In the video clip below from a recent episode of The AI Investor Podcast from 24/7 Wall St., the team discusses the biggest trends to watch in the coming year.
As we approach 2025, anticipation is building around the significant advancements in artificial intelligence. One of the most pivotal developments is the launch of Blackwell, NVIDIA’s next-generation GPU architecture. Blackwell is set to be a game-changer, offering a substantial leap in computational power that will enable more sophisticated AI models and applications. Early benchmarks indicate a performance increase of up to 3.7 times compared to its predecessor, the H100 GPU, with only a modest increase in cost.
This exponential growth in processing capability is poised to break current limitations in data centers, particularly in scaling GPU clusters beyond 100,000 units. The enhanced performance of Blackwell will not only impact NVIDIA but will have cascading effects across the entire AI industry, catalyzing advancements in networking, switches, and optics that are essential for large-scale AI operations.
Another major trend to watch in 2025 is the emergence and mainstream adoption of AI agents. These are autonomous AI systems capable of making decisions and performing tasks without human intervention. While the concept has been primarily theoretical until now, developments suggest that AI agents will soon play a significant role in various industries.
For instance, companies like Apple and Amazon could leverage AI agents to revolutionize customer service, offering personalized and efficient interactions that surpass current chatbot capabilities.
This shift could redefine consumer hardware and software, necessitating a complete overhaul of how these systems are designed and function. The transformation is comparable to the early 2000s when companies began integrating the internet into every aspect of their operations.
As AI agents demonstrate advanced reasoning and autonomy, they are expected to trigger a “ChatGPT moment,” capturing public imagination and driving widespread adoption. This evolution will not only enhance user experiences but also open new avenues for businesses to engage with customers, streamline operations, and innovate their service offerings.
Watch the Video!
Subscribe to The AI Investor Podcast on Apple or Spotify
Transcript
[00:00:00] David Hanson: Let’s actually jump into the trends that we’re looking at, the biggest trends going into hard to say that 2024 is nearing the end. We’ve only got a couple more months. So as we go into this next year, you said the reacceleration. Maybe it’s plateaued a little bit in terms of excitement. I think many would say that 2024 has been the year of AI.
[00:00:18] David Hanson: You’re saying next year could be even bigger. So let’s talk through some of these trends, which ones are new, which ones are continuations or just something that’s already in the marketplace now getting bigger. So where do you want to start on the trend side of things?
[00:00:30] Eric Bleeker: Let’s start with the launch of Blackwell.
[00:00:32] Eric Bleeker: Cause I think this is the fundamental jumping off point for a lot of the advancements in AI. I think AI has a little bit of a, you’d almost call it like a tick tock model where the amount of compute, the kind of processing performance makes a jump, which then allows a jump in the software the AI models, which feeds the next compute cycle.
[00:00:56] Eric Bleeker: So we’re at a point right now where we’re actually hitting a wall on the compute side where the biggest data clusters, they can only get to about. 100, 000 GPUs. And the reason for that, which again, this will dovetail with other concepts we’re going to talk about is that it’s not just the processors.
[00:01:17] Eric Bleeker: It’s about the entire stack of technology in a data center, about how you network them all together, the switches, the optics, and if everything’s not getting better. You’re not going to be able to make these data centers meaningfully more powerful. Blackwell is meaningfully more powerful. You talked about, you can benchmark things all kinds of ways.
[00:01:39] Eric Bleeker: We talked about six X last week. I was looking at the first real benchmarks I’ve seen came out for it on August 28th. And if you’re just looking. At a tokens per second running this is meta’s model. The B 200 was at 11, 264. The prior most advanced chip, the H 100 was at 3, 065. So that’s a performance jump of about 3.
[00:02:03] Eric Bleeker: 7 fold for pricing. That is not 3. 7 fold. It might be more like 50%. So again, this is going to give a massive performance jump and it’s also going to make. All of the other technologies. That are necessary for making these huge GPU clusters. It’s going to move them all forward meaningfully, which is another trend.
[00:02:27] Eric Bleeker: We’re going to talk about in a moment, but again, this is going to unleash that next generation of models. So when something comes out last week, we talked about the chat GPT moment and GPT for being this massive leap forward and how, you’re able to almost see AI reason the next generation of models are going to be another Kind of leap forward that begins a new hype cycle, a new, and Blackwell coming out is going to facilitate this kind of up leveling in capabilities so that impacts NVIDIA and that’s what almost all media is going to focus on, but it actually has cascading effects across the entire AI space.
[00:03:08] David Hanson: So you talk about the GPT moment where we see AI logic and, put things together, not to put you on the spot, and it might be that we don’t know all the capabilities of Blackwell, what are the kind of big unknowns or things that Blackwell might be able to unlock that we haven’t been able to achieve from just a functionality perspective?
[00:03:26] David Hanson: Is it the ability to have better memory of what’s going on? Or is it just, we don’t know yet and we won’t know until we start building the models with the new.
[00:03:34] Eric Bleeker: Let’s use that as an opportunity to go to a second kind of prediction for the biggest jumps. I think we have to talk about AI agents and so far AI agents has been pretty, almost theoretical.
