Aswath Damodaran: Big Tech Is Spending Trillions on AI. Can It Pay Off?

The AI Factory Analogy [00:07:54]

• The collective investment in AI infrastructure is compared to building the world's largest factory, costing over $2 trillion since late 2022.
• This investment dwarfs historical expenditures like the railroad boom or dot-com investments.
• Companies like NVIDIA profit from selling the chips (building the factory), while others like Meta and Alphabet aim to use the AI infrastructure for products and services (using the factory).

  "So there's this architecture companies that make their money in building the factory. But included in your mix are those companies that are spending the money on this architecture because they want to make products and services."

Revenue vs. Investment Discrepancy [00:11:05]

• Current collective revenues from AI products and services are estimated at $250 billion.
• This revenue is significantly lower than the $2 trillion invested in the AI factory.
• If revenues stagnate, the $2 trillion investment could be largely written off.

  "You have $250 billion in revenues. Let's play the worst case scenario. If revenues level off at $250 billion, that $2 trillion is almost entirely going to be written off."

Justifying the Investment: Future Revenue Needs [00:12:17]

• To justify the $2 trillion upfront investment, collective AI revenues in a steady state might need to reach $8-10 trillion.
• This requires significant profit margins and sustained growth.
• The core question is whether the AI product and service market can realistically generate such high revenues.

  "The revenues will have to be maybe $8 trillion, $10 trillion, because if you think about the profits and the expenses to produce the revenues and the profits and the tax and the profits, you're very quickly going to start to scale the ladder up to $10 trillion."

The 3P Test: Possibility, Plausibility, Probability [00:13:48]

• Achieving $10 trillion in AI revenues is currently considered possible but likely pushes the limits of plausibility.
• The probability of reaching this revenue target is assessed as low at present.
• It's crucial not to label the situation as a bubble without sufficient data, as the future potential is still unknown.

  "Right now, it's possible. You can have a $10 trillion revenue, and I'll explain why that possibility exists. Is it plausible? You're probably pushing the limits of what's plausible with $10 trillion."

Contamination from Existing Revenue Streams [00:14:37]

• Profitable, durable revenue streams from companies like Alphabet (advertising) and Microsoft should not be conflated with AI-specific revenues.
• These existing businesses provide cash flow to fund AI investments but are separate from the AI factory's output.
• Including them contaminates the analysis of AI's true potential.

  "And those should not be part of the AI discussion. Those should not, they are ways in which these companies are coming up with the cash flows to fund them."

Risk Exposure: Diversified vs. Concentrated Companies [00:16:43]

• Companies with strong existing businesses (e.g., Alphabet, Meta) have more contained damage if AI investments fail, as losses would impact shareholders via reduced dividends or buybacks.
• Companies like CoreWeave, with significant debt and fewer diversified revenue streams, are more exposed to potential write-downs of AI investments.
• The risk of AI failure is higher for companies solely reliant on AI for future growth.

  "But the companies that don't have the side businesses are going to be much more exposed if or when there's a write-off of the AI investment."

Investor Choices: Ride the Wave or Retreat? [00:18:40]

• Investors face a choice: pull money from AI-related stocks (mostly Mag-7) or stay invested.
• Pulling out might mitigate losses if a correction occurs but risks missing potential gains during the interim.
• Market timing is a perennial challenge, and historically, staying out of the market for too long after a correction can be detrimental.

  "The good news for you then is when the correction comes, you're going to be hurt less. But the bad news is it could take three years for the correction to happen and what you lose while you wait might vastly exceed what you benefit."

The Long-Term Horizon and Retirement [00:20:40]

• For long-term investors (e.g., 35-year-olds), even a potential AI bubble might not significantly alter their investment trajectory.
• Investors nearing retirement should consider de-risking their portfolios.
• Bonds currently offer attractive yields (5.5%), providing a safer alternative for those closer to needing their funds.

  "But if you're 35, I don't think it should alter the trajectory of your investing in any significant way, even if you believe there's an AI bubble."

