A few premises

a) Some animals matter

b) Not all animals matter, some extremely simple animals don't matter.

c) There are anti-correlations of the form: the more cows there are, the fewer insects (or rodents) there are. - that hold true in our world in which one of the species is substantially more cognitively capable than the other.

d) There is currently no consensus on how simple a mind or cognitive system has to be for it to matter. Arguably this consensus cannot be reached, since different hypotheses will use different correlates to try and find the morally worthy thing.

Conclusion

e) We do not know now, and won't know in the medium future whether increasing or decreasing consumption of farm animals is desirable from an utilitarian perspective.

I don't know where this argument fails. I've shown it to many EA's and no one saw a big problem so far. However, some people think this is just stupid, and I'm happy to see it proven wrong.

To say the same in less abstract form: World 1 and World 2 have the same amount of land. In World 1, people eat cows and raise cows, so there are 100 cows and 1000 insects. In World 2 people are vegan and there is more forest land, World 2 has 10000 insects and no cows. Which world is ethically better seems to hinge on the comparative moral worth of insects and cows. Given we don't know what it's like to be a bat, or a cow, or a bumblebee, we cannot decide which world is ethically more desirable. Therefore we have no reason to direct our actions to make our world more like 1 or 2.

You can change insects (or rodents) and cows for any pair of animals that are anti-correlated in nature, and cognitively dissimilar, and where the size of the anti-correlation is larger than your certainty about which animal is more morally worthy.