Rob Arnott on the Hidden Costs of Passive Investing and the Real Winners From AI

Ralph Lauren

• Rob Arnott · Research AffiliatesBullish

  Ralph Lauren is a smart AI user, using its 'Ask Ralph' app to sell product, not an AI innovator.

  Arnott cites Ralph Lauren's 'Ask Ralph' app, launched about a year earlier, which lets consumers get fashion advice from an AI trained on the founder's writing and interviews. He says it is not an AI revolutionary company but a smart user of AI, using it as another angle to sell product. He frames this as the type of AI user that captures value, consistent with his view that the big winners will be the rest of the economy.

  “Is it a smart user of AI using it as another angle to sell product? Yes.”at 9:08

Space Exploration Technologies

Across the shows: 20 speakers in other episodes, 14 bullish and 6 bearish calls, past 30 days

• Rob Arnott · Research AffiliatesBullish

  SpaceX will see forced index-inclusion buying after IPO, so he would buy at IPO and sell into indexation.

  S&P indexers hold over 25% of the float of each index stock. Once SpaceX is added, index funds must buy roughly 25% of its float, mostly from insiders and early holders who are unlikely to sell their favorite stock. Arnott expects a herd of index buyers to rush in at the closing price on the addition day to get zero tracking error. He was the only one of five asset managers asked who would buy at the IPO, because between the IPO and the index addition he could buy and sell into the forced indexation demand. This goes against his usual value discipline, but index flows create opportunities.

  “S and P indexers will have to buy 25% of the float from people who don't wanna sell. So I would buy and sell into the indexation.”at 22:37

Palantir Technologies

• Rob Arnott · Research AffiliatesBearish

  Palantir is the biggest frothy-multiple name his index deliberately leaves out.

  Arnott's index is lightly overweight the overlap stocks (the mag seven and dominant AI players) and excludes the frothy small businesses trading at high multiples, which make up about 5% of the S&P 500 and are not in his portfolio. He says Palantir is the biggest of those frothy names. The support for this view is brief, so the evidence is thin, but the implication is that Palantir's valuation is too high relative to fundamentals.

  “The little businesses trading at frothy multiples are 5% of the S and P. Palantir is the biggest of them for what it's worth.”at 35:27

NVIDIA

Across the shows: 26 speakers in other episodes, 21 bullish and 10 bearish calls, past 30 days

• Rob Arnott · Research AffiliatesBearish

  Nvidia lending to customers to buy its chips is a red flag, echoing dot-com and GM/GE vendor-finance risks.

  Arnott points to chip providers, with Nvidia as a prime example, lending money to customers to buy their products. He says that was a red flag in the dot-com bubble. Lending to buyers is not wrong in itself, since General Motors and General Electric did it and it helped them, but it sowed seeds of trouble in the global financial crisis. If a vendor must lend to customers to buy its product, the customers must deliver on what is built with those products, which in AI means chips.

  “many of the providers, NVIDIA is a beautiful example, are lending money to their customers to buy their product. And that was a red flag during the dot com bubble.”at 38:32

OpenAI

Across the shows: 17 speakers in other episodes, 7 bullish and 15 bearish calls, past 30 days

• Rob Arnott · Research AffiliatesBearish

  OpenAI's ChatGPT lead was matched by DeepSeek within two years, so its moat is weak.

  Arnott says OpenAI thought it had a moat around ChatGPT. Only about two years after launch, DeepSeek produced something as powerful as ChatGPT's then-cutting-edge version, at about one-hundredth of the investment. In his view, moats in AI are fragile, which weakens OpenAI's position.

  “OpenAI thought it had a moat around ChatGPT, and yet it was only two years after they introduced ChatGPT that DeepSeek came along and said, oh, we've got something as powerful as your current cutting edge ChatGPT version.”at 52:36

Anthropic

Across the shows: 33 speakers in other episodes, 21 bullish and 40 bearish calls, past 30 days

• Rob Arnott · Research AffiliatesUncertain

  Anthropic's valuation is astounding, but it has no moat unless it keeps its edge against rivals.

