the models, but the ones providing the telemetry and runtime security required to manage them. As Catherine Trebnick points out, the AI apocalypse for legacy security firms never happened. Instead, platform players like CrowdStrike and Palo Alto Networks are thriving by acting as the essential safety layer for frontier models. They are capturing value by solving the immediate pain of agent sprawl, a problem that most organizations did not realize they had until it impacted their bottom line.

Key Action Items

• Audit Your Agent Surface Area: If you are an enterprise leader, stop tracking software seats and start tracking AI agents. The gap between what you think you have and what is actually running on your network is likely an order of magnitude. (Immediate)
• Prioritize Edge Infrastructure: Over the next 12 to 18 months, shift focus from cloud-only AI investments to edge-compute capabilities. The companies that can perform inference locally, without relying on high-latency, high-cost cloud calls, will hold the competitive advantage in the physical AI era.
• Re-evaluate Supply Chain Exposure: For investors and operators, stress-test your dependency on critical minerals and rare earths. The onshoring trend is not a political talking point; it is a necessary hedge against the inevitable regulation of physical AI. (12 to 24 months)
• Look Beyond the Humanoid Hype: Don't be seduced by the humanoid form factor, which is primarily a recruiting and capital-raising tool. Focus on the thousands of other types of terrestrial and industrial robots that will actually perform the bulk of the world's physical work. (Long-term)
• Adopt Disciplined Conviction: As capital pools grow, such as Bessemer’s $6B fund, the temptation is to spray-and-pray. Focus on companies building the last mile of vertical AI, the tools that solve specific, high-stakes industry problems rather than general-purpose agents. (Ongoing)