Passing the Sleep Test: Portfolio Adjustments [00:21:58]

• Investors should adjust their portfolios to pass the "sleep test" – not lying awake worrying about their investments.
• Shifting from market-cap-weighted to equal-weighted indexes or factor tilts can reduce concentration risk.
• While not a major course correction, these adjustments can improve well-being.

  "And sleep test is if you lie awake at night wondering what your portfolio is doing, you fail the sleep test."

Personal Investment Strategies: Buying Low, Selling High [00:24:00]

• Professor Damodaran bought mega-cap stocks like Microsoft, Meta, and Tesla at different times when they appeared undervalued, not initially for AI potential.
• He emphasizes that even expensive companies like the Mag-7 have had periods of being cheap.
• He sold Tesla due to political entanglements and NVIDIA reluctantly after massive gains, as it was priced as the "greatest company ever."

  "I bought Microsoft when Satya Nadella became CEO in 2013 to 2014. I bought Facebook after the half of it after the fiasco they had in 2017 with Cambridge Analytica and the other half after the metaverse fiasco..."

The Role of Taxes in Selling Decisions [00:27:37]

• Taxes, especially capital gains taxes, can significantly influence selling decisions, particularly for large winners.
• Damodaran, living in California, faces high capital gains taxes (around 26-27%), making him reluctant to sell unless a stock becomes significantly overvalued.
• This tax consideration means he might leave money on the table but avoids the emotional distress of selling too early.

  "Taxes especially I think contaminate the investment process because they can affect when you sell. Because when you sell you have to pay taxes and that's going to be larger than your biggest winners."

Portfolio Concentration Limits and Risk Management [00:29:10]

• Damodaran imposes a strict rule: no single investment can exceed 15% of his portfolio.
• This rule is on autopilot, automatically triggering sales once a stock hits the limit.
• This prevents any single company, even a massive winner, from becoming an outsized risk.

  "I will not let an investment get above 15% of my overall portfolio. It's on autopilot I sell it once it hit 15% which means none of these stocks are in double digit levels now on my portfolio."

Staged Selling and Emotional Discipline [00:31:43]

• For investors with highly concentrated positions (like 70% in one stock), a staged selling approach is recommended.
• Selling in portions (e.g., a quarter every six months) removes the emotional burden of timing the sale.
• This disciplined approach prevents holding onto winners too long due to emotional attachment.

  "Do it in stages: sell a quarter of your Tesla stock you know put it on autopilot every six months I'm going to sell a quarter of my Tesla stock no matter what the price is because if you have to think through whether this is the right time to sell you will find a reason not to sell."

The Circularity of AI Investments [00:37:22]

• A concern is that many AI companies are leasing data center space from each other, creating circular revenue streams.
• For example, Alphabet investing in Anthropic, which then leases data center space back from Alphabet.
• This circularity makes it difficult to assess genuine external demand for AI products and services.

  "Alphabet has invested in Anthropic and then Anthropic is using those dollars or some portion of them to then buy lease you know data center back from Alphabet and that's going on all over this and that's where I have a lot of trouble building up that valuation of Alphabet and saying what's real here in terms of external demand..."

AI as a Tool vs. Replacement: Revenue Potential [00:41:05]

• If AI primarily acts as a tool to enhance productivity (e.g., for consultants), its revenue potential is limited as a cost to businesses.
• If AI replaces employees entirely, it could theoretically generate revenues up to $26 trillion (global salaries), but this scenario creates a macroeconomic paradox: displaced workers have no income to consume products.
• Realistic total addressable markets for AI are likely between $2-10 trillion, depending on whether it's a tool or a workforce replacement.

  "So let's say AI's pipe dream is it can replace every employee at every company I'll tell you in a minute why this is going to be a nightmare for the rest of the world right but potentially revenues could be up to 26 trillion."

The "Big Market Delusion" and Overinvestment [00:47:20]

• When a large potential market emerges, overconfident companies tend to overinvest collectively.
• Each company believes it will be a winner, leading to a situation where the total investment might exceed the market's capacity.
• This overconfidence can result in investors holding a portfolio of companies where only one or two succeed, potentially leading to overall losses.

