A Layman Thinks About AI

NVIDIA

Across the shows: 104 speakers in other episodes, 86 bullish and 58 bearish calls, past 30 days

• Howard Marks · Oaktree Capital ManagementBullish

  Nvidia's large profits show the value of its strong competitive position in AI chips.

  Marks cites Nvidia and TSMC as the big chip producers racking up large profits in the current environment. He takes this as evidence of the value of a strong competitive position in AI. He treats chip producers as among the clearer beneficiaries of AI spending, while noting that chip life and obsolescence remain open questions for the rest of the stack.

  “The big chip producers, NVIDIA and TSMC, are racking up the big profits these days, demonstrating the value of a strong competitive position.”at 9:56

Taiwan Semiconductor Manufacturing Company

Across the shows: 9 speakers in other episodes, 8 bullish and 0 bearish calls, past 30 days

• Howard Marks · Oaktree Capital ManagementBullish

  TSMC's large profits show the value of its strong competitive position in AI chips.

  Marks names TSMC alongside Nvidia as a big chip producer racking up large profits, which he cites as evidence of the value of a strong competitive position. He also notes that nearly all leading-edge chips come from TSMC, which underlines its central role in AI.

  “The big chip producers, NVIDIA and TSMC, are racking up the big profits these days, demonstrating the value of a strong competitive position.”at 9:56

AI infrastructure & data centers

• Howard Marks · Oaktree Capital Management

  Data center investing is not a safe bet: constraints and overbuilding risk, though supply curbs could help existing sites.

  Marks argues data centers look safer than AI developers because they are bricks and mortar, but most of their cost is GPUs whose obsolescence risk falls on the operator, not the building shell. Developers face site acquisition, regulatory hurdles and local opposition, plus bottlenecks in power, cooling, construction and financing, and financing costs could surprise. He sees picks-and-shovels exposure as a way to avoid having to pick AI winners, but overbuilding could create oversupply. Moratoriums on data center construction could make existing and future sites more profitable and valuable. He concludes the winners' and losers' results may be more pronounced than usual in infrastructure.

  “The key here is that data centers aren't a layup. The intelligent investor has to find opportunities that are insulated from the uncertainties just listed, and won't be subject to overbuilding and thus oversupply of compute.”at 17:47

• Howard Marks · Oaktree Capital Management

  AI needs about $1 trillion of new capital a year, with hundreds of billions likely raised as debt.

  Marks cites Goldman Sachs data showing roughly $600 billion of AI-related debt issued globally so far this year, around $400 billion of it in US investment grade, which is over a fifth of IG supply. From that pace he projects AI needing an average of $1 trillion of new capital per year in the period ahead, with hundreds of billions of it continuing to be raised as debt each year. This scale is the basis for his later concerns about rates, equity capacity and credit quality.

  “AI seems likely to require an average of $1,000,000,000,000 of new capital per year in the period ahead, and hundreds of billions of that are likely to continue to be raised each year through the issuance of debt.”at 22:46

AI: overall outlook

• Howard Marks · Oaktree Capital Management

  AI's direction is fairly predictable, but what investors get paid for from it is much less predictable.

  Marks asked Claude to rate AI predictability from 1 (unknowable) to 10 (fully predictable). It gave AI overall about a 3, split into a direction score of about 6 to 7, since the technology will keep improving and spread through the economy, and a score of about 2 for the things investors get paid for: how fast it happens, how big the profits are, and who keeps them. Marks adopts this framing and presents it as the core of his assessment, which explains why he thinks the investment outcome is unusually hard to predict even though the technology trajectory is clearer.

  “AI as a whole scores about a three. ... The direction is fairly predictable, maybe a six or seven. ... The things investors get paid for are much less predictable, around a two, how fast it happens, how big the profits are, and who keeps them.”at 28:45

• Howard Marks · Oaktree Capital Management

  AI will produce many surprises, but they needn't be unbearable, so investors should not ignore it.

  Marks says that AI is sure to produce many surprises, and that investors must accept they do not know everything about it. He argues the key question is the probability that an AI investment will disappoint versus the probability that it won't. Uncertainty is therefore a reason for careful probability-weighing rather than avoidance, and he frames AI as 'too important' to ignore. He gives no specific numbers or forecasts for AI outcomes, so this is a general view on how to approach the sector.

