3 ms·

> There is a long list of customers who'd kill for a chance to get (additional) capacity of the leading edge node process for years at a time Is there? Seems l

> There is a long list of customers who'd kill for a chance to get (additional) capacity of the leading edge node process for years at a time Is there? Seems like we're heading for a recession and reduced demand - at higher prices I wouldn't be surprised existing customers cut back on existing capacity.

sangnoir 4y ago
> Is there? Seems like we're heading for a recession and reduced demand Perhaps lower demand for consumer devices[1], but at Google and Amazon's scale, getting more efficient Tensor cores, ARM processors, and AI accelerators for their respective clouds will save them some OpEx by requiring less power (and cooling). It may also save CapEx as well, by reducing the need to build additional datacenters by getting +10% more compute[2] with the existing DC footprint, power and cooling budgets. Server costs are amortized over long periods - longer than whatever recession may or may not happen. Also, if they can lock-in TSMC for years, they'll continue to reap benefits long after the recession is over and shift capacity to Kindles or Pixel devices when consumer demand recovers. 1. Which may explain Apple developing cold feet. 2. Back of envelope calculations. I do not know what node mixes Google and Amazon have in their data centers, but I'm guessing it averages-out between 10 to 15nm. How much can they benefit by replacing their oldest processors with 3nm parts?