Q: Before we get into the themes, can you tell us a little about your background and how your experience across research, credit and investing has led you to this role?

Rob Bittencourt: I've been at Apollo for almost 20 years. What has prepared me well for leading the thematic investing group is a couple of things. One is that I've covered a lot of different sectors, many of which have been profoundly changed by both technology and other non-linear shifts. I had a front-row seat witnessing how these shifts could fundamentally reorder the economics of industries and the companies that operate within them. I also spent time managing a variety of research teams, which gave me a helpful perspective across industries and all the different cross-currents shaping them from the outside.

Q: What do we mean by thematic investing, and how has it added to Apollo's existing investment approach?

Rob Bittencourt: Thematic investing is a collection of secular forces generally concentrated around technological change and geopolitical change—exogenous forces that can reshape industries very quickly in non-linear ways. Apollo has always had a thematic view across the firm, whether in hybrid, private equity, or credit. But what created an urgency to build a dedicated group is that the pace of change seems to be accelerating. On the technology front, the level of change being driven today is really unlike anything we've seen, certainly in my lifetime. And in geopolitics, the world is more multipolar and more unstable than it's been since the end of the Cold War—more large-scale conflicts, more volatility. Having a team that can dedicate real research focus to these trends in real time, and then collaborate across the Apollo platform from an origination, underwriting, and risk management perspective, felt like a genuine force multiplier for the broader firm.

Q: You've described the development of AI as a series of distinct acts. Can you walk us through the stages and tell us where you think we are right now in the cycle?

Rob Bittencourt: The big bang, for large language models specifically, was the public launch of ChatGPT at the end of 2022. There's a saying that truly advanced technology is indistinguishable from magic—and the only other time I felt that way was when I first saw the iPhone in person at an AT&T store on Sixth Avenue. It was just such a leap relative to what came before that you almost didn't believe it.

The first real signal of the financial impact came from NVIDIA's Q2 2023 earnings, where revenue grew roughly 100% year over year. That opened everyone's eyes: this is more than a curiosity, it's going to drive a real investment cycle. That first phase extended through roughly Q3 or Q4 of 2024, when reasoning models arrived—the idea that pointing more compute at a query produces better outcomes. That created a new vector of demand, because reasoning models are more compute-intensive not just to train but at inference. Then came the DeepSeek scare in early 2025, where everyone worried models would be commoditized by open source. That concern was quickly drowned out by the sheer volume of demand from reasoning models and broader adoption. And then agentic AI really exploded earlier this year, propelling the next phase of the cycle.

I give you that background because it's always difficult to predict the future. But when you try to forecast even something like end-state demand for AI compute infrastructure in three or four years, the type of technological innovations I just described have consistently supercharged demand and made the technology more capable, which promotes more adoption—a genuinely virtuous cycle.

Q: Right now, the discussion really centers on implementation, the cost of tokens, and this AI sticker shock. How real is that concern, and what does it tell us about where adoption stands?

Rob Bittencourt: The industry is at a very interesting point in its lifecycle from an enterprise perspective—really starting to transition from experimentation to deployment. As the numbers get larger, they're eliciting more scrutiny. Companies are now in a position where they want to start seeing a tangible return on these larger and larger investments, which is exactly what you'd expect. The technology was new, companies saw the promise, wanted to onboard it and start validating it. But there's also a budget. As costs become more material, the industry is really going to have to demonstrate real ROI—productivity improvement, top-line expansion in some cases, leaner teams in certain departments.

There's a lot of work required to transition from experimentation to full deployment, especially for more complicated, less well-defined workflows that are very unique to individual companies. I think there will be tension between recognizing the technology's true promise and the challenges in fully unlocking it, balanced against a cost line item that keeps growing. Companies will use a variety of strategies—lower-cost models appropriate to specific tasks, open source alternatives, token budgets per employee, more targeted access for the employees who will be most productive. That's going to be a potential headwind, but ultimately I think the industry continues to grow.

Q: Estimates suggest roughly $5 trillion of AI-related digital infrastructure could be deployed globally by 2030. Two questions: how is all of this going to be financed, and what must happen for that investment to generate acceptable returns?

Rob Bittencourt: The primary patrons of the technology and the associated infrastructure buildout are largely the top five hyperscalers in the US—Meta, Microsoft, Amazon, Oracle, and Google. Consensus today is roughly $725 to $750 billion of CapEx spending in calendar year 2026, expected to go above a trillion. When the numbers were smaller, most of that investment was financed through free cash flow—these are very high-quality businesses generating over $800 billion in free cash flow. As the numbers get even larger, they're starting to look at alternatives. The most straightforward has been investment-grade publicly traded bonds: over $100 billion of net issuance last year, over $200 billion this year. The hyperscalers are becoming a larger and larger part of the IG index.

