AI Snapshot: Inside EPIK's Rise to #1 on the App Store

0:00So right now, of course, we have kind of this backdrop of a super strained AI chip supply chain, right? This is something no one can get enough chips. There's huge wait lists. They're very hard to get. And I think amid all of this, OpenAI, which is, of course, the number one AI startup right now, is reportedly contemplating an entry into the AI chip space. So this move comes as the demand for powerful AI processors to train their ever-advancing models. We've got ChatGPT. GPT4 is now the big thing. GPT5 is rumored to be coming out soon. Well, I don't know what the rumors are, but I think it's going to come out in December.

0:36And you can just, you know, I mean, that's my prediction based off of a bunch of stuff I've seen them trademarking some things around the space and some people saying that they're already in the training process. Could be completely wrong, but I think they might just hit it one year later in December GPT-5. But in any case, they're also just they've just launched Dolly 3, which is, you know, their upgraded image generator, which I think is a really good move. In any case, of course, all of these things are sucking a ton of compute. So OpenAI currently depends on GPU based hardware. like a lot of its different, you know, peer companies.

1:08And I think right now they may be looking to change that because GPUs have been essentially the backbone for AI development due to their like proficiency in handling parallel computations, which is really a necessity for training today's kind of top tier AI. But the rise of generative AI, of course, has really, really helped out companies like NVIDIA who are making money off of the chips. So however, I think it has kind of pushed this whole GPU supply chain to its limits. Microsoft recently cautioned about potential service disruptions due to like issues with server hardware because of AI. And also NVIDIA's high performance AI chips are reportedly, you know, sold out until 2024.

1:49So really it's, it's hard to get your hands on those things. And I think a lot of these AI companies are kind of feeling the strain because it's not just open AI that needs its own chips. There's all these other companies that are doing it. It probably feels like they're getting in their own way and it's like kind of a bottleneck to their own company so i'm sure open ai at this point is like you know doesn't want to put their uh their company in the hands of another enterprise so it looks like they're kind of eyeing this space so to give a little bit of context and maybe perspective on kind of the magnitude of demand a study by bernstein analyst stacy raskin highlighted that if ChatGPT queries scale to even a tenth of Google searches volume, an initial investment of roughly $48 billion in GPUs would be necessary.

2:35This is insane. And I'm going to say that again, because I do not think people fully understand this. We talk about like, oh, ChatGPT is awesome, yada, yada, whatever. And Google like should be worried it's going to get beaten. And Google's kind of doing this stuff with BART on the side. But like if Google searches really were replaced by ChatGPT, which to be honest, for a large part, I'm doing a lot of, I'm doing a lot of stuff on ChatGPT that I would have used Google search for in the past. But if that was to replace it, right, 10 % of Google searches would require almost$50 billion in GPUs to be purchased in order for OpenAI to facilitate that.

3:10So that's absolutely insane. If they were to do 100 % of Google searches, we're talking about somewhere like 500 billion dollars this is insane numbers if google was just to switch to completely be chat gpt instead of like a google search we're talking 500 billion dollars in gpus that would be needed so i think another really interesting number is that around 16 billion dollars worth of chips would be required annually for sustained operations right so even if they made like if they so let's say chat gpt is going to take over 10 of google searches 50 billion dollars up front on GPUs and then every year 16 billion dollars this is insane money and I think chat GPT and open AI really know that if they want to be able to scale this is a bottleneck for them right even if they were the most popular thing in the world this is a bottleneck and I think they're looking at you know how can we build things how can we build them better I'm cheaper so that we're not kind of stuck with this bill even if they could get it half price by making it themselves it would make a really big difference so open AI is it is by no means kind of pioneering the custom AI chip territory.

4:16Google's TPU tensor processing unit powers massive generative AI systems like Palm 2 and Imogen. Amazon provides its AWS customers with proprietary chips designed for both training, I think it's called Tranium, and Inference, which is Inferencia. So meanwhile, there's a whole bunch of rumors around Microsoft collaborating with AMD on an AI chip called Athena, which OpenAI is like reportedly testing right now. So I think with a venture capital injection of over$11 billion, that's what OpenAI has at the moment. And, you know, I think they're nearing a$1 billion in yearly revenue. Like that's kind of where they're at right now.

4:55They said, I think Sam Altman earlier this year said they forecast they're going to hit about a billion dollars in revenue this year. I think OpenAI's financing position appears like to be pretty solid. It's fairly robust. But recent, of course, murmurs from the Wall Street Journal suggest that the possibility of a share sale is going to catapult OpenAI's secondary market valuation to around$90 billion. They're kind of looking at doing that right now. So that would be very, very incredible. However, I think the road to AI chip development isn't without its bumps. Just last year, AI chip maker Graphcore saw its valuation plummet by around a billion dollars post a Microsoft deal fallout.

5:34Microsoft never ended up going through the deal and it really kind of hammered their company. So I think that was kind of leading to an announcement of job cuts owing to challenging, you know, they said this is like challenging economic situation stuff. So I think the past month also saw Graphcore grappling with declining revenue and mounting losses. Intel's AI chip subsidiary, which is called Habana Labs, had to let go of an estimated 10 % of its employees. Even tech giant Meta faced a bunch of turbulence with its custom AI chip endeavors, eventually axing some of its experimental hardware that they were kind of working on before.

6:07So it's not like it's all roses in this space and it's not like it's very easy, right? Of course, we have NVIDIA, which is absolutely crushing it, but there's a lot of people that are kind of hurting. Maybe that's because an NVIDIA was significantly better. There's a lot of people speculating on why that is. But I think while OpenAI's potential move into the AI chip domain is a significant development, the journey to bring a custom chip to market is gonna be really lengthy and might drain hundreds of millions annually. The critical question I think that remains is whether OpenAI's stakeholders, which of course is Microsoft that owns half the company, are ready to kind of place their bets on such an ambitious and uncertain venture, especially when Microsoft's already, you know, inking deals with AMD and others.

6:45I think only time is going to reveal if OpenAI's potential chip endeavors are actually going to be successful or not. And the one thing I would just stress finally is that these deals really do take a long time. If they were to design their own chip, if they were to try to launch this, building out the facilities to fabricate it building out you know the essentially all the manufacturing like this stuff takes years we'd be talking two three like two years would be fast right we're talking probably four years where they could scale this thing out and so really at the moment i think it's this is not a solution that really pulls them out of this problem um but you know maybe it's future revenue maybe it helps them in the future but at the moment i think everyone really is kind of stuck with nvidia and a couple of the other players and uh essentially just trying to make do with what they have currently on the market.