Artificial Intelligence – the new electricity?

Jennifer Martin
Welcome to the Angle from T. Rowe Price, where curious investors can gain an information edge on today's evolving market themes. Sharper insights on the forces shaping financial markets begin here. Better questions, better insights.
In this season of the angle, we're focused on how recent advances in artificial intelligence, otherwise known as AI, are impacting the global economy and financial markets.

I'm Jennifer Martin, a global equity portfolio specialist here at T. Rowe Price Associates, and I will be your host as we learn more about this intriguing and rapidly evolving area of the global economy and financial markets.
It may be an overstatement to say that AI will change everything, at least for now, but we’ll talk with our experts about how it is already shaping the technology sector and how many of the world's largest companies do business.
I'm really excited to kick off this episode with two of our technology portfolio managers, individuals who I've known a long time, and are in the center of how AI is reshaping industries.

Dom Rizzo, global technology manager at T. Rowe Price Associates, whom I have often described as half man, half semiconductor. And when you hear his enthusiasm, you will understand why. And Jim Stillwagon, communications and technology manager at T. Rowe Price Associates, and a wonderful partner, and the voice of reason as we navigate AI.

Dom and Jim, welcome.

Dom Rizzo
Thanks, Jennifer.

Jim Stillwagon
Excited to be here. Thanks, Jennifer.

Jennifer Martin

 Dom, let's start with you. For a while now, you have been making a bold claim. That generative AI may prove to be on par with harnessing electricity, in terms of its impact on human productivity. Could you explain?

Dom Rizzo

Thanks, Jennifer. Yeah, we've been talking about this for a while now, and I finally feel vindicated because Jamie Dimon wrote in his annual letter that AI has the potential to be on par with electricity. But let's take a step back and think about the technologies that have enabled this step change in true innovation that we've seen because of artificial intelligence.

There's, there’s, really two changes that we've seen that have allowed AI to capture the global zeitgeist the way that it has. The first is on the hardware side, and that's the rise of the GPU, the graphics processing unit from the likes of NVIDIA . The second is on the software side and that's the adoption of these large language models, specifically the transformer architecture of the large language models.

That's the “T” in ChatGPT. And if you've heard me talk before, you've probably heard me use this analogy, but I really love it, so I'm going to use it again. If you think about what AI is doing, you can think about reading the book A Tale of Two Cities by Charles Dickens. And if we test a traditional CPU, a central processor unit, from the likes of Intel, to read the book and identify every time Charles Dickens says the word “the,” it would start on page one and it would read the book.

“It was the best of times. It was the worst of times.” In that first sentence, we obviously have two “the’s.” The GPU, the graphics processing unit from the likes of Nvidia , on the other hand, would rip the book up into 100 pages and pull out all of the “the’s” simultaneously. On the software side, with the large language models, these transformer architectures give all that data context.

So effectively, to extend the analogy, the large language model with the transformer architecture, allows you to say, “Hey, if Charles Dickens uses the word “the” on the first page, it's most likely preceded by ‘it was”: It was the best of times. It was the worst of times. So those are the two innovations that have allowed for this incredible artificial intelligence takeoff that we've seen over the past year.

Now, why can this be bigger than electricity? It really comes to the vast amount of use cases for these two new technologies. I shouldn't really say new technologies because they've been with us for a while. But the applications of these technologies that we've really harnessed in the past year.

So it's not just word summaration that we've seen or even interacting with the computer with the likes of ChatGPT, but we've seen tremendous productivity-enhancing applications in the likes of code writing, in the likes of image generation and the likes of cybersecurity. Go analyze every threat that's ever happened in your enterprise, and pull out all of the different detections that we need to make going forward.

And potentially the biggest application that we're going to end up seeing for this new artificial intelligence technology is re-envisioning the ad tech stack. And I know Jim's going to talk more about that.

So, with all these different applications, you can easily get to the claim that, hey, if we put this generative AI in so many different end markets, we have a productivity-enhancing technology for the global economy that is on par with electricity.

