Bearish on Long-Term Treasuries: What DoubleLine's Sherman Call Means for the Future of AI

On the surface, a bond manager saying he is still downbeat on long-term U.S. government debt looks like a story for Wall Street traders only. It is not. That single view, DoubleLine's Sherman staying bearish on long-term Treasuries, sends a signal that runs straight through the middle of the AI boom. It touches the price of money, the cost of building data centers, the mood of venture capital, and the way every company budgets for artificial intelligence tools.

Here is the simple version. Long-term Treasury yields are the reference point for the cost of money over many years. When that reference point stays high, everything that pays off far in the future gets cheaper on paper today. And almost nothing in the modern economy pays off further into the future than AI infrastructure. A data center built now may run for a decade or more. A frontier model may take years to turn a profit. The bearish long-bond call is, in practice, a statement about how expensive patience has become.

So what does it mean for AI?

The Basic Mechanics: Why Long Bonds Matter to Everyone

When people talk about "long-term Treasuries," they mean government bonds that pay back over many years, often ten, twenty, or thirty. Their yields move in the opposite direction of their prices. A bearish view on these bonds means you expect prices to fall and yields to stay high or climb.

Why does that spread? Because long-term yields act like gravity for the whole financial system. Mortgage rates, corporate borrowing costs, and the discount rate investors use to value future profits all lean on them. When the long end of the curve stays elevated, the future gets discounted harder. A dollar of profit expected in 2035 is worth less today than it would be in a world of very cheap long-term money.

That is the core of it. AI is the most future-heavy bet in the global economy right now. So a permanent shift toward higher long-term yields hits AI harder than it hits most sectors.

AI Is the Ultimate Long-Duration Asset

Think about what an AI build-out actually requires. Land. Power. Cooling. Silicon. Fiber. And years of construction before a single dollar of revenue arrives. This is the profile of an infrastructure project, not a software app. Software scales with a credit card. Infrastructure scales with a loan.

That matters because infrastructure is financed with debt, and debt is priced off long-term rates. When the long end stays high:

• Borrowing gets more expensive. Data center developers and the companies leasing to them pay more to raise money. Projects that penciled out at lower rates stop penciling out.
• Payback windows stretch. If a project needs eight years to earn back its cost, a higher discount rate makes that wait feel much longer to investors.
• Capital gets pickier. Lenders and equity partners start asking harder questions about utilization, power contracts, and customer commitments before they write checks.

None of this stops AI. But it changes its shape. Instead of a broad gold rush, you get a narrower, better-financed race. The companies with the strongest balance sheets and the cheapest access to capital pull ahead. The marginal, thinly financed entrant struggles.

Channel One: The Private Markets and AI Funding

Much of the AI ecosystem runs on private capital, venture funds, growth equity, and private credit. A persistent bearish view on long bonds tends to keep the whole risk-free benchmark high, which means investors demand a bigger premium for taking risk on an unproven company.

What does that look like in practice?

First, valuations for late-stage AI firms get more disciplined. The very top names will still raise money. The middle of the pack will find that each round comes with more conditions and more dilution.

Second, the bar for revenue quality rises. Investors want recurring revenue from paying customers, not pilots and proofs of concept. "We have great technology" becomes a weaker pitch than "we have signed contracts."

Third, private credit steps in where equity does not. Expect more AI infrastructure to be financed with structured debt arrangements tied to specific hardware or specific customer contracts. That is a healthier model in some ways, it forces discipline, but it also concentrates risk in fewer hands.

Channel Two: Corporate AI Budgets

For the average business, the interest rate story is more subtle but just as real. When long-term rates stay high, the cost of capital for the whole company stays high. That makes every big spending decision compete with a very attractive alternative: paying down debt or holding cash that actually earns something.

Executives will start to ask a sharper question about AI projects. Not "will this transform us?" but "what does this return over the next eighteen months?" That is a fair question, and it is a healthy one. It pushes AI vendors to sell outcomes, not promises.

