The ceiling and the floor.

In 1959, Richard Feynman gave a lecture with a famous title: There’s Plenty of Room at the Bottom. In the new issue of MIT Technology Review (Nov/Dec 26), Volkan Gurses (MIT and Nvidia) revisits that phrase to ask a question that anyone working in digital strategy should keep on their desk: how far can we push computing?

His answer has two sides. And it helps explain why the debate about data centers, energy, and technological sovereignty is often framed poorly.

The descent has stopped

For decades, progress in computing came almost for free. Transistors halved in size, and chips doubled in density. Then the empirical laws describing that progress began to bend. Moore’s law is slowing down. Dennard scaling ended around 2004, when clock speeds plateaued above 3 GHz. Koomey’s, Kryder’s, and Keck’s laws, concerning energy efficiency, memory density, and fiber capacity, are flattening out.

Today, progress is bought mainly by building more. According to the article, six major hyperscalers are on track to spend more than $700 billion on computing infrastructure this year, roughly twice the 2025 record. The compute used to train AI has doubled approximately every six months since 2010, much faster than Moore’s law would allow. It is brute force: concrete, copper, silicon, megawatts.

But the floor is still a long way down

The second message is the most interesting. Scaling laws are observations about systems we have engineered, not laws of physics. The actual physical limits, set by quantum mechanics, relativity, and information theory, lie much further down.

• Energy: since the 1940s, energy per operation has fallen by 17 orders of magnitude. We are still several hundred thousand times above the Landauer limit.
• Communication: optical interconnects are about a million times away from the Holevo limit. This is probably the first threshold the industry will approach.
• Memory: almost 30 orders of magnitude separate today’s devices from the Bekenstein bound. DNA stores one bit in about 30 atoms, but writing it is still too expensive.
• Speed: 35 orders of magnitude separate us from what quantum mechanics would allow a one-kilogram computer to do.

Three paths, not two

Gurses distinguishes between a ceiling and a floor. The ceiling is how much we can build: chips, energy, capital, land, cooling water. The floor is how much a joule, a second, and a gram of matter can compute. In the last century, we followed both paths together. Today, we are following almost exclusively the first, and the ceiling is coming into view.

The third path is to start with a physics experiment operating close to the floor and scale it up. It happened with the transistor, the laser, and giant magnetoresistance. It is happening with reversible logic: in 2025, the startup Vaire Computing tested its first chip, which recovers some of the switching energy instead of dissipating it.