I was going to disagree with you, but a back of the envelope calculation shows that the energy of computer production is probably equivalent to that of a few hundred hours of use, and so if you can save 10% on energy you are probably environmentally justified in switching to a new device. This is very counterintuitive to me.

To first order the upper bound for carbon cost of production/recycling is going to be a multiple of the energy required to melt the components. Call that multiple four: production of raw materials + reshaping in factory + recycling of materials at EOL + energy cost for assembly. So a rough upper limit of the energy to switch a computers would be four times the energy to heat the computer's mass in water (4 J/gC) up to the melting point of silica (1900 C), and a rough lower bound would be the same using the heat capcity and melting point of iron (10x lower heat capacity, 1500 C), and neglecting assembly. For 1 kg of material, that gives a range of 1.8 and 32 MJ. Now you can compare to the energy used. For myself, I'm on a 2 kg laptop that draws up to 250W, but conservatively 50W. That adds up to 32 MJ/kg in 400 h. If I can reduce the energy use by 5W with a new model, that will pay itself back in 2 years of workday computation. Crazy.

Caveat: there will be additional non-CO2 environmental costs. Most components are not recycled, and resource extraction is also environmentally damaging.