E83:AI的年中回顾：代币瓶颈

Recorded: July 1, 2026 Published:July 9, 2026 Host: Rob Rowe,GlobalHead of Research, Citi Guest: Heath Terry, Head of Technology and CommunicationsResearch, Citi Transcript: Rob Rowe (00:00) Today on our podcast, we're going to be addressing a mid-year review of everything AI andinnovation. Welcome to the Research @ Citi podcast. I'm your host, Rob Rowe, Interim Head ofResearch at Citi. Today, we're really excited and fortunate to have Heath Terry with us, who is our Head ofTechnology and CommunicationsResearch at Citi, and who is absolutely a leader in this field.Heath, thanks for being with us. Heath Terry (00:28) Thanks for having me, Rob. Always glad to be here. Rob Rowe (00:31) Fantastic. Maybe we can kick in and just look at market performance in terms of AI right now,think about the market in general. I know that there's a lot of questionscoming up in regard tohow is this transformation going? I think, surprisingly, this market has been on the upside forsome time now. It's still going. So, you know, we had a debate sometimes about whether this isa cycle or whether this is just a continuous move in terms of development and innovation. Howare you seeing performance in the marketplace right now? And how are you seeing thatdevelop? Is this a cycle,or is this more just a new level that we're going to? Heath Terry (01:13) Look, this is a new technology that is fundamentally different from any major technologytransition that we've seen before, right? We've been through several of these. We've beenthrough cloud,we've been through mobile,we've been through internet. And this one dwarfs allof those in terms of scale, as we've talked about a lot. The impact that it's going to have on theeconomy, on corporates, on the competitive stack within industries,is greater than any we'veseen before. And within technology—and this is what you're seeing reflected in the market—this is thebiggest shift of economics we've ever seen. We are taking 90% incremental margin dollars out ofsoftware and internet and information services and we are moving them into AI models that areright now about 50% incremental margins. And that delta, that 40-percentage-point delta, goesinto the pockets of all of the infrastructure providers, all the data-center infrastructure that hasto be built for this, which is whyyou've seen such a dramatic spread in terms of performancebetween the semi/cap-equipment companies, the hardware companies, the infrastructureproviders themselves, and all of the software, internet, information-services companies that areon the other side of it. So this, I think, goes beyond a cycle because cycles are usually tied to economies or products.This is a fundamental shift in technology, the likes of which we really haven't seen before. Rob Rowe (02:44) AndHeath,to that point:In terms of adoption,there’sbeen a lot of argument about adoptionone way or the other. How are firms adopting this? How is the world adopting this? How areindividuals adopting it? Some people would perceive that adoption to take longer than peopleexpect. But do you think that, infact, it's accelerating? How are businesses looking at adoption?Where is adoption right now? I think that's the other leg that people are also looking for. Heath Terry (03:11) Yeah, absolutely. Adoption has been accelerating. We have been saying this forwell over a yearnow, just in terms of the pace of acceleration:It keeps steepening,the bend in the S-curvekeeps getting steeper. The best proxy for measuring this,to us,is the hyperscaler backlognumbers. We've gone from 28% year-over-year growth last year to 32%, to 53%, to 91%, to 143%in Q1 of this year. And that acceleration that you're seeing reflects the adoption of companies at the enterpriselevel:healthcare companies, financial-services companies, certainly technology companies,industrial companies trying to put this technology to work in a way that is generating ameaningful return on them. And we are seeing that meaningful return show up in higher revenueper employee. We're seeing it in margin expansion across those early adopters. And then, ofcourse, you're seeing it in the returns that the hyperscalers are generating on the investmentsthat they're making into this infrastructure. Rob Rowe(04:17) And are there particular industries where you're seeing that acute development happening morerapidly than others? There are. This is a technology that is very unevenly distributed byindustry and within industry.So,there are companies that are further ahead on this than their peers and their competitors.And there are certainly industries. Soif you look at the top in this—and, you know, this isn't going to surprise anyone,but AI hasbeen best for software development:the most performant, the most effective tool. And so nosurprise, the companies that have the most software developers are the ones that are benefitingthe most from this.Soit starts at the top with technology companies, then you get down tofinan