Switching FrictionWide moat

Switching friction is the sum of all the other CUDA threads made concrete at the moment of decision: when a customer weighs a rival chip, they must reckon not with the chip's price but with the total cost of leaving Nvidia's world. Rewriting and re-validating software, retraining engineers, rebuilding tooling, and accepting the risk that something breaks in production together form a wall of cost that a merely cheaper or faster chip must overcome before it wins the sale.

The friction is decisive because it reframes the entire comparison. A rival can offer a chip that is twenty percent cheaper and lose anyway, because the customer, totting up the migration cost and the risk, concludes the cheaper chip is not actually cheaper once everything is counted. Nvidia does not have to be the best value on the spec sheet; it only has to be close enough that switching is not worth the upheaval — a far easier bar to clear.

This is why Nvidia's pricing power is so robust in the near term. As long as the friction of leaving exceeds the savings of switching, Nvidia can hold premium prices without losing the customer, and the friction is largest for exactly the biggest, most complex, most deeply invested customers — the ones with the most software, the most trained people, and the most to lose from a botched migration.

For the owner, switching friction is where the abstract moat becomes a dollar figure on a customer's spreadsheet, and it is the mechanism that converts CUDA's accumulated advantages into pricing power. It holds firmly today; the watchful question is whether the various efforts to lower each component of that friction — Nvidia's 2024 licence terms already bar CUDA translation layers¹ — can, in time, add up to a wall low enough to climb.