Jennifer Martin

Jim, what are you seeing in terms of AI's impact to date on the major consumer Internet platforms?

Jim Stillwagon

Well the companies funding this AI arms race, primarily the tech mega-caps’, spending billions of dollars on GPUs and data center buildouts—eventually, they need to earn a return to justify that level of investment.

Thankfully, we can already see the early proof points of AI’s transformative impact on the largest consumer internet platforms. Look at how AI is driving engagement on social media, from TikTok to Instagram reels, to YouTube shorts. It's all about the rise of short form video, paired with AI powered recommendation algorithms. If you think about a traditional social feed, your Facebook experience circa 2015. That feed would sort against a pretty narrow slice of content, posts from your friends and family. But then there's this step change in complexity with the shift to high velocity short form video. The algorithm needs to select from billions of pieces of content across the entire global creator economy, not just from your direct connections. That's a massive AI and machine learning exercise. And it turns out that short form video is just much more engaging than a traditional social feed in isolation. And then if we fast forward a few years and imagine what happens when content creators fully embrace generative AI, what that could mean for the quality and the quantity of their output. Remember user generated content. It's the substrate of social media. Look at the rate of progress in areas like text to video generation. How you can type a scene description into open AI’s latest Sora model, and have it create a video from scratch. Three years ago, Sora’s predecessor for AI video creation, it was called Dall-E, produced these uncanny valley type of videos. Technically impressive at the time, but not the type of content you would want in your Instagram feed. But now we have these text to video diffusion models approaching near cinematic fidelity, to the point where the recent Hollywood strikes invoked limitations on the studio's ability to use A.I.

So, we could be on the cusp of an AI enabled content wave, which would make the most performant recommendation engines in social media even more valuable, in terms of sorting and selecting engagement worthy content.

Jennifer Martin

So, a quick follow up question, Jim. How does that engagement uplift actually translate to monetization? What does AI mean for the platforms’ business models?

Jim Stillwagon

Well, there's this old marketing adage advertising follows eyeballs. It still applies to social media. More consumption equals more ad dollars, all else equal. But the P&L impact from AI goes way beyond engagement. We're talking about a complete re-architecting of the ad tech stack from brand creative to campaign execution, to targeting and attribution. Meta and YouTube already have AI tooling that will automatically adapt existing ad campaigns to different surfaces with different aspect ratios.

So, a traditional YouTube ad will reformat to vertical video shorts, or an Instagram Stories ad will be recut for Reels placement, all with minimal input from the marketer. We’re even seeing Spotify experiment with AI for podcast advertising, text to audio ad creation, or even authentic translations for different listener demos.

But you know, and then when it comes to ad targeting and attribution, we all know how data privacy norms are evolving. The old approach of tracking individual users based on a device or browser ID, that's giving way to more probabilistic cohort- based models. Google has what they call Performance Max, Meta has Advantage Plus. These are all just campaign formats that leverage AI, to deliver better ad performance with less-intrusive tracking.

Think about the end game here as abstracting away complexity for small businesses. Allow an unsophisticated marketer to show up with some basic brand creative elements, set their return targets, and hand the keys over to Google and Meta. If you lower the bar to experiment with performance marketing, you will bring more demand to the ad auction. And that's actually one of the main lessons from Meta’s success over the past decade. They grew their active advertiser base from 1 million in the early 2010s, to over 12 million today. Smaller social media peers, like Snapchat and Pinterest, have less than 1 million actives. Broadcast networks have a couple hundred. So, we think AI could be that next major unlock for auction density and in turn ad monetization.

Dom Rizzo

Jim, if I had to take a step back and sum it all up, is it fair to say - Look, AI allows for an abundance of studio-like quality content that we're going to see all over the internet, in image and videos. With that is going to attract more eyeballs, more users time. And then AI really opens the floodgates to allow anyone to become a marketer, whether it's your small pizza shop down the street or your small laundry mat.

