April 16, 2024
George Lee, Co-Head of the Goldman Sachs Global Institute, joined Tony Pasquariello, Global Head of Hedge Fund Coverage, on the Breaks of the Game podcast to discuss the AI revolution — where we are in the cycle, what’s the sentiment among consumers and enterprises, and what are some of the concerns and bottlenecks.
This conversation was originally recorded on March 15, 2024. Below are excerpts from the discussion. The full podcast episode is available to GS Institutional Clients on Marquee here.
You’re back from a four-week tour of planet Earth. You met with a broad set of investors and using that as a sample set, how would you characterize sentiment right now?
There’s an extraordinary level of unanimity about the importance and the potential of the AI revolution. It’s hard to find an antagonist to that core theme.
The degree to which this technology just dropped on the face of the Earth and seemed immediately consequential to people is quite unusual. And that degree of unanimity is building up, rather than dissipating.
One thing I would observe is that there is as stark a differential as I’ve seen in my 30 years around technology between the trajectory of the technology itself and the rate and capability of adoption by enterprises and individuals. The technology’s moving at wire speed of improvement — it feels like every week there are new exploits, new models released, new research findings. And yet, people are still early in the phase of deploying and instrumenting this technology.
That divergence does set up for potential troughs of disillusionment here. I think, given the nature of the speed of the technology, those troughs will be of shorter periodicity than what we’ve seen in the past.
You made a comment to me that in your travels, there was a conventional wisdom around AI — all people of all ages are buying into this.
Yes, it’s unusual. You meet young founders, they’re super indexed to this technology. They’re building it. And for these sorts of revolutions, you often find a generational lag around this. But here, I met people even older than me, that are running legacy businesses, and they were as excited, as focused on the impact of this technology, as their younger peers. And that feels very different to me than prior revolutions.
You’ve nodded to this previously: We’re witnessing a massive capital spend in the absence of real application-level ROI. Does this remind you of that big telecom bonanza in the late ’90s? And should we be concerned that we’re in late innings on this?
I do think this is an unusual investment super cycle at every layer of the stack. In prior revolutions, investments have been very focused on one layer or another, or they’ve been chained together over time. Here, there is an enormous amount of simultaneous investment in everything from the fundamental power layer, to data centers, to the gear that’s inside data centers, to the foundation models themselves, to the applications.
I wouldn’t say there’s investment in the absence of any line of sight to the ROI. I think everyone recognizes that there is a significant ROI to be had. What I find super interesting about the enterprise decision making that I’m detecting in my discussions, is people are feeling compelled to invest, even without perfect clarity on ROI, because this is so competitively consequential.
If you’re not keeping up with your competitors’ investment and ability to leverage this technology, you’re falling behind. And the potential for this technology to disrupt is causing people to invest with a little bit of a leap of faith to those ROIs.
So does that mean the biggest names in a given industry are the main beneficiaries here?
George Lee: Not necessarily. In some ways I think it’s a fascinating opportunity for more agile competitors, who have slightly more mobile and adaptable enterprises that can adopt things quicker or take a little bit more risk on this. In my travels, I met three or four leaders within a certain industry segment, and in one example that stands out in my mind, the number three player by far was the most aggressive, had the most ambitious plans, was moving quickly.
If we bring this back to markets, the impact of this theme on broader equities in general has been immense. What are you paying attention to here?
I am fascinated by the second- and third-order impacts of this technology in the ecosystem. Behind every great big tech player lies a phalanx of suppliers and enablers. And we are now seeing investors explore interesting ideas below the waves.
When it comes to the AI commercial cycle, which phase are we in right now?
I think we are in the “experiment and instrument” phase, which is between the “build” and “deploy” phases. 2023 was the year of people and enterprises discovering the potential and the power of this technology. 2024 seems to be the year where we are figuring out how to actually deploy. We’re only getting to the point where there are enterprise-ready architectures to really take advantage of this, so I think 2025 is going to be the year of “deploy and scale.”
There has been an immense demand for power from things like EVs, the strong economy, the cloud. But this has of course been supercharged by the AI theme. So how do you think about this power and energy issue?
The unique nature of these workloads, which are very power consumptive and extremely heat generating, demand infrastructure that many traditional data centers cannot support. You are seeing a transition for data center providers to more location-independent and cost-insensitive solutions.
At the same time, we are increasingly seeing really smart engineers focus on efficiency improvements, and you are starting to see that in some of the recent models. But there are some opposing forces at work here — we can mitigate the slope demand by being much more efficient about the way things work. With the marginal cost of “intelligence” trending to zero eventually, you would think that there’s just massive cost savings that ensue. But it actually creates this second wave of demand and increase in volume, which sort of offsets to some degree the efficiency gain of the technology. A very complex calculus to try and figure out.