For Apollo, this is very relevant. A huge area of our focus is private IG—working with blue-chip, IG-rated companies that have specific capital needs tied to capital investment strategies. We can directly or indirectly finance the infrastructure buildout. The $35 billion deal with Broadcom, with Anthropic as a tenant, is a good example. We've also been involved in infrastructure financings with SpaceX through xAI, which I would argue is either already or on the verge of becoming the sixth hyperscaler in the US. It's going to be an all-of-the-above strategy: IG bond markets, private credit solutions, asset-backed financings, some equity, plus the organic free cash flow of the businesses themselves.

On the second question—what you need to believe—we've done some math on this. If roughly $5 to $6 trillion will be spent through 2030, you need to believe the industry will generate roughly $2 trillion a year in AI-related services revenue. To be clear, that's money end users are spending on AI services, not infrastructure. Just for context: the global online advertising market is $750 to $800 billion; the enterprise software market is over a trillion. So essentially what's being underwritten is the creation, over the next four-plus years, of a business collectively larger than the global enterprise software and online advertising markets combined. Being generous, we might be at a $250 billion annual run rate today. There's a lot of wood to chop to get to $2 trillion. And the probability that the final number lands above $5 trillion is higher than the probability it comes in below, which means the required revenue threshold only grows.

Q: You follow a group of companies you call AI canaries—typically software or business services companies where AI disruption would show up first. What is your view right now based on the metrics you track?

Rob Bittencourt: The market has been very forward-looking in pricing AI disruption risk—because absent a handful of examples, there are very few instances where I can point to a company or sector and show tangible signs of disruption in actual quarterly results. Yet you've seen a re-rating in parts of the software sector, business services, call center operators—categories viewed as potentially vulnerable to AI disintermediation. Multiples have contracted significantly. At some point, if that contraction was warranted, you'll have to see it in the numbers. And you're not seeing it even in consensus forecasts. We track about 200 companies across business services and software, and roughly 10 of those have consensus EBITDA declines over the next couple of years connected to AI disruption.

Why do we focus on these names? A couple of reasons. First, it goes back to what you need to believe to justify the current pace of investment. I don't want to suggest AI is purely zero-sum—that the only way the industry generates real revenue is by taking business from incumbents. But there will be some of that, particularly within this group. The longer you don't see any tangible impact, the more concerned we'd get that the technology is struggling to diffuse through the economy at the pace currently expected. So these companies give us micro-level insights with macro implications for the state of AI diffusion. Second, from a risk management perspective, understanding how AI is changing the unit economics and business models of these companies gives us data points and precedents we can use to think about how AI will impact a broader set of industries.

Q: AI-related issuance is now spreading across nearly every corner of capital markets—investment-grade hyperscaler debt, high yield, leveraged loans, IPOs, secondary offerings. How should investors assess concentration risk across their portfolios?

Rob Bittencourt: It's clearly a huge part of equity markets given the index concentration, and it's going to become a more concentrated part of fixed income indices as well. My suggestion would be: maintain a high-level view of your overall exposure, but don't approach the hyperscalers monolithically as carbon copies of one another. These are unique companies with diversified businesses. The underwrite of Amazon is not just tied to AI—it's their cloud business, their e-commerce business, their advertising business. Same for Google, Meta, Oracle, Microsoft. I would look at each uniquely and not view them blindly as pure AI plays, while still recognizing there's some correlation.

The good news is these are really, really good businesses—diversified, high-margin, continuing to grow outside of AI. Under current assumptions around their spend plans, they will remain high-quality counterparties. Where I'd be more concerned is outside the hyperscaler universe: single-site projects without high-quality counterparties, situations with real execution or technological risk where the credit quality is ultimately predicated on being a relative winner in the AI race. That's a harder bet to make.

Q: Let's move to the second theme: the changing world order. What channels do you expect it to impact markets?

Rob Bittencourt: The changing world order is an umbrella idea where many of the norms that have existed since World War II are starting to shift in fundamental ways. Trade is a clear example—the post-war trend toward free trade and globalization is starting to unwind as countries, not just the US, become more protectionist. The second is the move to a multipolar world. When the US was the de facto single superpower, the world was, at a large scale, more predictable geopolitically. That's no longer true, and it's going to lead to more volatility tied to geopolitical events. I don't think this ends with Ukraine or the Middle East conflict.

Specific to the US, you're starting to see real support develop around reindustrializing the country in strategic sectors—semiconductors, defense, pharmaceuticals. That's going to be a huge impetus for investment. So it touches trade, industrial policy, geopolitics, and conflict—all of which have far-reaching implications across markets, companies, and sectors.