[00:03:47] Eric Bleeker: Basically an AI agent is an AI model and algorithm that can autonomously make decisions. You’re seeing a lot in things like coding, but it’s not going very mainstream. So we had Brett Taylor, or he was on a podcast named invest like the best recently, and he’s building out an AI agent company and where they’re starting as customer service.
[00:04:15] Eric Bleeker: And he had to quote that basically the interface and how you interact with every single company will increasingly be driven by agents that are, they’re going to autonomously be able to make decisions. You could actually be talking on the other end to an AI, agent that it’s not just trying to query from a select number of questions, it’s actually able to make decisions for a company itself.
[00:04:40] Eric Bleeker: And it reminded me a lot. When we worked on Motley Fool, David Gardner, who’s one of the founders of the company, he had this quote that You know, in the early two thousands, he said, eventually every company will be an internet company. And at the time the internet was so nascent, but he had seen around the bend and he had seen that not only will every single company have a website, but how you do your CRM will be based on the internet now with so many remote companies, right?
[00:05:08] Eric Bleeker: The internet is the basis for how companies collaborate. So that was a true, truly prescient and true quote. And I think we might be looking at something similar for agents. You could see how you interact with the company, go to its website. It’s just an agent. What do I need to do today? You could begin interacting with Amazon around how you want to be shopping for anything.
[00:05:29] Eric Bleeker: Apple intelligence. You can see how almost all consumer hardware is going to be remade. Cause you think about something like, Siri, why is Siri stink? Like everyone hates Siri, right? It’s not just because it can’t understand what you’re saying. It’s also because the missing thing with voice has always been, you need to be the UI and you need to know how to use it.
[00:05:51] Eric Bleeker: But if you’re an agent that knows how to be predictive and helpful for you as tying into, all the other aspects of your life, all of a sudden that assistant can be truly breakthrough. So I think agents will mean that. Every single type of consumer, consumer hardware and software is going to need to be fully essentially rewritten and redefined in how it functions maybe within the next five years.
[00:06:16] Eric Bleeker: And I think the first true examples we’re going to start seeing about AI being able to make these autonomous decisions happens in 2025. So I think that’s going to be a chat GPT moment where it’s like, Oh, it’s not just AI and me going to chat GPT. And saying something, here’s an example of where AI is being able to take actions.
[00:06:37] Eric Bleeker: It’s basically AI for the first time to a wide audience, showing reasoning and being able to act autonomously. So I think that’s going to be a huge fundamental thing in the next year.
[00:06:49] David Hanson: Yeah. Cause even with a chat GPT, that was a type of consumer facing application, but people still had to come to it, right?
[00:06:56] David Hanson: It was, you had to seek it out versus it just naturally being. In your everyday life. When I think you’re, I know the customer service example is a common one because it’s something that almost every company has to deal with. I think it was what Klarna came out last year and made some sort of statements that, Hey, we figured out customer service and we’re going to be able to automate things.
[00:07:15] David Hanson: I think it’s still TBD on how exactly widespread that is. But when you talk about, I’m sure what everyone has an experience with customer service, let’s take, you’re calling the airline because your flight is delayed. You get the little automated thing. You try to solve it. They’re not answering your question.
[00:07:30] David Hanson: You’re eventually just ah, just let me like, please let me talk to an agent. Please. Just like a human agent, like just 0, operator. And then you talk to someone and they solve your question. What you’re saying is. With something like Blackwell with better models, the logic can get so good that we might actually be able to get our situation resolved without having to hit zero, zero, zero, zero, and talk to a real agent
[00:07:52] Eric Bleeker: in
[00:07:52] David Hanson: the
[00:07:53] Eric Bleeker: future.
[00:07:53] Eric Bleeker: A hundred percent. And I think it’s a hundred fold jump in capabilities. And it’s not just bringing it’s because it’s able to show reasoning where it’s not just a predefined list. It’s able to be so much more helpful laterally across all questions, but also that we can get to a point where the AI makes a decision without having to ever get a human involved.
[00:08:17] Eric Bleeker: If you want to return an item, you could entirely do it. And there’s also beyond customer service. Like I said, there’s. It could just be going to a website or interacting with a brand and being the face of that brand, meaningfully help you if you’re going to Nike and describing exactly what kind of item you want.
[00:08:36] Eric Bleeker: It might be that kind of chat box that instead of you ignoring it, you choose to begin your path with that company there because it is genuinely more helpful.
[00:08:47] David Hanson: Yeah. And like I said, I think so many of us have had These experiences with quote chat bots. Yep. And what nine times out of 10, it stinks. And you’re just like, Oh my gosh, can I just, cause it is following just that predetermined path of, okay. If they say this, then this so I agree. I think that’s, that certainly seems like it does have the opportunity to be that moment because suddenly, if it becomes very widespread, you’ll have people interacting with the AI and be like, Oh wow, this truly is different.
[00:09:14] David Hanson: So certainly one, one to watch. I think we’re all. Rooting for never having to talk to a donation when your fly is delayed again. So I root for that one.
Thank you for reading! Have some feedback for us?
Contact the 24/7 Wall St. editorial team.
Latest from 24/7
