Oakmark Fixed Income Market Q2 2026 Commentary

Learning from history

Lately I’ve found myself reading more history than fiction. What keeps pulling me back isn’t the history itself. It’s the timeless lessons about how thoughtful leaders make decisions when the stakes are highest and every option appears unattractive.

I recently finished One Minute to Midnight, Michael Dobbs’ account of the Cuban Missile Crisis. What struck me wasn’t how close the world came to nuclear war. It was how quickly the debate became framed as a binary choice: invade Cuba or accept Soviet nuclear missiles ninety miles from Florida.

But President Kennedy rejected that framing. Rather than accepting one of the two choices presented, his administration created a third option—a naval quarantine that bought time, preserved flexibility and ultimately helped avoid catastrophe.

The lesson I took away wasn’t about geopolitical prowess. It was about decision-making. The most dangerous mistakes often begin by accepting someone else’s framing of the problem.

Unfortunately, investors fall for this mistake all the time.

One of the traps investors fall into is believing there are only two possible outcomes: a business either succeeds or fails, an industry either thrives or disappears, or a transformational technology either changes everything or changes nothing.

I think of this as the “Certainty Trap”.

The reality is that the world is rarely binary. Outcomes fall along a spectrum, and the businesses and industries that matter most are usually shaped by many possible paths, not two.

At Harris | Oakmark, uncertainty doesn’t cause us to retreat. It causes us to become more disciplined. As the range of outcomes widens, we spend less time predicting the future and more time identifying securities where today’s price already compensates us for that uncertainty.

Artificial intelligence – The Great Certainty Trap

Artificial intelligence may be today’s best example of the Certainty Trap. Depending on who you ask, AI will either change everything or prove to be another technology bubble.

My personal view is that AI has the potential to become the most important technological advancement since the Industrial Revolution. The potential applications are staggering: scientific discovery, drug development, medical research, supply chains, manufacturing, software development and everyday decision-making. If given sufficient time to mature—and absent major political or regulatory constraints (neither of which should be taken for granted)—I believe AI’s long-term economic impact could ultimately prove far greater than we can reasonably appreciate today.

Potential, however, is not inevitability.

Comparisons to the dot-com era are common today. Although those comparisons can be useful, they are also incomplete.

Certainly, the internet ultimately transformed the global economy. But many of the companies leading its first wave didn’t survive long enough to participate in the eventual success. Often, the problem wasn’t the pursuit of bad ideas or even bad execution. They simply ran out of time.

The biggest difference between then and now isn’t enthusiasm. It’s financing.

Twenty-five years ago, most technology companies depended on equity markets to fund growth. When speculative capital disappeared, many promising business models disappeared with it.

Today’s AI investment cycle begins from a dramatically different position. The companies making the largest AI investments collectively generate well over half a trillion dollars of annual operating cash flow and possess balance sheets capable of funding hundreds of billions of dollars of additional investment without materially impairing their credit quality.

That distinction matters. Twenty-five years ago, the capital markets largely determined how long companies could pursue ambitious growth plans. Today, the companies themselves do.

That means AI, which is still in the 1.0 era, is likely to have many more years to evolve, compared to the internet companies that disappeared during the dot-com bust. But more time doesn’t guarantee a better outcome. It simply gives the technology a longer runway to prove its economic value.

That doesn’t eliminate uncertainty. It just changes where the uncertainty lies.

The question is no longer whether the largest AI investors can afford to keep building. The question is whether the returns ultimately justify what they build.

Uncertainty remains everywhere—from energy generation and semiconductor supply to regulation, enterprise adoption, competition and inference economics. Some of the most important drivers of AI’s future value, by definition, haven’t even emerged yet.

That’s what makes AI such a dangerous Certainty Trap. Its future is unlikely to be defined by a single outcome, but by a wide range of possibilities—some extraordinarily positive, some meaningfully disappointing and many somewhere in between. The mistake isn’t having an opinion about the outcome. It’s building a portfolio as though one outcome is certain.

Harris | Oakmark Fixed income approach to AI – Avoid the trap

As uncertainty expands, the instinct shouldn’t be to increase conviction. It should be to become more selective. Completely ignoring AI would, in my opinion, be just as imprudent as betting the portfolio on one narrowly defined outcome. Zero exposure assumes AI won’t matter. An all-in approach assumes we already know exactly how it unfolds. I believe both positions require more certainty than the evidence supports.

Instead of asking, “Who wins AI?” we ask a different question: “Which businesses can create value across the widest range of AI outcomes?” That question has shaped every AI-related investment we’ve made.

Meta’s (META) bonds are a good example.

Meta reaches more people every day than any other consumer platform in the world, serving approximately 3.56 billion daily active people across Facebook, Instagram, WhatsApp and Messenger. With that scale, Meta has multiple paths to creating value from AI. Better advertising, greater engagement, improved content discovery and higher productivity are all meaningful opportunities. Plus, Meta’s success doesn’t depend on the success of one specific large language model.

Mark Zuckerberg has repeatedly demonstrated an ability to adapt the business as technology evolves. Whether AI ultimately creates value through proprietary models, open-source innovation or embedded applications, Meta has multiple ways to benefit.

