Richard Feynman’s 1959 Vision: The Lecture That Launched the Race to Engineer at the Atomic Scale

On December 29, 1959, physicist Richard Feynman stepped up at an American Physical Society meeting at Caltech and sketched a future in which engineers would work at scales that, at the time, sounded almost absurd. In a lecture later known as “There’s Plenty of Room at the Bottom,” he argued that physics did not forbid us from compressing vast amounts of information into microscopic spaces—down to the point of placing atoms exactly where we want them. If life could store a cell’s operating instructions in a slender strand of DNA, Feynman reasoned, then human technology should also be able to pack knowledge densely. The constraint wasn’t nature’s rules; it was our lack of tools.

For much of the following half-century, the industry seemed to validate his optimism with surprisingly little resistance. The transistor—the basic on/off switch underlying modern chips—kept shrinking. As it did, more transistors fit into the same area, and computers became simultaneously smaller, faster, and cheaper. Machines that once required a room of hardware moved onto desktops, then into pockets, and ultimately onto wrists.

That era of comparatively “easy” progress is now faltering. One of the most visible symptoms is the explosive growth of data centers. Six of the largest hyperscalers—companies operating the biggest cloud and AI computing facilities—are projected to spend more than $700 billion this year on computing infrastructure. That figure is about double the record they set in 2025. The investment is pouring into new buildings and equipment—concrete, copper, and silicon—at a pace that is pushing against how quickly electrical grids can supply power.

This expansion delivers gains in the most straightforward way possible: by deploying more of the hardware we already know how to build. The quieter problem is that the kind of progress that historically made each generation of chips dramatically better is slowing.

Several famous “laws” that once described exponential improvements are no longer keeping their old pace. Moore’s Law—the observation that transistor density on silicon rises predictably over time—has decelerated. Dennard scaling, which held that power density could remain roughly constant as transistors shrank, has broken down, leaving heat and energy consumption as hard constraints. Koomey’s Law, charting the long-term rise in computations per joule, is losing momentum. Kryder’s Law, which tracked rapid increases in magnetic storage density, has nearly stalled. And Keck’s Law, associated with the growth in how much information a single standard optical fiber can transmit, is flattening as well.

Yet these are not inviolable laws of nature. They are patterns that held while particular engineering approaches kept working. The deeper boundaries—the ultimate limits imposed by quantum mechanics, general relativity, and information theory—are still far beyond what today’s data centers confront. In effect, the scaling trends describe the distance computing has already descended into the small and the efficient; fundamental physics describes the remaining headroom.

Exploring those limits can still be practical, even if some are unreachable for ordinary industry. Attempts to build proof-of-concept systems near a physical boundary can do double duty: they can break open stalled technological paradigms while also functioning as experiments that test our understanding of the universe.

Feynman himself tried to spur that kind of invention with a pair of $1,000 prizes, offered from his own pocket. One reward would go to whoever could build a working electric motor smaller than a cube just 0.4 millimeters on a side. He expected the feat would require a fundamentally new microtechnology. Instead, about a year later, Caltech-trained engineer William McLellan arrived with a tiny motor that ran. The device weighed about a quarter of a milligram, was roughly grain-of-sand sized, and had been painstakingly assembled by hand under a microscope using a toothpick and a watchmaker’s lathe.

Why it matters

Computing is entering an era where more capability increasingly comes from scale—massive capital spending and electricity-hungry infrastructure—rather than effortless chip-level gains. That shift affects everyone who relies on cloud services and AI, from businesses to consumers, because costs and energy constraints become central. The next breakthroughs may come from new architectures and physics-driven approaches that move beyond the familiar scaling playbook.