Jim Stillwagon

Yeah, it's more engagement, more monetizable consumption, and a more effective ad tech stack, all working together.

Jennifer Martin

Jim, thank you. Very clear on AI’s impact on the consumer internet platforms. Dom, this is a great segue to you as someone who used to be a semiconductor analyst. Could you explain to those of us without advanced degrees in computer science what's so special about AI chips?

Dom Rizzo

Let's take a step back. You know, I don't think I've ever really stopped being a semiconductor analyst. It's a bit like the Hotel California. You could check out any time you want, but you can never leave.

I don't think we have seen companies grow profits at such a rapid pace, at such a scale that we have seen in the last couple of years, like we've seen with the digital semiconductor companies. And why is that? Digital semiconductors are almost the perfect incremental operating margin business model. You only have to do the R&D once, and then everyone kind of knows who you're buying from. The customer is also well known. Who's the customer? It's Microsoft, Amazon, Google. You don't need a ton of marketing. And then what we've seen with an AI is a complete explosion in demand of silicon. The silicon intensity of AI, it’s, it’s, unfathomable. And just to put some numbers around it, AMD, the fablus semiconductor company based in Santa Clara, has said that they expect the AI chip market to go from roughly $45 billion in 2023, to roughly $400 billion in 2027. That's over a 70% CAGR. It's huge.

Jennifer Martin

I think you've been really clear that AI’s probably, this AI cycle, is probably bigger than we expect.

Dom Rizzo

And longer. That's the thing that I think that people might not fully understand yet, because this is really a completely new way of doing computing. So, it's not just that you need to go and upgrade all your current data centers. You have to go and build all new greenfield data centers in order to make this work properly. And we are still in the very early stages of that.

Jennifer Martin

Dom, thank you. And so maybe, Jim, what happens to search business if we're all spending more time directing our questions to AI chat bots instead of typing them into a search box? Could you expand?

Jim Stillwagon

Well, historically, search has been viewed as one of the most deeply moated business models in all of tech. Yet suddenly it became this flashpoint for AI disruption in the wake of Chat GPT’s launch. But to this question of whether AI entrenches or disrupts the search incumbent, we have to think through several risks. What happens if we have an AI enhanced search box on every digital service, from enterprise to consumer? We've all had those days. When you close out of your browser and your desktop workstation audibly labors to shut down 50 chrome tabs. A lot of those tabs were one off searches. Does some of that query volumes slip away to chat bots or copilots over time?

The search results page has been AB tested down to the pixel for over two decades now. From the ad load to the font size, this is one of the most optimized pieces of real estate in all of consumer internet, and it's about to go through the biggest interface overhaul since the desktop to mobile transition.

And then finally, the cost to serve risk. AI powered responses cost orders of magnitude more to generate than traditional search results. Lower revenue per query at a higher cost to serve. That's not a great combination at face value. But if we set aside those business model questions for a minute and just focus on the end user, does the consumer want a search engine, or an answer engine? We all know what the former looks like. It's the canonical text box and ten blue lengths of search. The latter is still in its infancy. More of a personal assistant than a search engine, more conversational and context aware, you know, providing you with actual answers instead of outbound links. In truth, users probably want a mix of both. An AI powered super assistant, plus a best in class search engine.

The incumbent has all of the ingredients to compete here in terms of data distribution, developer talent, infrastructure. I mean, they have huge swaths of proprietary data. Every day google search handles over 8 billion queries. YouTube serves over a billion hours of watch time. We can list off search, YouTube, Android, Chrome, Gmail, the Google Play Store. Those are distribution channels ready to push latest AI offerings into the hands of consumers. And then we have to acknowledge that daily utilities that achieve noun and verb status like Google, they're difficult to unseat. So we want to have a balanced perspective on the future of search, both the risks and the opportunities in this AI era.

