the power wall
For decades, each new chip generation simply ran at a higher clock speed, and software got faster for free — 1 megahertz became 100, then 1000, then thousands. Then, around 2005, the climb stopped dead. Clock speeds plateaued and have barely moved since. The reason they hit a ceiling is the power wall: you cannot keep pumping more power into a chip, because you cannot get the heat back out fast enough without it melting.
The wall is set by heat. Dynamic power grows with frequency, and pushing higher frequency needs higher voltage, where power grows with voltage squared — so chasing clock speed makes power and heat climb faster than linearly. A chip can only dissipate so many watts through its package and cooler before the silicon overheats; a typical desktop limit is on the order of 100-some watts. Past that point a faster clock would cook the chip. So frequency stopped rising not because we forgot how to make transistors switch faster, but because we could no longer afford the power to run them all flat out.
The power wall is one of the two great turning points (with the end of Dennard scaling) that reshaped all of computing, and getting the causation right matters. Because a single core could no longer be clocked faster within the power budget, designers turned to putting many cores on a chip (multicore) and to specialized accelerators that do more work per joule (domain-specific architectures, GPUs, TPUs). The free lunch of automatically faster serial programs ended; from then on, more performance had to come from parallelism and specialization, which software must be rewritten to exploit. The power wall is why your laptop has eight cores instead of one core at 30 gigahertz.
Imagine wanting a 30 GHz single core. To clock that high you would raise voltage; with power growing as voltage^2 times frequency, the dissipation would soar into the hundreds or thousands of watts — far beyond what any air or even water cooler can remove. So instead of one 30 GHz core, vendors ship eight cores near 3 GHz that fit the same power budget.
The power wall caps clock speed by capping heat — which is precisely why multicore and accelerators replaced ever-faster single cores.
The power wall did not stop transistors from getting faster in principle; it stopped us from powering and cooling them all at full speed. That is the real, physical reason the 'free' clock-speed era ended around 2005.