  Arnott notes Anthropic was launched around 2022 or 2023 and is already thought likely to be a $2 trillion company, which he calls unprecedented. He asks whether it has a moat and answers no, unless it keeps its edge. That is very hard when competitors know as much as it does. He thinks moats are fragile at best.

  “But does it have a moat? Not unless it ... keeps its edge, which is really, really, really hard to do when you've got committed competitors who know as much as you do.”at 53:34

Microsoft

Across the shows: 6 speakers in other episodes, 5 bullish and 0 bearish calls, past 30 days

• Rob Arnott · Research AffiliatesBullish

  Microsoft's leadership has put it about 2% a year ahead of the S&P 500 this century.

  Arnott says Microsoft struggled in 2000, regained its footing in the early 2010s, and surpassed the S&P 500 in 2018. After an 18-year wait it is now ahead by about 2% a year. He credits its leadership since the early 2010s with being very intelligent about how to run things. In his view, a durable moat comes from leadership that understands clients and delivers more value than it charges.

  “It regained its footing in the early two thousand tens, and it finally surpassed S and P in 2018 century to date. You had to wait eighteen years to be ahead of the S and P, and now you're ahead 2% a year. The leadership of Microsoft since the early two thousand tens has been really spot on.”at 55:40

US stock market outlook

• Rob Arnott · Research Affiliates

  Large-cap growth's valuation premium will mean-revert; small-cap value should beat it by 6-8% a year.

  Since 2014 the top 500 US companies outperformed the next 500 by about 4,000 basis points. Their relative valuation went from a 15% discount to mid-caps in 2014 to an 80% premium today, while the next 500's cash flow grew about 3% a year faster. So the gap is all revaluation alpha. Either fundamentals of the top 500 must catch up, or relative valuations must mean-revert. Arnott says the spread between large-cap growth and small-cap value is wider than ever, including the dot-com bubble, so the 1,400 companies in Russell 2000 Value are highly likely to outperform large-cap growth by roughly 600-800 basis points a year for the coming decade.

  “I see that spread widening to a point that has never ever ever been seen before, including in the dot com bubble. ... These 1,400 companies in Russell two thousand value are highly likely to outperform... by perhaps six, 800 basis points per annum for the coming decade”at 12:01

• Rob Arnott · Research Affiliates

  Index rebalancing creates hidden drag on index investors, costing about 25 basis points a year since 1989.

  Arnott argues index funds underperform because they trade at index-change prices. A thought experiment: a closet S&P 500 that traded at the close before each announced addition would have beaten the index by about 25 basis points a year since 1989. Tesla rose roughly 50% between its November 2020 announcement and its addition, and about 900% from March lows to addition, showing how index flows move prices. The drag is still there, just not visible in expense ratios, and passive index investors pay it through the price effect.

  “If you had a crystal ball and you knew S and P is gonna announce the addition of Tesla at the end of trading today... Your performance over the last... Since 1989 would have been about 25 basis points per annum better.”at 27:14

• Rob Arnott · Research Affiliates

  Fundamentals-based index weighting has beaten the S&P 500 by about 70 basis points a year for 35 years.

  Arnott's firm tested a strategy that buys companies once they become one of the thousand biggest businesses by fundamentals (business momentum) and sells them when their business erodes, rather than buying stocks because they soared. Going back 35 years with 500 stocks, it beat the S&P 500 by about 70 basis points a year. The strategy also has lower turnover, and its losers are sold at smaller average losses (about 20%, not 70%) than the cap-weighted approach's sales.

  “it beats the S and P 500 by 70 basis points per annum for thirty five years.”at 30:15

• Rob Arnott · Research Affiliates

  Small cap value is trading dirt cheap with a CAPE near 10, a better place than AI winners.

  Arnott contrasts the crowded AI trade with unnoticed parts of the market. Small cap value, trading dirt cheap at CAPE ratios of around 10, is one example he gives for where investors could rotate after gains in AI.

  “But if you make gobs of money on something, why not take some chips off the table and put it on the unnoticed parts of the market? Small cap value, emerging markets value, trading dirt cheap, cape ratios of 10.”at 45:09

• Rob Arnott · Research Affiliates

  Treating R&D and some G&A as capital doubles the book-value factor's historical edge.