  "I call this the big market delusion. It happens every time there's a big market and you have overconfident businesses looking at that big market collectively they over invest."

Index Funds vs. Individual Stock Picking [00:53:40]

• The increasing uncertainty and complexity in the AI space strengthen the argument for index fund investing.
• Index funds offer diversification and allow investors to participate in market growth without the burden of individual stock analysis.
• Historically, investors who avoided the Mag-7 due to perceived high valuations have underperformed.

  "I mean if you think there was an argument for index funds before all of this craziness I think the argument just got stronger rather than weaker for index funds because the more uncertain you feel about how processes play out and winners and losers the better off you are letting your money ride with a bunch of all companies in market hoping that you catch some of them."

Diversifying with Multiple Index Funds [00:56:03]

• Instead of relying on a single index fund, a diversified portfolio can be built using various index funds (e.g., S&P 500, small-cap, emerging markets).
• This approach provides broad market coverage and allows for customization based on investor goals.
• The key is to be comfortable with matching market returns rather than trying to significantly outperform.

  "The advantage of index funds and ETFs is you can create as diversified a portfolio as you want to with no upfront flotation transaction costs."

Reverse Discounted Cash Flow and Revenue Focus [01:00:30]

• A reverse discount cash flow model can be used to estimate the revenues required to justify current market valuations.
• Framing the analysis around revenues, rather than complex cash flow metrics, facilitates broader conversation between finance and tech professionals.
• This approach helps identify companies that might still be viable investments even if the overall AI sector is overvalued.

  "My suggestion I would make is reframe this not in terms of free cash flow but in terms of revenues because I think that's really what this big debate is what will the revenues look like in this business and are they large enough to justify what we're investing in what we're building right now."

The Societal and Economic Consequences of AI [01:11:20]

• Widespread AI adoption, particularly if it leads to mass job displacement, could have severe macroeconomic consequences, including a global depression.
• The optimistic AI narrative often overlooks the dystopian outcomes for society, such as increased inequality and unemployment.
• Companies pushing AI need to consider the broader societal impact and develop a more positive narrative, akin to AT&T's investment in Bell Labs for social goodwill.

  "So I think this is the problem you let 25 to 30 year olds without adult supervision build trillion dollar companies without somebody pushing back and saying this makes no sense."

AI's Impact on Learning and Daily Life [01:21:40]

• The long-term effects of AI on learning processes and daily human interaction are a cause for concern.
• Past technological shifts (PCs, internet, social media) have had unforeseen negative consequences.
• There's a worry that AI might diminish human capabilities and lead to a net negative outcome for society.

  "I look at my grandchildren I wonder what the world will look like and how they will learn a world full of ai I am not particularly happy about the kinds of things ai will do to their learning processes..."

Personal Portfolio Bets and Risk Tolerance [01:26:50]

• Mindy Jensen acknowledges her significant concentration in SpaceX (43% of her portfolio) and expresses discomfort with it.
• She plans to discuss with her husband, Carl, about carving off a portion of the position for living expenses and experimenting with the rest.
• The hosts discuss the difficulty of divesting from highly successful, high-conviction investments.

  "I pulled up our spreadsheet and SpaceX specifically is about 43 I am starting to get a little uncomfortable with that much in one stock and it has skewed all the rest of the percentages because so much is in that one stock."

AI as a User: Portability and Price Sensitivity [01:31:25]

• A power user of AI tools observes continuous model improvements (Grok, GPT-4, Claude 3) and increasing monthly costs.
• The strategy is to build workflows that are portable to the next generation of AI models, avoiding dependence on any single provider.
• Price sensitivity and usage will likely drive adoption and competition in the AI market.

  "I'm building everything I'm I'm working on I build it to be portable because I'm betting on the next AI model coming from out of nowhere for the next provider maybe grok leaps rocks it in a few months maybe gemini does in a couple of months maybe perplexity or maybe all these guys get arrows in their backs as is common in technology pioneering and a new one that doesn't exist yet takes over in two years."