Jennifer Martin

Jim, really helpful. And I was struck by your comment. Do we want a search or an answer engine? And I think that really will depend on what you need and probably generation and how we all have kids at home. They're doing things so differently than we did. And so sometimes that will probably drive the adoption.

Dom Rizzo

Well, my kid is nine months year old, Jennifer, so he's just learning how to eat strawberries for the first time.

Jennifer Martin

Well, he's going to be very AI enabled by the time he's walking, that's for sure. You know, I think, pivoting a little bit, kind of keeping the discussion going. But one of the pieces in the news that we saw was really related to the Apple / Google relationship. And there was discussion that Apple is

debating using Google's large language model, Gemini. And so maybe Dom why don't you start with expanding on this business relationship.

Dom Rizzo

Absolutely, Jennifer. If I think about these two companies, they are unequivocally and completely tied at the hip. We learned from the DoJ case that Google pay’s Apple $20 billion a year to be the default search engine on the iPhone. And why is that? There's many benefits and advantages to being the default search engine on a phone that over a billion users carry in their pocket. And what do we know about iPhone users? In general, they have a higher propensity to spend than Android users. It’s also why I’m not surprised at all to see the press reports that Open AI is looking to partner with Apple on the upcoming iPhone.

And what's the vision? Jim and I talk about this all the time. It really is for a super assistant. You can say, Hey, Siri, please book me an Uber at 2 p.m. to BWI Airport. And with that, Siri will go right through the app, book the Uber for you.

Siri Interruption

“It doesn't look like you have an app named BWI Airport and that will go through the app book Uber. You can search for it in the App store.”

Jennifer Martin

I deliberately didn't wear my watch today because she talks a lot. She loves to talk to me.

Dom Rizzo 

How funny. This is exactly why she needs a refresh.

Jim Stillwagon

Definitely. But there are many reasons why these companies would work together beyond the search default partnership. Apple understands that smartphones are a maturing category at this point. They need a technical breakthrough in AI to help accelerate the replacement cycle. From Google's perspective, the chance to partner with Apple on AI would be a strategic coup after several self-inflicted PR missteps, and unflattering comparisons to open AI over the past year. Just the sheer query volume coming off of iOS devices has the potential to reshape the AI arms race.

And remember, if Apple does end up partnering with a foundation model provider, that doesn't preclude Apple from dual tracking their own internal AI model development, particularly for low latency privacy safe on device purposes.

Dom Rizzo

That's right, Jim. The race for the AI smartphone is on. And there are really three areas I think these companies are going to compete, outside of the large language models that sit in the cloud that's going to differentiate the personal assistant on the smartphone. The first is at the silicon level. Which company can design the best silicon in terms of power performance and area for AI applications.

The second is with the developer ecosystem and the tools offered to those developer ecosystems to make those AI applications seamless and easy to use today. One tool that I think a lot about is the accessibility tools. It's amazing. Someone who's vision impaired can operate a smartphone

seamlessly today because of those accessibility tools. That's very similar to what machine learning vision can look like on the phone itself.

And then finally there's the lightweight large language models that will actually sit on the device. If these are efficient and can thread all the different apps together, that's how you get the perfect assistant who can properly answer the question. Please book me an Uber at 2pm to BWI airport.

Jennifer Martin

So, I think, if we move into, kind of our final stages of some of the questions that we had talked about, one of the areas that we wanted to really hit on is, “What are the other areas we're very excited about, when it comes to AI?” And so, maybe Dom, we could start with you.

Dom Rizzo 

Yeah. All of this artificial intelligence is built on top of one thing: data, data, data, data. And it is a cliché to say, but data is really the new oil. And with data you need a place to store it and optimize it and play with it. And that's why we've seen the rise of the lakehouse architecture, from companies such as Databricks. And what's unique about the Lakehouse architecture. It can handle both structured and unstructured data, and that's important because it allows you to run both backward looking analytics, but also, more importantly, forward looking, predictive generative analytics for your enterprise.

