00:00When you look at the chips that are out there right now what is the lifespan. Well that's that's been
00:05a really interesting thing because it used to be we'd say like this kind of data center compute would be
00:10have a lifespan of two to four years. And people expected like the H100 build out to really depreciate. But
00:18they're still commanding pretty good prices. So the longevity of the build and like how it's monetized is gone further.
00:27But that's also partly because of growth.
00:29And partly because how the market's evolving. The thing I made a point about is if you look at what
00:36we talked about for AI two years ago people hardly talk about it.
00:41Nobody talks about Lama 70 B and nobody talks about a whole bunch of models. The models got really big
00:46really fast really effective. And the big the big issue is change and what's going to happen next.
00:53Well give me a sense of what you think will happen next particularly not just in terms of the type
00:57of chips but the idea of what exactly the use cases will be and the type of technology that will
01:03be needed specifically for that.
01:06Yeah it's really interesting. So one of the things that AI is great at is programming. And there's two big
01:11reasons for that. One is it was written by programmers and developed to help programming.
01:17So that was a really early access. It's also really good at search and summarizing. And again you can think
01:23of that as it was trained on the internet.
01:25And it's become really good at that. But as we start to refine the use cases I think we're going
01:32to use it for more more science and engineering more more use cases and analysis.
01:38I think everybody already knows. I'm sure you guys do this. If you want to go look up what's going
01:43on with a certain market segment you do an AI search first.
01:46Right. Because it'll give you a good overall sense. Yeah no it's a fair point here. I do want to
01:51you know sort of game plan out this scenario where you call it these stranded assets dead silicon dead data
01:58centers that you know the world we could be skating into.
02:01In your view. I mean what what survives in that sort of system where you do see basically technology advancing
02:10and some of the the the chips that we're working with right now do become obsolete for lack of a
02:16better word.
02:17Yeah. Yeah. I'll give you two different takes on this. One is it's been very stable for a long time
02:2340 years that the computer world's divided into you know compute you know general purpose computers.
02:30We even use the word general purpose. Memory and IO like you know IO is networking you know flash you
02:36know disk drives compute the CPUs and then memory is all kinds of memory SRAM DRAM flash.
02:43Right. And then when there's changes in the market you see a proliferation of new companies and new ideas.
02:49The internet boom spawned a whole bunch of network startups. The GPU and gaming boom spawned I think 35 different
02:56GPU companies.
02:57And for a while HPC generated floating point accelerators Cray you know there's all kinds of computers there. So in
03:05the growth phase there's experimentation but then there's always consolidation and I think we'll see the word the phrase general
03:14purpose AI will come up.
03:15Because today everybody's talking about disaggregated AI and specialty hardware. That's a sign of an inflection but that's not usually
03:24the long term landing place.
03:25So from experimentation of course to consolidation I'd love to hear you know how you see tense torrent sort of
03:33fitting into this picture.
03:35All right so I think we're one of the few doing real general purpose computing. So it's a balance of
03:41memory you know DRAM SRAM flash drives compute and then the networking to hook a lot of processors together.
03:50We see the software applications is pretty distinct. So if you look at you know the big LMS versus small
03:57video image speech to text. They're fairly different applications but underneath there we think there's a pretty common hardware method.
04:06So we're building software to differentiate but the hardware platform is actually one general purpose AI computer.
04:13Jim I do have to ask you just about what your plans are for this company. There's been a lot
04:18of talk about consolidation in this industry. Just a few minutes ago we had breaking news that on
04:23semi was buying synaptics. There was a report in the information yesterday I believe that seemed to intimate that there
04:29had been some sort of discussions between your company and Qualcomm.
04:33Sure. Our mission right now is we're going to grow like we are building computer. We're in production. We're driving
04:41volume. We're really interested in the high value
04:45end of the AI token segment. So that's large LMS prefilled decode large video models really fast acceleration. We see
04:54a big opportunity there. We think that's where the most value for our company is.
04:57We also think that's that's the center of the general purpose compute and general purpose AI that's coming. So like
05:05we're interested in partnerships.
05:08Yeah. And we'll see what happens. But with regards to just the standard that you're using obviously it's a more
05:12open standard relative to what we know
05:14NVIDIA is making. Are you trying to go directly up against NVIDIA or are you trying to sort of be
05:19complimentary to it. The market's really big.
05:22And and people asked you know for a couple of different things. Our AI software stack is open sourced. Our
05:29both AI processor and support processor is built on the risk five architecture which is open.
05:34We actually use an AMD processor is our host node in our galaxy server. So we're pretty flexible about the
05:41compute architecture. But the openness you know the best models in the world today.
05:46Many of them are already open source. And having the software stack itself be open source is a compliment to
05:52that. And people want to own their compute.
05:54And they want to pass this you know sovereign AI. And so we that's the business that we're in.
05:59Amen.
Comments