The practical result? AI adoption keeps growing but shifts toward use cases with fast, measurable payback:

• Automating repetitive back-office work with clear labor savings
• Cutting customer support costs with measured resolution rates
• Speeding up code review and testing with tracked cycle-time gains
• Improving demand forecasting with reduced inventory waste

Big, vague, five-year transformation programs will be broken into smaller phases with checkpoints. That is not a retreat from AI. It is AI growing up.

Channel Three: Energy, Power, and the Cost of Compute

There is a third channel that is easy to miss: power. AI data centers are becoming some of the largest single electricity consumers in the world, and much of the new generation being planned is long-lived infrastructure financed over decades.

Higher long-term rates make that generation more expensive to build. When the cost of financing a power plant rises, so does the price of the electricity it sells. That flows directly into the cost of compute.

So a bearish long-bond view does not just raise the cost of the chip. It raises the cost of the electricity that runs the chip, the cooling that protects it, and the transmission that connects it. Over time, this pushes the industry toward efficiency. Better model architectures. Better chips per watt. Smarter scheduling of workloads to where power is cheapest. Scarcity is a powerful teacher.

What This Means for the Future of AI

If long-term yields stay high for years, as the bearish case suggests, the AI industry of the late 2020s looks different from the one many people imagined. Here is the most likely shape of it:

1. Consolidation around strong balance sheets

The giants with huge cash reserves and top-tier credit can borrow cheaply relative to everyone else. That advantage compounds. Smaller players will increasingly rent compute rather than build it, which pushes the market toward a utility-style structure.

2. Compute as regulated infrastructure

As AI capacity becomes more capital intensive and more debt financed, expect it to look more like telecom or electricity, long contracts, predictable pricing, and heavier scrutiny from regulators and lenders alike. That is a real shift in how the industry operates.

3. Smaller, cheaper, more efficient models

When capital and power are expensive, you optimize. You do not throw the largest model at every problem. You route simple tasks to small models and reserve the heavy lifting for where it truly pays. This is good news for efficiency and good news for the planet.

4. A higher bar for "AGI-ish" spending

Blue-sky research does not disappear, but it concentrates in a handful of well-funded labs. Everyone else focuses on applied AI with measurable returns. The frontier still moves. It just moves in fewer places.

Actionable Playbook: What to Do Now

For business leaders, the message is not to slow down on AI. It is to change how you buy and build it.

• Model AI projects with a realistic cost of capital. If your finance team assumes cheap long-term money, redo the numbers. Use a higher hurdle rate for projects with long payback periods.
• Favor payback under twelve to eighteen months. Prioritize use cases where you can prove savings or new revenue within a year. These survive any rate environment.
• Lock in energy and compute contracts early. If you expect to need substantial compute capacity, long-term contracts protect you from future price rises. Do not assume prices only fall.
• Build a capital-light path. Buy or rent rather than build where you can. Owning infrastructure is only smart if you can finance it well.
• Track the long bond, not just the model release. Long-term yields are now a leading indicator for AI valuations and funding conditions. Make them part of your quarterly review.
• Measure ruthlessly. In a higher-rate world, the companies that can prove return on AI spend will get the funding. Everyone else gets questions.

The Bigger Picture

It is worth stepping back. A bearish long-bond view is fundamentally a bet that the era of ultra-cheap, ultra-long money is not coming back soon. Whether that proves right or wrong, the AI industry is already adapting to the possibility.

That adaptation is healthy. The first phase of the AI boom was about capability. The next phase is about economics. Can this thing pay for itself? Can it run profitably at scale? Can it survive a world where money costs something again?

Answering those questions well is not a downgrade for AI. It is what turns a technology wave into an enduring part of the economy. The companies and leaders who plan for a higher cost of capital today will be the ones still standing when the next cycle turns, and they will own the infrastructure everyone else has to rent.

The bond market's message is blunt: the future is not free. The smart response is to build an AI strategy that can pay for itself.