  Arnott argues that GAAP treats R&D as an expense, like a bonfire, while it treats a desk as a capital asset. Companies only fund R&D when they expect it to pay off. His paper proposed counting R&D and a third of G&A as intangible capital depreciated over about ten years. Over the last 60 years, a simple price-to-book factor would have made a value investor four times as rich as a growth investor. Adding R&D and some G&A doubles that to eight times, and the effect is strongest over the last 30 years as the economy has become more intangible.

  “if you treat it as just like you do a desk, a purchase of a capital good that depreciates over maybe ten years, then the efficacy of a book value based value factor doubles. It's twice as effective.”at 49:04

• Rob Arnott · Research Affiliates

  Value has lagged only because its relative valuation fell further; mean reversion could make it stupendous.

  Arnott says value as a factor is out of favor because its performance has been poor. Looking at revaluation alpha, value has underperformed because its relative valuation has fallen even further. That means mean reversion could produce strong value returns over the coming decade. He argues that people extrapolate past performance, which is the mistake.

  “Value has underperformed, but it's underperformed because its relative valuation has fallen even further, which means any mean reversion could mean, would mean, that value could be absolutely stupendous in the coming decade.”at 1:07:53

AI: overall outlook

• Rob Arnott · Research Affiliates

  AI will be among the most powerful productivity enhancers for the global economy in our lifetimes.

  Arnott argues the AI revolution is real and very powerful, probably the most powerful productivity enhancer for the global macro economy in our lifetimes. He compares it to the advent of computers and the internet, saying AI takes that revolution to a new level. He grounds this in his firm's own experience of adopting large language models since 2023.

  “The AI revolution is incredibly powerful. It is very real. It is probably going to be the most powerful productivity enhancer for the global macro economy in our lifetimes.”at 2:50

• Rob Arnott · Research Affiliates

  In tech revolutions, customers and AI users capture more value than the innovators who build the technology.

  Arnott argues that every technological revolution requires customers to win more than innovators, because computer users made more money than computer makers. AI users find their time leveraged and get more work done, so AI is more valuable to them than to the AI innovators. Capitalism requires providing a product more valuable to the customer than its price, so the big winners will be the rest of the economy, not the AI builders.

  “Every technological revolution requires that the customers win more than the innovators.”at 7:51

• Rob Arnott · Research Affiliates

  So far AI produces adjacent ideas, not independent new ones; it may change within ten years.

  Arnott has worked with AI tools in his research and has not seen them come up with any independent new ideas. They surface work that is adjacent to new ideas and very close to existing writing, which is useful. He hedges that this probably will not be true in ten years, and he has not yet seen AI be creative.

  “In my experience, and this probably won't be true in ten years, but in my experience, I haven't seen AI come up with any independent new ideas in the times I've worked with it.”at 1:05:09

AI & jobs

• Rob Arnott · Research Affiliates

  AI will create more jobs than it destroys, as past technological revolutions did.

  Arnott says job losses to AI are overstated because technological revolutions always create more jobs than they destroy. He cites the Luddites and mechanical looms, the human 'computer' occupation eliminated by office computers, and the fact that nobody regrets those changes a generation later. He notes workers may lose jobs to people who use AI better, not to AI itself.

  “Technological revolutions always create more jobs than they destroy.”at 3:59

• Rob Arnott · Research Affiliates

  AI makes Research Affiliates' programmers three to four times as productive, with far fewer needed.

  Arnott says his programmers are probably three or four times as productive as before. He knows because the firm now has a third or a quarter as many programmers and gets more done. He also describes AI as a tool that leverages people's time across roles, and says staff who do not learn to use it are rolling the dice with their careers.

  “Our programmers are probably three or four times as productive as they used to be. I know because we have a third or a fourth as many programmers, and we're getting more done.”at 1:01:50

AI chips & semiconductors

• Rob Arnott · Research Affiliates

  Chipmakers' profits need AI customers to win, yet the market prices those customers as irrelevant.

