Increasing the storage capacity of lithium electrodes, without altering their cyclability, is one of the key challenges for modern ion-based batteries. For graphite-based anodes, the well-known capacity limit is ∼370 mAh/g, which corresponds to a lithium composition of Li1C6. Lithium intercalation is accompanied by a volume expansion of ∼10%. In the present work, accurate first-principles methods are used to investigate the performance of different bulk sp2 carbon allotropes as anodes in lithium-ion batteries. Compared to graphite, which is an alternated stack of graphene layers (Bernal stacking) exhibiting a perfect hexagonal tiling, the layers of the other stacked systems considered are constructed from various polygonal carbon rings, such as squares, pentagons, hexagons, heptagons, octagons, and dodecagons. These sp2 allotropes, which appear locally in defective graphene and grain boundaries, can exhibit a substantial increase in specific capacity with respect to graphite (up to a factor of two, i.e., Li2C6) with only a relatively small volume expansion (at most 25%). The mechanisms for this predicted increase in the number of lithium atoms that can be hosted in these still hypothetical carbon crystals are analyzed in detail, yielding global strategies for improving lithium capacity in sp2 carbon-based batteries. In addition, these results offer an insight on the local mechanism of Li incorporation in randomly defective graphite.