As few as distribution genuinely requires, which is usually one to begin with and rarely more than three. Every additional chain multiplies the operational surface. Supply accounting, bridge or mint-burn controls, redemption paths and attestation scope all grow, while demand does not multiply with it. The chains worth adding are the ones where the holders you actually want already are, not the ones with an ecosystem grant.
The cost that is underestimated is supply accounting. A single-chain token has one authoritative supply figure. A multi-chain token has one per chain plus a canonical total, and every one of those has to agree with the reserve at all times. Reconciling that continuously, rather than at attestation time, is the work, and it does not get easier with practice.
Bridges are the sharpest edge. A token bridged rather than natively issued introduces a second trust assumption on top of the issuer's, and holders on that chain are exposed to the bridge's failure as much as to yours. Native issuance on each chain avoids it but multiplies the mint and burn controls that must be individually correct.
Redemption is where multi-chain issuance is genuinely tested. A holder on the least liquid chain must be able to redeem on the same terms as one on the busiest, or the peg fragments by chain and the token trades at different prices in different places. Which is a failure of the product even while the aggregate reserve is fully intact.
The disciplined sequence is to launch on one chain, reach the point where redemption is boring, and only then add a second. With the accounting and attestation extended before the first token is minted rather than after.