How far can we push computing?

In 1959, physicist Richard Feynman gave a lecture titled 'There’s Plenty of Room at the Bottom', arguing that physics allows manipulating matter at the atomic scale. He claimed no physical law prevents writing the entire Encyclopædia Britannica on a pinhead or storing all human knowledge in a speck of dust. Feynman’s ideas inspired decades of progress in miniaturization, proving that biology already achieves such feats—cells store genetic instructions in tiny DNA strands. His vision set the stage for modern computing, where transistors, the tiny switches powering chips, shrank dramatically, enabling computers to evolve from room-sized machines to devices small enough to wear. Feynman even offered $1,000 prizes to demonstrate these possibilities, one of which was claimed by an engineer who built a motor the size of a grain of sand using only a toothpick and a microscope.

Moore’s Law, named after Intel co-founder Gordon Moore, observed in 1965 that the number of transistors on a chip doubles roughly every two years, driving rapid improvements in computing power. For decades, this trend fueled exponential growth in chip performance, but it has now slowed significantly. Around 2004, another principle called Dennard scaling ended; it had ensured that smaller transistors used less power while running faster. Today, progress relies more on brute-force methods—adding more chips, memory, and power—rather than shrinking transistors. This shift is evident in artificial intelligence, where the compute power used to train AI models has doubled every six months since 2010, far outpacing Moore’s Law. However, this progress comes at a cost: data centers, operated by tech giants like the six largest hyperscalers, are spending over $700 billion in 2026 to expand infrastructure, straining electrical grids and relying on scaling up rather than innovation.

The article outlines four fundamental physical limits to computing, derived from quantum mechanics, general relativity, and information theory. The first limit, set by quantum mechanics, caps computational speed. For a one-kilogram computer (about the mass of a laptop), physicist Seth Lloyd calculated in 2000 that the ultimate speed is 5 × 10^50 operations per second—35 orders of magnitude faster than today’s leading chips. The second limit, from thermodynamics, was discovered by Rolf Landauer in 1961: erasing a single bit of information must release a minimum of 2.9 × 10^-21 joules of heat at room temperature. Today’s chips waste hundreds of thousands of times more energy per operation. The third limit, from general relativity, was found by Jacob Bekenstein in 1981: the amount of information a given space can hold is capped by its size and energy. For Lloyd’s ultimate laptop, this limit is about 10^42 bits. The fourth limit, from information theory, was determined by Alexander Holevo in 1973: the maximum data rate for a one-watt light signal is about 10^17 bits per second, far exceeding today’s optical links.

The largest gap between current technology and fundamental limits lies in computational speed, with a 35-order-of-magnitude difference—equivalent to the time between a nanosecond and 200 million times the age of the universe. The energy gap is smaller but still significant: today’s chips operate about six orders of magnitude above the Landauer minimum, the theoretical lowest energy required to erase a bit. Memory density also faces a gap of nearly 30 orders of magnitude compared to Bekenstein’s limit. Quantum computers, which manipulate quantum bits (qubits), offer a partial solution by potentially using every degree of freedom in matter for computation. However, current quantum computers are inefficient, requiring trillions of atoms per qubit and thousands of qubits to achieve reliability. Bridging these gaps will require entirely new approaches to computing, as traditional methods are approaching their physical boundaries.