  “AI is sure to produce a lot of surprises, but they needn't be unbearable. The thing to ponder is, what's the probability an AI investment will disappoint, and what's the probability it won't?”at 40:05

AI bubble & tech valuations

• Howard Marks · Oaktree Capital Management

  AI technology isn't an illusion, but whether AI spending and prices are justified is largely a matter of faith.

  Marks defines a bubble as irrational exuberance leading to imprudent behavior, excessive investment and too-high prices, and says it must be judged on three counts: whether the technology is overhyped, whether spending is excessive with inadequate returns, and whether investors pay too much for stakes. He believes AI will be powerful and consequential, so the technology is not an illusion. But whether the trillions in capex will earn adequate returns depends on how AI will be used and who will profit, and until those are known, neither the spending nor the prices of stocks and debt can be judged. Talking to knowledgeable people, he concludes this is largely a matter of faith.

  “The conclusion I draw from my conversations with people I consider knowledgeable is that, at this point, answering this question is largely a matter of faith.”at 36:36

• Howard Marks · Oaktree Capital Management

  For AI, judging intrinsic value and whether an investment is cheap is much harder because of unresolved questions.

  Marks argues that his usual method of investing, which depends on price relative to intrinsic value, is much harder to apply to AI. He says a high level of uncertainty can be acceptable if the price is low enough relative to intrinsic value, but for AI, estimating intrinsic value and therefore cheapness is 'greatly complicated' by open questions. Those questions include who will profit from AI, how much power availability constrains buildout, and what residual asset value will be. Because those cannot be answered definitively, the valuation of AI investments is inherently hard to pin down.

  “Note, however, that in the case of AI, thinking about intrinsic value, and thus, cheapness, is greatly complicated by the questions just raised.”at 38:53

AI pricing & unit economics

• Howard Marks · Oaktree Capital Management

  If AI models become commodities, vendor profits get ground down and the gains go to users.

  Marks argues that if several AI models become equally capable, their service could become a price-driven commodity. In that case profitability and return on investment would be ground down, and the economic benefits could flow more to AI users than to vendors. He points to declining prices per token and cheaper, open-source and Chinese models as pressures on vendor pricing. The argument is framed as a question and carries a hedge, so it is a tentative view.

  “Will their service become an undifferentiated commodity that sells on the basis of price? If it does, will profitability and ROI be ground down, in which case the economic benefits may go more to the users of AI than to the vendors?”at 9:16

AI & jobs

• Howard Marks · Oaktree Capital Management

  AI will likely replace many workers, so it is probably going to be taxed like the people it displaces.

  Marks argues that one of AI's most likely effects is replacing workers. Governments will need to replace tax revenue that came from those workers, and many top earners already face combined rates of 50% or more. He asks what would stop AI from being taxed like people, and what tax bracket it would fall in. This is a tentative view framed as a question, and he notes he has heard little discussion of it.

  “Given that one of AI's most likely effects will be replacing workers, and given government's need to replace the tax revenues that previously came from taxing the people AI puts out of work, what's to say AI won't be taxed like people?”at 10:12

US growth & recession risk

• Howard Marks · Oaktree Capital Management

  AI capex is carrying most of US economic growth while the rest of construction declines.

  Marks argues AI is a small part of the US economy but is responsible for an outsized part of economic growth. AI's contribution to GDP growth currently comes overwhelmingly through capital expenditures, mainly data center construction, which is rising rapidly while the rest of construction is in decline. He adds that AI-related workers' spending on rent, food and cars carries AI's effect beyond the sector. This concentration is why he sees the economy as dependent on a single factor.

  “AI represents a small part of The US economy but is responsible for an outsized part of economic growth. AI's contribution to GDP growth currently comes overwhelmingly through capital expenditures, primarily for construction of data centers.”at 18:45

US stock market outlook

• Howard Marks · Oaktree Capital Management

  The S&P 500 is concentrated in a few AI and tech companies, leaving the market vulnerable to AI setbacks.

  Marks cites JPMorgan data showing 42 AI-related companies generated about two-thirds of S&P 500 price gains since January 2024. He also notes that Magnificent Seven earnings have outgrown the rest of the index for 14 consecutive quarters, and that AI capex is carrying US GDP growth while other construction declines. He argues that because the economy and markets rely on a few undiversified factors, AI has become a 'fault line' through portfolios. Negative developments in chip life, corporate strategy or compute demand could have far-reaching consequences, so the market is more vulnerable than it would be with broad-based drivers.

