That was a different situation. Tesla cancelled the S40 before any of the cars were made. For the very small number of preorders for the S40 they provided a software locked 60 rather than cancelling the orders.
Indeed, as I alluded to. But they've done it in other cases, as well. 60 to 75. AP hardware on all cars. There's a cost for those who don't buy the upsell, but there are also benefits to Tesla. Data in the AP hardware case.
In this case, it's increased cell volume / lower per-cell cost, plus simplifying the line while working a ramp that they're having trouble ramping to plan.
Again, I don't think they'd do this here, as even if you assume Tesla's got a 50% margin on the LR battery, that's a $4500 hit for anyone who doesn't buy the upsell. But the cost isn't as high as many seen to think. For example, let's say that was their margin, and they save say 10% in cell volume margin improvements plus a few hundred bucks/car by simplifying/improving throughout of the line via not changing out pack size during production. Further, assume 15% of SR owners upgrade during ownership. Tesla would charge a grand more for the upgrade than at build time, so that's a $10k charge against $4500 cost on 15% of SR vehicles. $5500 addl profit on 15% of cars. Say, $350 improvement to margin on *all* cars by simplifying the line and increasing volume.
Let's assume the steady state production is evenly split between SR and LR. This would mean Tesla's cost before upsell is $4150 on SRs ($4500 - $350 line savings) and additional profit of $350 on LRs. Net of $1900/car across the 3 line.
Now add back the $10k from 15% of SR buyers, or 7.5% of the fleet. That's $750/car. Net cost is now $1150/car across the line.
Now, some portion of those cars will be traded in, and Tesla resells them all as LRs, making back further some of the up-front cost (realistically all of it for the vehicles they get back as trades, as LRs will surely sell for $4k more than SRs). If 40% of 3s are traded back to Tesla, we've now made additional profit across the line.
So what at first looks like a $4500 hit on each SR car may very well be a net-even or long-term profitable proposition.
These numbers were pulled from thin air, are likely wrong, etc. I don't think they'll do this, as I said above. But don't toss it out as a crazy thought as the long-term cost is far lower than it seems when just taking the $9k sticker price and assuming Tesla eats it.