Hacker News Reader

Comment by gmerc | Hacker News Reader

If NVIDIA could be easily replicated, Intel would have done it.

Given enough time they will and won't be able to get 70%+ margins for their hardware.

There are multiple initiatives to eat their lunch which is based on Cuda.Their advantage isn't even in hardware, but rather how they've invested years and years in having integrated software + hardware solution for AI.

“Given enough time” assumes competition will catch up while leader stays idle.

If leader is still innovating at a pace greater than any competition, then competition will never catch up.

This is assuming everything continues in the same direction the dominating company is innovating in though right?

Maybe AI is here forever, maybe there are different approaches that will blindside them?

ATM as far as as datacenter GPUs good the price/performance is the only metric that counts (power costs are almost insignificant relative to how much a H100/etc. costs).

So Intel only needs to come up with something relative cheap that's good enough (it can easily use > 4 times more power to perform the same task) to force Nvidia to cut their insane margins.

Your idea of "easy" is stupid. Possible yes easy , no. And all that money they spend to catchup just to find Nvidia used that time to RD the next step. At this level it's a patent game. Also at this level the companies literally employ the best in their field. The best out of 8 billion. Replicating talent is impossible.

There’s many hot takes like this, most of which ignore the bottlenecks in the supply chain.

Reply on news.ycombinator.com