That only works out if there are enough investors willing to pay for those futures. If the new factory can make a billion drives but they only have 2 of those futures contracts sold (that is 200k drives) they don't build the factory. Remember too if they sell those contacts they are on the hook to deliver - if it is just investors they will accept the street value of 100k drives in 2028 but some of the people might be buyers demanding physical goods.

Every year a few farmers realize they are contracted to deliver more grain than they have in their bins and so have to buy some grain from someone else (often at a loss) just to deliver it. This isn't a common problem but it happens (most often the farmer is using their insurance payout to buy the grain - snip a very large essay on the complexities of this)

> If the new factory can make a billion drives but they only have 2 of those futures contracts sold (that is 200k drives) they don't build the factory.

But the AI companies are flush with cash and trying to buy everything, right? Why wouldn't they buy up as many futures contracts as the fab company needs to justify more fabs?

> Every year a few farmers realize they are contracted to deliver more grain than they have in their bins and so have to buy some grain from someone else (often at a loss) just to deliver it.

This is most commonly because they sold a futures contract for X bushels expecting to grow 2X but 75% of the crop failed and they only have 0.5X.

Semiconductor fab yields aren't as susceptible to how much it will rain next year and the companies generally have a pretty good idea of what their yields are for a given process node.