Howard Marks: CNBC The Exchange (2026)

Summary

Howard Marks argues that markets can exaggerate both optimism and fear, so investors should not treat market moves as reliable signals of what is fundamentally true. He sees AI as unusually difficult to forecast: it is unclear how products will be used, which companies will succeed, or what future revenues and margins will look like. He therefore favors measured, risk-aware exposure over trying to predict specific winners.

Marks draws a sharp distinction between investing in equity and lending. Equity can participate in potentially large gains, while a lender’s upside is generally capped at interest and principal repayment—and losses are possible if repayment fails. Investors seeking AI-related upside may be better served by stocks, he says. Lenders should participate only when repayment is highly likely and compensation is adequate for the risk.

Oaktree does invest in AI-related infrastructure, especially data-center projects, but Marks describes a preference for safer approaches. He sees energy supply as a potential constraint on AI infrastructure expansion. He also notes that technology-company debt gives credit investors more choices, but apparently reliable borrowers may not offer attractive returns or terms.

On interest rates, Marks declines to predict the Federal Reserve’s next move. He expects rates to remain around current levels over the next few years, sees little chance of a return to the unusually low rates of 2009–2021, and allows that inflation could push rates somewhat higher. He thinks the AI investment cycle is still relatively early—perhaps around its “third inning”—but emphasizes that future profitability and winners remain uncertain.

Assets, sectors, and instruments mentioned

• Equities and stocks: Marks sees these as a way to obtain substantial upside from AI, while acknowledging the risks.
• Debt and credit: Includes investment-grade and high-yield credit, as well as technology-company borrowing. Marks emphasizes repayment risk and questions whether some apparently reliable issuers offer adequate returns.
• AI and technology: AI models, software, computing capacity, and “neo-cloud” businesses.
• AI infrastructure: Data centers and AI “factories.”
• Energy: Identified as a potential bottleneck to AI infrastructure expansion.
• Companies and organizations: Anthropic and OpenAI (noted as private), Meta, Microsoft, Brookfield, Oaktree, the S&P 500, and the Federal Reserve.
• Other references: Claude, an AI tool, in Marks’s discussion of workplace use.

Marks does not endorse any specific public stock or credit investment, and says he does not know enough to argue for particular company picks.

Investment framework and risk management

• Separate equity from credit risk: Equity offers participation in growth; lending generally offers capped returns, with downside exposure if repayment fails.
• Require confidence in repayment when lending: Lend only when repayment is highly likely and the income compensates for uncertainty.
• Prefer safer exposure in highly uncertain areas: Marks favors safer ways to invest in AI despite its potential, including selected infrastructure projects.
• Assess project constraints: Oaktree seeks data-center projects in less congested locations, where restrictions can limit excessive construction.
• Do not rely on market sentiment: Marks says markets exaggerate developments, and volatility can reflect shifting optimism rather than new fundamental information.
• Retain human judgment: AI can help organize data, write code, and work with spreadsheets. But Marks says the key advantage is drawing sound conclusions from the data—not delegating that judgment to a model.
• Avoid overconfidence and unsupported bets: Marks cites the danger of believing something “for sure” when it is not true. He says he will not make risky bets without more information.

Key numbers, timelines, and market context

• AI product outlook: Marks says it is difficult to know how companies’ products will be used five years from now, or to forecast sales, prices, costs, margins, and winners.
• Energy demand: He raises the question of where three to ten times more energy could come from as AI infrastructure expands in the coming years.
• AI-cycle stage: Marks describes the cycle as near its beginning, while also using the analogy of being around the “third inning.”
• Interest rates: He expects rates to remain around current levels over the next few years, with a possible modest increase if inflation concerns persist. He considers the low-rate period from 2009 to 2021 anomalous and says this has been his thesis for about four years.
• Corporate AI adoption: Marks says AI use remains concentrated in a relatively small number of firms. He cites an estimate that S&P 500 companies generally spend about one-tenth of one percent of revenue on it, while noting that adoption may rise substantially.
• Market volatility: Marks describes recent swings—including episodes over the prior year or year and a half—as useful interruptions to excessive optimism. He does not claim that these moves establish that AI investment is a bubble.
• Bubble assessment: He distinguishes the reality of AI from whether valuations or spending are excessive. He says he cannot determine the latter without knowing future earnings and profitability.

Disclosures and cautions

• Marks makes no specific prediction about the Fed’s next move and says he does not believe in forecasts, particularly his own.
• He does not identify specific companies where he believes his view differs from the market.
• He says there is not enough information to know AI’s eventual winners, losers, or profitability.
• No explicit “not financial advice” disclaimer appears in the subtitles.

Presenters and sources: Howard Marks, Oaktree co-founder and co-chairman; Kelly Evans, interviewer on CNBC’s The Exchange.