This makes sense. The next generation GPUs are expected to be made on TSMC 7nm/6nm fabrication technology which is expected to have only 20% greater transistors per unit area. To put things in perspective, when going from Turing to Ampere we had a 80% increase in transistors per unit area.
Also, power efficiency improvements are proportional to the square root of area. So sqrt(1.8) = 1.35 which is about right. Ampere was around 35-40% more power efficient than Turing.
Going from Ampere to Lovelace we will get sqrt (1.20) = 1.09. Only 9% improvement in power efficiency won't be enought to generate a generational improvement in graphics performance. So the only other option is to increase power consumption.
RTX 3080 has 8704 cores and consumes 320 watts. RTX 4090/Ti, if it has 18,432 will have more than double the cores but will consume less than double the power. The RTX 4070 should have comparable performance to 3090 but should manage it around 250-300 watts (lesser than 3090).