Latest / DX Today | No-Hype Podcast & News About AI & DX / Apollo Economist: AI Paid Off Only for the Magnificent Seven

Transcript

Apollo Economist: AI Paid Off Only for the Magnificent Seven · DX Today | No-Hype Podcast & News About AI & DX

• 0:00
  SPEAKER_00: Welcome to the DX Today Podcast, your daily deep dive into the AI ecosystem.
• 0:05
  SPEAKER_00: I'm Brian, and joining me as always is Jessica.
• 0:08
  SPEAKER_01: Thanks, Brian.
• 0:09
  SPEAKER_01: Today we are digging into a short note from one of the most widely read economists on Wall Street.
• 0:14
  SPEAKER_01: And it lands like a bucket of cold water on the entire AI boom.
• 0:18
  SPEAKER_00: Set it up for us, because I saw the headline over the weekend, and my first thought was that somebody finally said out loud what a lot of operations people have been muttering for months.
• 0:27
  SPEAKER_01: So the economist is Torsten Slock, the chief economist at Apollo Global Management, the giant private equity and credit firm.
• 0:35
  SPEAKER_01: He publishes a daily note called The Daily Spark.
• 0:38
  SPEAKER_01: And on Saturday it was titled No Signs of AI Payoff Outside the Tech Sector.
• 0:43
  SPEAKER_00: That is not a subtle title.
• 0:44
  SPEAKER_00: And I gather the body of the note was not subtle either, which is part of why it spread so fast across finance and tech circles this weekend.
• 0:51
  SPEAKER_01: Not subtle at all.
• 0:53
  SPEAKER_01: His line was, and I am quoting, despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech.
• 1:05
  SPEAKER_01: Zero is a big word.
• 1:07
  SPEAKER_00: Zero is a very big word, and I want to come back to it.
• 1:10
  SPEAKER_00: But first, explain how he is drawing the line between tech and everybody else.
• 1:15
  SPEAKER_00: Because that framing matters a lot here.
• 1:18
  SPEAKER_01: He splits the 500 biggest companies in the main American stock index into two groups.
• 1:23
  SPEAKER_01: One group is the Magnificent Seven, meaning Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla.
• 1:32
  SPEAKER_01: The other group is everyone else, which traders now call the 493.
• 1:36
  SPEAKER_00: So the 493 is basically the real economy.
• 1:39
  SPEAKER_00: Banks, retailers, insurers, manufacturers, airlines, the kind of companies where most people in our audience actually work every day.
• 1:48
  SPEAKER_01: Exactly.
• 1:49
  SPEAKER_01: And he looks at two simple measures for each group.
• 1:52
  SPEAKER_01: The first is revenue per worker, adjusted for inflation.
• 1:55
  SPEAKER_01: The second is profit margins.
• 1:57
  SPEAKER_01: For the seven big tech names, both lines are climbing.
• 2:00
  SPEAKER_01: For the 493, both lines are flat.
• 2:02
  SPEAKER_00: Do we have actual numbers on those charts?
• 2:04
  SPEAKER_00: Or is this mostly about the shape of the lines?
• 2:07
  SPEAKER_00: Because I always want to know how big the gap really is before I get worked up about it.
• 2:12
  SPEAKER_01: The note is mostly charts, so the figures are readings from those charts by other analysts.
• 2:17
  SPEAKER_01: Roughly, the seven bring in something like$780,000 per employee, while the 493 sit around$525,000 in a band they have held for years.
• 2:26
  SPEAKER_00: And margins?
• 2:28
  SPEAKER_00: I would guess the gap there is even more dramatic, given how much pricing power those software and platform businesses have compared with the grocery chain.
• 2:36
  SPEAKER_01: It is dramatic.
• 2:37
  SPEAKER_01: Margins for the 7 run around 30%, while the rest of the index sits near 10%.
• 2:42
  SPEAKER_01: And the key point for Slug is not the gap itself.
• 2:46
  SPEAKER_01: It is that the 493 line has not moved since generative AI arrived.
• 2:50
  SPEAKER_00: He also brought in a productivity measure, right?
• 2:53
  SPEAKER_00: Because margins can be flat for a hundred reasons that have nothing to do with technology at all.
• 2:58
  SPEAKER_01: Right.
• 2:58
  SPEAKER_01: He used a total factor productivity series from the San Francisco Federal Reserve, which tries to measure how much more output we get from the same labor and capital.
• 3:07
  SPEAKER_01: His rate is that there are still no signs of AI boosting productivity growth.
• 3:11
  SPEAKER_00: Okay, so here is where I put on my skeptic hat, but aimed at the skeptic.
• 3:16
  SPEAKER_00: Aggregate data like this is slow and blunt.
• 3:19
  SPEAKER_00: Could the payoff simply be too small and too early to show up in an index of almost 500 companies?
• 3:25
  SPEAKER_01: That is the strongest pushback, and SLOC half anticipates it.
• 3:29
