00:00Physical AI, the humanoids, the most advanced in the world,
00:04whether it's Tesla, Figure, Techman,
00:08all the chips here built in the Taiwan ecosystem.
00:13And then, of course, cloud AI,
00:17whether it's the TPU racks, the racks by NVIDIA,
00:23Graviton, everything here, as I mentioned,
00:27100% of our ecosystem is built in Taiwan.
00:30So, without Taiwan, there really is no ARM.
00:35Thank you again.
00:42Now, what seems like a long time ago,
00:45and in the world that we're living in with AI,
00:48we're living in light-year speed,
00:50we did an event called ARM Everywhere back on March 24th.
00:54And at that time, we were looking at what was going on
00:59relative to the growth of agents and agentic AI.
01:04And at that time, and this is March 24th, not so long ago,
01:09showed a slide about the growth of OpenClaw
01:12relative to Linux and Kubernetes.
01:15GitHub stars on the left are exactly what you think they are.
01:19They are stars that rate the popularity
01:22or the stickiness of a certain application.
01:26OpenClaw reached levels almost beyond parabolic
01:30in terms of the takeoff.
01:32And this is back in March 24th.
01:34And what that told us was that
01:37the growth of these agentic platforms
01:40were driving demand for CPUs
01:43in a way we had not seen before.
01:46And the logic behind that is quite simple.
01:49GPUs, XPUs are amazing at generating tokens.
01:52That is their purpose.
01:55Whether it's training to generate the learning
01:57or inference to deliver the tokens,
02:00the token machine, the token factory,
02:01is the accelerator.
02:03But agents, unlike humans,
02:06don't sleep.
02:07And agents beget agents that beget agents.
02:10And all of those tokens
02:11that need to be distributed,
02:14managed, orchestrated,
02:16delivered to the destination,
02:17that's only a workload that CPUs can do.
02:21CPUs, of course, in conjunction
02:22with a full system design.
02:24So we made a comment back on March 24th,
02:27and I think we were probably
02:29one of the very first to do this.
02:31It said, we believe going forward
02:33that four times the number of CPU cores needed
02:37in the same power envelope going forward.
02:41Now, that multiplier,
02:43I ended up getting so many questions
02:45relative to show me the math
02:46and how do you figure that out,
02:47and not long after that,
02:49you started hearing numbers of 4x, 8x, 10x.
02:53It's a hard number to predict
02:54just based upon the growth rate of these agents.
02:57But what we do know is as follows.
03:01If we look at today
03:03what we're seeing in terms of agentic growth,
03:06even fast-forwarding from the 24th of March,
03:12this is just exploding.
Comments