00:00OpenAI just announced its first custom artificial intelligence chip developed in partnership with Broadcom,
00:06which shows up to 50% more processing cost savings in stores now.
00:12Get yours while supplies last.
00:15Sure, I cannot afford one.
00:17Let's discuss with Bloomberg Tech host Ed Ludlow.
00:19Ed, a variety of things to discuss.
00:22Maybe we start with just that breaking news there.
00:25It feels like everyone is trying to look to lower their costs.
00:27Companies are pushing back.
00:28Is this just Broadcom trying to solve that riddle?
00:32All of the above, but from OpenAI's perspective, the point of having Jalapeno, that's what it's called, is to A,
00:40diversify.
00:41OpenAI, just like everyone else, is still very dependent on NVIDIA compute, but everyone is also in an environment where
00:48they are compute constrained.
00:50They need more, and so seeking alternative sources, including custom silicon or ASICs, is a good route.
00:57The pitch of working with Broadcom on ASICs is that on the measures that matter, a dollar per kilowatt basis
01:06or a dollar per token basis, it is a better performing system than what is out there from the market,
01:12i.e. NVIDIA right now.
01:13But that's all on paper, so the plan is for OpenAI to deploy these chips in time in existing Microsoft
01:20data centers.
01:21But as you guys know, the economics require scale.
01:24So this is something they've moved very fast on, faster probably than anyone has taken an accelerator from design through
01:31to deployment on.
01:32But, you know, it's just a diversifying strategy rather than a put your eggs in one basket strategy.
01:37The key, I guess, is that the cost of compute continues to move lower.
Comments