Thanks, this is exactly the comment I was looking for. In addition to the 70% loss due to the solar panel efficiency, I think we should also lose some efficiency in the conversion to light via leds (although I expect that’s much more efficient, perhaps at like 80%).

I’m curious what is physically possible, if we assume we can achieve the max possible efficiency. I’m guessing there’s behavior like a Carnot engine, and the energy transfer can only be up to ~86% efficient (but please correct me if I’m wrong!!). In that case, conversion from light to energy via solar panels -> conversation back to light via leds should be 0.86*0.86 = 73% efficient in best case. And the full effect should be (800/300)*0.73 = 1.94, about twice as good as growing plants with the sun’s direct light. I’m surprised that seems possible!

p.s. My efficiency guesses are probably wrong. Please correct me.

By using multiple junctions and stacking them, top one converting the most energetic photons, then the second-most etc, one can approach the theoretical limit of about 95% or whatever it is. However in practice it's very expensive and difficult as I understand. AFAIK the current state of the art is about 6 stacked junctions at around 60% efficiency, at great cost.

And as you say the LEDs aren't 100% efficient either, though both deep red and bright blue are among the most efficient, about 85% there.

So that leaves you with about 50% overall just from those two.