There’s an argument for fewer (~3) competitors in top end chip design.
This isn’t the 90s. The sheer scale of R&D needed now means it basically doesn’t work unless they can sell to like a quarter of the planet. Re-inventing the wheel for a smaller slice of the pie would cost buyers.
On the other end, having fewer hardware targets makes life infinitely easier for middleware devs, and things more optimized, features more common.
And yeah, I know, every monopoly argues this. But chip design (and fabbing) really is different. We are talking many hundreds of millions just to tape out a single die in a whole lineup.
…To be clear, Nvidia’s current market share is not okay, not even close.
But they have more competition at the top end than you think. Cerebras is real, Google TPUs are more mature than most realize, and most of all, the Chinese ASICs are going to take over that part of the world.
There’s an argument for fewer (~3) competitors in top end chip design.
This isn’t the 90s. The sheer scale of R&D needed now means it basically doesn’t work unless they can sell to like a quarter of the planet. Re-inventing the wheel for a smaller slice of the pie would cost buyers.
On the other end, having fewer hardware targets makes life infinitely easier for middleware devs, and things more optimized, features more common.
And yeah, I know, every monopoly argues this. But chip design (and fabbing) really is different. We are talking many hundreds of millions just to tape out a single die in a whole lineup.
…To be clear, Nvidia’s current market share is not okay, not even close.
But they have more competition at the top end than you think. Cerebras is real, Google TPUs are more mature than most realize, and most of all, the Chinese ASICs are going to take over that part of the world.