Original title: The Physical Turn: How Edge Robotics Reclaims the Real World
Author: Elena Rostova
Publisher: MIT Press, 2026, 312 pages
In The Physical Turn, Elena Rostova puts forward an audacious thesis: after ten years of purely software-based AI, innovation is returning to the physical world. According to this MIT researcher, the $4.35 billion injected in 2025 into physical robotics marks more than just a new investment cycle. It signals that the true economic revolution will come from machines that manipulate matter, not from algorithms in the cloud.
This shift toward edge robotics could resolve the Solow paradox that still haunts the American economy: how can spectacular productivity gains in AI fail to translate into macroeconomic growth? Rostova has an answer: as long as artificial intelligence remains confined to screens, it cannot transform the sectors that carry the most weight in the real economy.
The Essentials
- Investments in physical robotics surged 335% between 2022 and 2025, with Figure AI raising $1 billion and Physical Intelligence $400 million
- Edge robotics already represents 15% of total AI investments, compared to 3% in 2022
- According to Rostova, this shift explains why American manufacturing productivity has grown three times faster than services since 2024
- The author predicts a golden age for heavy industry in the next five years, driven by autonomous machines capable of repairing, assembling, and building
The Author
Elena Rostova has directed the Embodied AI laboratory at MIT since 2019. A former engineer at Boston Dynamics and researcher at DeepMind, she has published fifteen articles in Nature Robotics on reinforcement learning applied to physical tasks. Her career has taken her from theoretical algorithms to the concrete challenges of robotic manipulation in uncontrolled environments.
Rostova wrote this book after observing the relative failure of generative AI’s promises in business. Despite impressive linguistic capabilities, ChatGPT and its competitors struggle to sustainably increase productivity beyond simple cognitive tasks. The author sees this as proof that AI disconnected from the physical world remains fundamentally limited.
The Central Thesis: AI Must Reclaim a Body
“Intelligence without embodiment is merely a sophisticated parlor game,” writes Rostova. Her central thesis fits in one sentence: generative AI has reached its economic limits because it cannot act on the physical world. The most significant productivity gains will come from machines capable of seeing, touching, manipulating, and repairing in real environments.
The author relies on investment data to support her argument. In 2025, edge robotics captured $4.35 billion in private investments, a 335% increase since 2022. Figure AI, which develops humanoids for industry, raised $1 billion. Physical Intelligence, specializing in AI for object manipulation, collected $400 million. These amounts now exceed investments in pure conversational AI.
This financial shift reflects, according to Rostova, a growing realization: the real economy depends on the physical transformation of matter. Agriculture, construction, manufacturing, and logistics represent 40% of global GDP but have barely benefited from advances in AI. “We have created systems capable of writing perfect emails but incapable of screwing in a bolt,” the author summarizes.
Manufacturing Productivity as Empirical Evidence
Rostova draws on American productivity statistics to validate her thesis. Since 2024, manufacturing productivity has grown at an annual rate of 4.2%, compared to 1.4% for services. This divergence, the largest since the 1990s, coincides with the arrival of the first collaborative robots equipped with AI in factories.
The author documents three emblematic cases. At Tesla, the new Optimus robots assemble batteries 30% faster than traditional automated assembly lines. In Amazon warehouses, robotic arms piloted by AI sort packages twice as efficiently as barcode-based systems. At BMW in Germany, welding robots automatically adapt to variations in parts without human reprogramming.
These examples illustrate what Rostova calls “the physical advantage”: unlike pure algorithms, intelligent robots improve their performance through interaction with their environment. They learn from every manipulation, accumulate experience with materials, and develop expertise that software systems cannot replicate.
The Predicted Industrial Golden Age
Rostova’s forward-looking analysis sketches a historical reversal. After forty years of tertiarization in developed economies, Industry 4.0 would make heavy industry attractive again. Autonomous robots would eliminate the labor constraints that pushed production toward Asia.
The author cites reshoring projects announced by Intel, Apple, and GM. Intel plans to produce 80% of its chips in the United States by 2030, relying on fully roboticized fabs. Apple is testing autonomous assembly lines that would reduce dependence on Chinese factories. These relocations are becoming economically viable thanks to robots that work 24/7 without salary increases.
Rostova predicts this dynamic will extend beyond high-tech. Construction, the emblematic sector of productivity stagnation, could experience its first revolution since the invention of reinforced concrete. Bricklaying robots developed by Construction Robotics already lay 3,000 bricks per day, compared to 500 for a craftsman. In agriculture, John Deere’s autonomous harvesters optimize harvesting field by field using embedded AI.
The Blind Spots in the Analysis
Rostova’s demonstration contains gray areas that weaken her most optimistic conclusions. The author underestimates the regulatory and social obstacles to massive robot deployment. In Europe, the precautionary principle slows the authorization of autonomous machines in public spaces. In the United States, unions negotiate conversion clauses that hinder adoption.
The economic analysis also has gaps. Rostova extrapolates from limited examples without accounting for threshold effects. The productivity gains observed at Tesla or BMW concern specific tasks in controlled environments. Nothing guarantees they will generalize across the entire manufacturing industry, much less to services, which represent 70% of employment.
The author also neglects the geographical inequalities that her scenario would create. If robots relocate production to wealthy countries, what becomes of emerging economies that depend on manufacturing industry? AI is already widening the North-South gap before having delivered on its promises in developed countries. Physical robotics could exacerbate this divide.
Finally, Rostova glosses over the energy question. Autonomous robots consume ten times more electricity than traditional automatons. At a time when data centers are already saturating electrical grids, this additional consumption poses sustainability challenges that the book does not address.
Why Read It
The Physical Turn offers an original framework for understanding the next phase of AI. Unlike works that stop at cognitive capabilities, Rostova analyzes artificial intelligence as an economic and industrial phenomenon. Her approach through investments and productivity illuminates trends that purely technological analyses miss.
The book stands out for its empirical documentation. Rostova does not merely speculate: she compiles investment data, industrial case studies, and macroeconomic statistics to support each argument. This factual rigor often lacks in prospective essays about AI.
The work will particularly interest industrial executives seeking to anticipate mutations in their sectors. Rostova provides concrete reference points for evaluating the impact of intelligent robotics on value chains. Her analyses of Tesla, Amazon, and BMW constitute instructive case studies for any automation strategy.
Beyond the professional audience, the book illuminates a public debate often polarized between naive technophilia and technophobia. Rostova demonstrates that AI can create real economic value, but only if it moves beyond screens to transform the physical world. This nuanced perspective helps transcend the usual caricatures about artificial intelligence.