Jennifer Martin

And I...and I guess I want you to expand on that really quickly because we're appreciating that you need a lot of energy to run these AI enabled data centers.

Dom Rizzo

It's one of the most compelling advantages that the United States has in the AI arms race. We have incredibly abundant energy resources in this country, relatively low-priced natural gas, some access to nuclear power. We'll see if that ever expands as regulation is able to be updated. And with that, we have the ability to power these data centers, which are effectively the same power consumption of small cities.

Jennifer Martin

Great. Well, Jim, are there areas that you're excited about outside of data and energy?

Jim Stillwagon 

Well, we've been talking about this concept of a super assistant, whether that could be the killer app for consumer AI. Now, to be clear, we're still in the picks and shovels phase of this AI boom. We know the early beneficiaries of the compute layer. It's still early to start picking winners and losers at the application layer.

If you back to the early days of mobile, many of the apps that have earned first-pane status on all of our smartphones—Instagram, Uber, WhatsApp—those companies launched around 2009-2010, several years after the first iPhone in 2007. So we're not there yet with AI applications, but if we had to conceptualize a potential killer app for AI, the super assistant is a great starting point. Basically, delivering on the original premise of Apple's Siri or Amazon Alexa. Setting kitchen timers and playing music is fine, but we want an assistant with real utility.

Let's expand on that example of Siri helping you book an Uber to the airport. Your AI assistant should know that you prefer Uber Comfort over Uber Black. It should have access to your calendar and realize that you need to arrive early to Dulles for a preflight call. It should be providing you with dinner recommendations based on your prior business trips. It should have a prep packet for who you're meeting with on that trip.

That's the real potential for an AI super assistant, but we have many hurdles to get there. We'll need AI models with agency, with memory, with access to personal data. And if an assistant can act on your behalf, that requires bright-line safeguards. You don't want gen AI hallucinations when you're giving a model the ability to click around unsupervised with credit card information.

But there's a reason why the super assistant vision captivates the tech sector. If you have consumer apps and the open Web plugging into a new AI intermediation layer, that's an incredibly valuable and strategic position when we think about the future of personal computing. 

Jennifer Martin

That's what's really exciting, Jim, is that the consumer is really going to start seeing some incredible benefits from the improvements in AI and you highlighted some great examples.

Dom Rizzo

I'm just looking forward to the day that I don't have to read 500 emails every morning and the AI can tell me the 20 most important. 

Jennifer Martin

That will be beautiful. So, AI may be the new electricity. How do you think about technology valuation in the context of this really new technology paradigm or environment?

Dom Rizzo

Absolutely, Jennifer. Look, I think with our conversation today, we led credence to this argument that we've had for a while that AI does has have the potential to be the biggest productivity enhancer for the global economy since electricity. 

And if we look at other productivity enhancing technologies, historically, those have often resulted in speculative stock market bubbles, whether it's the railroads or renewables or China. China was an incredible productivity unlock for the global economy. All of those were coupled with speculative stock market bubbles. The mother of all of them was, of course, the internet bubble where we saw productivity statistics actually improve, step by step, along with the relative performance of the Nasdaq. So where we in that cycle today and how do we navigate it responsibly internally.

Jennifer Martin

Dom, really helpful. You highlighted how important it is to put AI into perspective two previous cycles and how well, thoughtful evaluation of valuation will allow us to navigate this technology paradigm responsibly.

Dom And Jim, I want to thank you very much for today's discussion on artificial intelligence and how it's reshaping the technology industry and the broader economy. Really appreciate your energy and time today. 

Dom Rizzo

Thanks, Jennifer. Jim, this was such a fun conversation. 

Jim Stillwagon

Thanks again, Jennifer. Great discussion. 

Jennifer Martin

If I was to summarize today’s discussion, first and foremost, AI is going change a lot of businesses, in ways that you may not expect. For example, Jim discussed how small businesses may soon be releasing cinematic-quality ads on social media, just by typing in a description of what they want to communicate. 