Q: On US reindustrialization—what's the state of play, and why might private capital markets be better suited to financing it than traditional public markets?

Rob Bittencourt: At peak, manufacturing contributed about a quarter of US GDP. We're below 10% today. Getting back to even early 2000s levels would require several trillion dollars of investment. The push toward reindustrialization is being driven by a few converging forces. The pandemic scrambled supply chains and made very apparent that the US—and other countries—had real vulnerabilities and dependencies that needed addressing. Technology is also reshaping national security and defense in ways that require domestic industrial capability. And technology is also an enabler: things like additive manufacturing and AI can at least collapse the cost delta between US-based and lower-cost offshore manufacturing to make it economically feasible.

What does this mean for financing? Investment in US manufacturing facilities has already more than doubled over the past couple of years, driven largely by semiconductor investments—and I suspect that broadens into other sectors. These investments are capital-intensive by definition—you're building hard assets. The US hasn't been in the business of building out this ecosystem at scale, which requires flexibility to manage the project-oriented nature of the process. Private capital is very well positioned for that: you're lending against a hard asset, you're working hand-in-hand with a company to understand their investment plan and rationale in detail—something that's harder to do on a publicly syndicated basis.

Q: The third theme is the future of finance. What forces are shaping it, and how will they change the way people earn, spend, save, borrow, and invest?

Rob Bittencourt: There are a couple of underlying trends. Technology is an obvious one—and not just AI. Tokenization, digital assets, the rise of stablecoins. These are new financial primitives that didn't exist 10 years ago and are being adopted particularly among younger generations. The second is demographics: an aging population creating demand for stable fixed income, while at the same time, we're seeing the early stages of a large generational wealth transfer to younger generations whose relationship with finance looks quite different.

That said, the financial industry has historically demonstrated a remarkable ability to adapt—to take new technologies, graft them onto existing businesses, and evolve. If you look at the most valuable publicly traded financial companies today, there's a lot of similarity to what that list looked like 50 years ago, which is not the case in most industries. So while disruptive new businesses have been created in finance, the incumbents have proven more resilient than in many other sectors.

Q: You've written about a major shift in venture capital over the past decade—it's almost become a different asset class, with so much capital flowing into AI, defense, and advanced manufacturing. How does that shift change the opportunity for Apollo?

Rob Bittencourt: We compared a snapshot of the VC industry in 2025 to 2015. A few things jumped out. It's a much larger industry overall. It's more concentrated—large AI-focused companies approaching trillion-dollar valuations are absorbing an enormous share; roughly 80% of VC investment in the first half of this year has been in AI. But the third thing that jumped out is how different the nature of the companies being funded is. In 2015, the biggest names were Uber and Airbnb—amazing businesses, but asset-light by design. Software was also a dominant sector. These were high-margin, capital-light businesses. Fast forward to 2025: the dominant categories are AI and defense tech, which in many cases are industrial businesses. You actually need to build factories to produce what you're designing.

That's very relevant to Apollo. As these businesses scale, we've been partnering directly or indirectly with companies like xAI, SpaceX, and Anthropic through the Broadcom deal to help finance the capital intensity associated with AI. I think similar opportunities will extend well beyond AI into advanced manufacturing and parts of the defense tech sector. We can bring efficiently priced capital versus a straight venture equity raise, with experience in customizing solutions to specific projects, which is exactly the kind of financing a lot of these capital-intensive startups actually need.

Q: Final question, on a more personal note—you're an avid reader. What have you been reading lately that you'd recommend?

Rob Bittencourt: Two books—one I'm reading, one I'm listening to. The first is Skunkworks, a memoir by the former head of Lockheed Martin's secretive R&D lab that developed iconic aircraft like the U-2 spy plane and the SR-71 Blackbird. We've been talking a lot about reindustrialization, and one thing that gives me genuine confidence and optimism is that the United States has a rich history of building leading-edge techno-industrial businesses. The engineering that produced the SR-71 in the 1950s was arguably as magical at the time as ChatGPT was in 2023.

The second is The Power Broker, Robert Caro's biography of Robert Moses, who built much of the highway and bridge infrastructure of the New York City metro area. It spans the first three quarters of the twentieth century and gives you a remarkable historical perspective on how dramatically things changed during that period. It's not without controversy—there's a genuine dark side to Moses' story—but for me, the most compelling thing is that it's essentially a historical anthology of New York City across more than 50 years.

Note: This transcript has been edited for clarity. It does not represent a verbatim transcript of the podcast. Podcast recorded on July 15, 2026.