  Arnott's logic runs down the AI value chain. AI users must value the product above its cost so AI vendors can value chips above their cost, which lets Nvidia, Micron and other chipmakers earn a profit on AI hardware. The end customer therefore has to succeed. Yet the market prices chipmakers richly and the end customers at deep discounts, as if AI will push them aside. Arnott calls that inconsistent, since the customers must succeed for the whole chain to work.

  “the AI customers have to value the AI product by more than it costs in order for the AI vendors to value the chips for more than they cost, in order for NVIDIA and Micron and the rest of those to be able to turn a profit ... the market is pricing things as if the end customer is irrelevant.”at 41:00

• Kai Wu · Sparkline Capital

  If enterprises see no AI ROI and stop buying, chipmakers and memory stocks will suffer.

  Kai argues that AI infrastructure makers cannot earn a profit on what they build unless customers buy it, and customers will only buy if they see ROI. Early on, companies experiment. Eventually enterprises will ask whether an AI tool is worth about a million dollars a year, and if not they will shut it down, hitting memory stocks and chipmakers. He says the market cannot have it both ways, either AI transforms the economy and everyone wins or it does not.

  “the fix and shovels can't actually drive a profit on ... all the infrastructure they're building unless people actually buy it. ... And no one will buy it if they don't see ROI.”at 42:08

AI infrastructure & data centers

• Rob Arnott · Research Affiliates

  Hyperscalers' 5-7 year chip depreciation is too long; a $1T AI spend implies $500B of write-offs.

  Chris Brightman's paper noted that a buyer of about $40 billion of chips was depreciating them over five to seven years, as most hyperscalers do. Arnott argues that Moore's law keeps producing faster chips, so older ones become uneconomic in about three years because of power and cooling costs, not because they break. Depreciating over six years therefore overstates earnings. On a $1 trillion annual spend, about $500 billion would be written off within three years. That write-off has not yet reached the books, but he expects it.

  “Let's say there's a trillion dollars of spend in a given year on AI tech. That means 500,000,000,000 of it will be a write off in three years. And that write off hasn't begun to hit the books, but it will happen.”at 44:52

• Kai Wu · Sparkline Capital

  AI chips go obsolete fast, making data center capex a treadmill with a narrow ROI window.

  Kai argues that buildings, land and fiber last a long time, as in past infrastructure booms, but GPUs become obsolete quickly. Companies must keep reinvesting just to stand still, and the window in which they can earn returns on this spending is narrow and must come quickly. He calls it a treadmill, a tough business, and suggests AI companies could come to resemble utilities or telecoms, with AI tokens as a commodity.

  “But the GPUs themselves, because of how fast technology progresses, they go obsolete pretty quick. ... You have to just keep investing in these ships just to stand still ... It's a treadmill. It's a very, like, tough business to be in.”at 45:50

AI bubble & tech valuations

• Rob Arnott · Research Affiliates

  Tech revolutions bring bubbles; AI leaders are priced as if dominance is assured, which history doesn't support.

  Arnott argues every technological revolution comes with a bubble, and the market treats today's AI leaders as having moats that will protect them. He cites the dot-com era: of the 10 most valuable tech stocks in 2000, only Microsoft beat the S&P 500 over the following quarter century, four no longer exist, and two had negative cumulative returns. Moats last until they don't, and AI leaders are already disrupted quickly (DeepSeek hitting OpenAI, OpenAI pressuring Google search). Some leaders will stay dominant, but they are priced as a foregone conclusion.

  “But every technological revolution is accompanied by a bubble. ... AI is likely to be no exception.”at 5:31

International equities

• Rob Arnott · Research Affiliates

  Emerging markets value is trading dirt cheap and a better place to move from AI winners.

  Arnott suggests that investors who have made a lot of money on AI should take some chips off the table and move into unnoticed parts of the market. He names emerging markets value and small cap value, both trading very cheaply, with CAPE ratios around 10.

  “But if you make gobs of money on something, why not take some chips off the table and put it on the unnoticed parts of the market? Small cap value, emerging markets value, trading dirt cheap, cape ratios of 10.”at 45:09