A high-end graphics card can now consume more power than an entire gaming PC did a decade ago. The growth is not carelessness; it follows from how chips have stopped improving.
Smaller transistors no longer save proportional power
For a long stretch, shrinking transistors let a chip use less energy per operation at the same rate, so each generation could do far more work inside the same power budget.
That relationship has weakened considerably. Newer manufacturing processes still pack more transistors into an area, but the energy each one needs falls far more slowly than it used to.
The consequence is that adding computation now mostly means adding power draw, because the efficiency gain that used to absorb it has largely gone.
Clock speed is an expensive way to buy performance
Running a chip faster requires a higher voltage to keep it stable, and power scales with the square of that voltage rather than in step with it.
Pushing the last portion of a card's clock range therefore costs a disproportionate share of its total consumption for a modest gain in frames.
Manufacturers do it anyway, because reviews compare products at their factory settings and a small measured lead is commercially worth a large amount of heat.
Memory and data movement carry their own bill
Rendering at high resolution moves enormous quantities of data between the memory and the processing units, and moving data costs energy independently of the calculations performed on it.
Wider memory interfaces and faster memory types both raise bandwidth and both raise power, and the memory system on a large card is a meaningful fraction of the total.
This is why cutting resolution often reduces power draw more than cutting other settings. It reduces traffic, not just arithmetic.
Cooling dictates the physical size
Every watt consumed becomes heat that has to leave the card, and air cooling scales mainly through surface area and airflow rather than through cleverness.
That is why cards have grown to occupy three or four expansion slots and why heatsinks now extend well past the circuit board underneath them.
Case compatibility has become a genuine purchase constraint, and cards routinely need physical support to avoid sagging under their own weight.
The rest of the system has to keep up
Power supplies are now sized around brief spikes rather than average draw, because graphics cards can demand far above their rated figure for very short intervals.
A supply that handles the average comfortably can still shut the machine down when a spike exceeds its protection threshold, which looks like an unrelated crash to the owner.
Undervolting has become a common response. Reducing voltage slightly while keeping clocks near stock often cuts consumption noticeably while costing very little performance.