  “It's incontrovertible that disproportionate reliance on a single sector makes the economy and stock market more vulnerable than they would be if the factors underlying them were broad based.”at 20:59

US Treasuries & yields

• Howard Marks · Oaktree Capital Management

  Rate-insensitive hyperscaler borrowing may push up the interest rates the US government must pay.

  Marks compares AI borrowing with the classic crowding-out effect, in which heavy government borrowing makes it hard for the private sector to get capital. Hyperscalers and frontier labs can't afford to slow AI investment because rates are high, so they keep borrowing at any price. He thinks this rate-insensitive demand for debt could raise the rates the government pays. He asks whether it has already contributed to the recent rise in rates, which suggests he believes it has, but he presents this as a question rather than a firm conclusion.

  “At present, it feels to me that the demand for debt from rate insensitive hyperscalers willing to pay whatever it takes has the potential to elevate the interest rates the government must pay.”at 23:42

IPOs, M&A & deal activity

• Howard Marks · Oaktree Capital Management

  Equity markets may not supply the hundreds of billions of new AI equity needed yearly, and AI debt and equity will compete.

  Marks notes that traditional US IPO proceeds were below $50 billion in eight of the last ten years, with exceptions of about $80 billion in 2020 and $140 billion in 2021. He suspects there is only so much new capital available in a given year. He also argues that AI's demand for debt and for equity, which AI requires alongside company cash flow, will ultimately compete for that same limited pool. Together these points suggest the equity market may struggle to fund AI's needs.

  “I suspect there's just so much new capital available in a given year. Won't AI's appetites for both debt and equity ultimately compete against each other for it?”at 24:24

Credit markets

• Howard Marks · Oaktree Capital Management

  Lenders to AI face concentrated, fast-changing debt that depends on compute revenues, so they must be vigilant.

  AI-related bonds are about 12% of the Bloomberg USD IG corporate index and about 6% of US high yield. Some data center debt rests on hyperscaler tenants' investment grade ratings, while other debt is collateralized by chips. Marks warns that data center leases can let tenants terminate if uptime falls short of 99.9%, and that chip collateral may not last as long as the debt. He also flags circular vendor financing and special purpose vehicles that keep debt off hyperscaler balance sheets, and asks whether hyperscaler creditworthiness and ratings may change as they add leverage. He concludes lenders face a new era and must scrutinize borrowers and deal terms closely.

  “Lenders to AI face a new era, marked by debt concentrated in a single industry, rapid change in industry fundamentals, and a high degree of reliance on the level of future revenues from sales of compute.”at 27:19

Frontier AI labs & model race

• Howard Marks · Oaktree Capital Management

  Frontier labs like OpenAI, Anthropic and xAI are practically unknowable, with huge capital needs and unclear moats.

  Marks cites Claude's rating of frontier labs (OpenAI, Anthropic, xAI) at 2 out of 10 predictability, which he reads as practically unknowable. He calls them impressive with big leads, but notes they are close to binary bets with huge capital needs, falling prices per unit of capability, and unclear lasting moats. He points out that these labs were unknown only a few years ago, so today's leaders could be displaced, and that the most uncertain propositions seem priced at a premium in private markets. He flags that OpenAI and Anthropic have filed for IPOs and that the public pricing of their shares will be telling.

  “For me, the surprise was the Frontier Labs score of two, which I take to mean 'practically unknowable'.”at 30:27

US–China AI & chip race

• Howard Marks · Oaktree Capital Management

  US limits on advanced chip sales push China toward cheap models that undercut US models.

  Marks notes that nearly all leading-edge chips come from TSMC in Taiwan, and that the US restricts sales of the most advanced chips to China. He argues this encourages China to offer low-cost models that can undercut more powerful US models. He asks what happens to US competitiveness if this changes, whether Chinese chips could catch up, and whether China could restrict Taiwanese chip sales to the US. These geopolitical risks are a key source of unpredictability for AI investors.

  “The US restricts sales to China of the most advanced chips. This encourages China to offer low cost models that can undercut sales of more powerful US models.”at 31:45