To us, that’s an attractive way to gain AI exposure. We do not think Meta requires one specific AI outcome to succeed, and neither does our investment.

Oracle (ORCL), which we touched on briefly last quarter, is another example.

In our view, its enterprise relationships, proprietary software and decades of customer data position the company to benefit regardless of how the competitive landscape evolves.

Oracle isn’t a singular bet on OpenAI, despite many attempts to pigeonhole it as one. Our investment depends on Oracle’s ability to adapt the pace of its AI investments as demand evolves while continuing to leverage one of the world’s deepest enterprise software ecosystems and one of the largest repositories of mission-critical corporate data.

We aren’t underwriting one large language model’s fate. We’re underwriting an investment-grade balance sheet, one of the world’s deepest enterprise software ecosystems, decades of mission-critical customer relationships, highly recurring cash flows from Oracle’s legacy software franchise, and the opportunity to create value by helping existing customers apply AI to the data, applications and workflows already residing on Oracle’s platforms.

We’ve also selectively invested in investment-grade, single-tenant data center bonds in support of hyperscale computing. In simple terms, these are facilities built for a single customer—typically one of the world’s largest technology companies—under long-term contractual arrangements.

Here, we’re not underwriting the success of any single AI application. We’re underwriting the resilience of the anchor tenant, only a modest level of ongoing demand for AI compute, and the contractual protections supporting our bonds. Even if demand for AI compute were to decline meaningfully, these facilities represent the first generation of AI infrastructure and only a small fraction of the compute capacity expected over time. While this isn’t true of every single-tenant data center bond, the investments we own benefit from meaningful principal amortization before maturity, further reducing the risk of principal impairment under adverse scenarios. In these investments, we’re able to earn nearly twice the credit spread of the underlying anchor tenant while underwriting what we view as fundamentally similar long-term credit risk. The Certainty Trap isn’t just about the investments we choose to own. It’s also about the investments we choose to avoid.

Several high-yield NeoCloud issuers illustrate this point. NeoClouds are a new generation of infrastructure companies that sit between the hyperscalers and AI developers, aggregating and reselling computing capacity to customers building AI applications. These are rapidly growing businesses meeting a very real need while demand for AI compute continues to exceed available supply.

What gives us pause is the combination of a high debt load and their role as what could ultimately become the marginal suppliers of compute capacity. WeWork offers a useful analogy. Flexible office space wasn’t the problem. It addressed a genuine need as the workforce became less tied to traditional offices. The problem was the economics of being the marginal provider of new supply while carrying a highly leveraged balance sheet. When demand fell short of expectations, the newest and most leveraged capacity became uneconomic first. Equity holders were wiped out, and creditors suffered substantial losses.

We see some of those same characteristics in portions of today’s NeoCloud market. These companies often carry high debt loads while building capacity that, if today’s supply shortage eventually normalizes, could become the marginal layer of AI compute. Many of their unsecured bonds also lack the structural protections that can help preserve principal if conditions deteriorate. If AI infrastructure demand ultimately proves less robust than today’s market expects, bondholders could face meaningful losses on principal.

That dynamic is very different from the AI investments we own. A slowdown in revenue growth is manageable for many businesses. For companies like Meta, Oracle, or the single-tenant data center investments discussed earlier, lower future revenue expectations would not materially change our assessment of their medium-term default risk. In the case of our single-tenant data center investments, we also benefit from structural protections that, even under our most pessimistic AI demand scenarios, should allow recoveries at or near par.

We believe the question is straightforward: Is the additional yield available in these securities, currently on the order of a few hundred basis points, sufficient compensation for the incremental risk of capital loss relative to the AI investments we already own? In our view, many of these highly indebted issuers require a high degree of certainty that AI infrastructure demand will remain persistently strong to justify what is ultimately a modest increase in yield.

For us, that’s the Certainty Trap. When the range of plausible AI outcomes is unusually wide, we’re reluctant to reach for a modest amount of additional yield if doing so requires underwriting one very specific future. We’d rather own bonds where today’s price already compensates us across a wide range of possible outcomes.

Conclusion

The temptation is to reduce uncertainty to a simple binary choice.

“AI changes everything. AI is a bubble. Buy. Sell.”

Those make for great CNBC soundbites. They rarely make for great investment decisions.

Artificial intelligence will almost certainly surprise us. Some expectations will prove too optimistic. Others won’t be optimistic enough. Some companies will create extraordinary value. Others will disappoint despite participating in the same technological revolution.

Fortunately, our investment process doesn’t require us to know exactly how that story unfolds.

Our responsibility isn’t to predict a single future. It’s to identify investments where today’s price more than compensates us for the uncertainty ahead. We look for opportunities that can succeed across a broad range of plausible outcomes rather than those that depend upon one very specific future.

That’s how we seek to avoid the Certainty Trap.

Adam D. Abbas, Head of Fixed Income and Portfolio Manager

Original Post

Editor’s Note: The summary bullets for this article were chosen by Seeking Alpha editors.