Second, AI is poised to change how we operate as consumers and in our daily lives. Dom and Jim both described for example their vision of super assistants that will take care of all the details of arranging transportation and getting us to our flight on time.

Finally, the common theme here is AI’s role in enhancing human productivity. But big innovations in the past like this have often led to stock market bubbles—the dot-com era is the recent poster child. But it’s a pattern you see throughout history. That’s one reason why industry experts like Jim and Dom are essential in helping us navigate through what could be, both in the best of times, and the worst of times—to bring us full circle! 

In future episodes, we will broaden our scope and look at how AI will impact the global economy and other sectors like real estate and energy.

Thank you for listening to the angle. We look forward to your company on future episodes. You can find more information about artificial intelligence on our website. Please rate and subscribe wherever you get your podcasts. 

Disclaimers

This podcast episode was recorded in April 2024, with subsequent recordings in May 2024. 

This podcast is for general information and educational purposes only, and outside the United States is intended for investment professional use only.

It does not constitute a distribution, offer, invitation, recommendation, or solicitation to sell or buy any securities in any jurisdiction, or to conduct any particular investment activity. 

This podcast does not provide investment advice or recommendations, nor is it intended to serve as the primary basis for an investment decision. Prospective investors are recommended to seek independent legal, financial, and tax advice before making any investment decision.

The views contained herein are those of the speakers as of the date of the recording and are subject to change without notice. These views may differ from those of other T. Rowe Price companies and/or associates. Information is based upon sources we consider to be reliable; we do not, however, guarantee accuracy. 

Investing in technology stocks entails specific risks, including the potential for wide variations in performance and usually wide price swings, up and down. Technology companies can be affected by,

among other things, intense competition, government regulation, earnings disappointments, dependency on patent protection, and rapid obsolescence of products and services due to technological innovations or changing consumer preferences. 

International investments can be riskier than U.S. investments due to the adverse effects of currency exchange rates, differences in market structure and liquidity, as well as specific country, regional, and economic developments. 

Growth stocks are subject to the volatility inherent in common stock investing, and their share price may fluctuate more than that of income-oriented stocks. Diversification cannot assure a profit or protect against loss in a declining market. There is no guarantee that any forecasts made will come to pass. 

Where “uncanny valley type of videos” was mentioned – this refers to the discomfort humans feel when they encounter robots that demonstrate human-like qualities. Although this term is largely relates to human interactions with robots, the uncanny valley can also occur with digital avatars and CGI used in films. 

Where “P&L” is mentioned, this refers to the profit and loss of a company’s balance sheet. 

Where “Ad Tech stack” is mentioned, this refers to ad tech which helps advertisers and agencies deliver the right content, at the right time, to the right audiences based upon first-party and third-party insights. This ensured advertisements are reaching engaged audiences that may be interested in the product or service. 

Where “auction density” is mentioned, this refers to the process that happens with each internet search to decide which ads appear for that specific search, and in which order those ads will show on the page (or whether or not any ads will show at all). Each time an ad is eligible to appear for a search, it goes through the ad auction. 

Where “CAGR” is mentioned, this refers to the compound annual growth rate (CAGR), which is the mean annual growth rate of an investment over a period longer than one year. The source of this data derives from the Advanced Micro Devices (AMD) Advancing AI Event on December 6, 2023. 

Where “AB testing” is mentioned, this refers to split testing, or bucket testing. It compares the performance of two versions of content to see which one appeals more to visitors and/or viewers. 

Where DoJ is mentioned, this refers to the U.S. Department of Justice. 

The mention of "fablus semiconductor" refers to Fab's, which is where semiconductors are born. Fab is short for “fabrication”, which means to produce, and refers to semiconductor production facilities in the semiconductor industry. 

This podcast is copyright 2024 by T. Rowe Price. 

ID0006876 

_{202405-3502067}