  SPEAKER_01: He argues the effects stay very limited as long as agentic AI is confined to a small corner of the economy.
• 3:36
  SPEAKER_01: So he is not saying it can never pay off.
• 3:38
  SPEAKER_01: He is saying it has not paid off yet.
• 3:40
  SPEAKER_00: That is a much more reasonable claim than zero forever.
• 3:44
  SPEAKER_00: But the headline writers obviously went with zero because zero gets clicks and reasonable nuance does not get anyone to open a newsletter.
• 3:52
  SPEAKER_01: Of course.
• 3:52
  SPEAKER_01: But the part I find most interesting is his explanation of why.
• 3:56
  SPEAKER_01: He wrote that AI may be a great calculator, but in business, intelligence is mostly judgment.
• 4:01
  SPEAKER_00: I like that line, and I think it is mostly right.
• 4:04
  SPEAKER_00: Most of the decisions that actually move revenue at a regional bank are judgment calls about people, risk, and timing, not raw information problems.
• 4:13
  SPEAKER_01: He goes further.
• 4:14
  SPEAKER_01: He says AI does not understand accountability, trust, culture, unwritten rules, ethical trade-offs or fairness, and that those things shape reputational risk and ultimately revenue growth and margins.
• 4:27
  SPEAKER_00: Now I will push back on him a little.
• 4:29
  SPEAKER_00: Plenty of human managers do not understand unwritten rules or fairness either, and their companies still make money.
• 4:36
  SPEAKER_00: The question is whether the tool helps the people who do have judgment.
• 4:40
  SPEAKER_01: Fair.
• 4:41
  SPEAKER_01: And that is where the research gets interesting.
• 4:44
  SPEAKER_01: There is a well-known customer support study where AI assistants raised issues resolved per hour by about 14% overall and by about 34% for the newest, least experienced workers.
• 4:56
  SPEAKER_00: That matches what I have seen in practice.
• 4:58
  SPEAKER_00: The tool helps the new person on the help desk sound like someone with three years of experience.
• 5:04
  SPEAKER_00: It does much less for your best people who already know the answers.
• 5:07
  SPEAKER_01: So the gains are real at the task level.
• 5:10
  SPEAKER_01: The puzzle is why they do not add up to anything you can see in margins.
• 5:14
  SPEAKER_01: And there is a pretty compelling answer from the people who study adoption inside companies.
• 5:18
  SPEAKER_00: Let me guess where this is going, because I have lived it.
• 5:21
  SPEAKER_00: Somebody saves an hour drafting a report and then somebody else spends an hour checking the draft because nobody fully trusts it yet.
• 5:28
  SPEAKER_01: That is exactly it.
• 5:29
  SPEAKER_01: Surveys keep finding that time saved on drafting gets partly eaten by reviewing, correcting, and redoing the output.
• 5:36
  SPEAKER_01: Software teams have test suites to catch bad output, but most industries do not have an automated way to check.
• 5:42
  SPEAKER_00: And that is the Monday morning reality.
• 5:45
  SPEAKER_00: In a claims department or a legal team, the verification step is a human reading every line.
• 5:51
  SPEAKER_00: So you have moved the work around without removing very much of it.
• 5:54
  SPEAKER_01: There is another data point in the note that I think gets overlooked.
• 5:58
  SPEAKER_01: Sluck leans on spending data from RAMP, the corporate card and bill payment company, which tracks AI purchases across tens of thousands of American businesses.
• 6:06
  SPEAKER_00: That is useful because it is actual purchase data instead of a survey where executives tell you how innovative they feel.
• 6:14
  SPEAKER_00: What does the RAMP data show about who is spending the money?
• 6:18
  SPEAKER_01: It shows the spending is wildly concentrated.
• 6:21
  SPEAKER_01: Among the top 1% of AI spenders, roughly 80% are venture-backed companies, compared with only about 13% for a typical spender.
• 6:29
  SPEAKER_01: The heavy users are mostly startups, not the real economy.
• 6:32
  SPEAKER_00: So the people paying top dollar for frontier models are often other tech companies, many of them funded by the same investors who are betting on AI.
• 6:41
  SPEAKER_00: That starts to sound a little circular to me.
• 6:44
  SPEAKER_01: Sluck actually used that word, circularity, earlier last week when describing some venture-backed token spending.
• 6:51
  SPEAKER_01: And he made a related point that mainstream corporate buyers do not need the latest model and are unwilling to pay for it.
• 6:57
  SPEAKER_00: That one rings true.
• 6:59
  SPEAKER_00: If you are a mid-sized insurer, you do not need the smartest model on earth to summarize a phone call.
• 7:04
  SPEAKER_00: You need something cheap, reliable, and approved by your security team.
• 7:08
  SPEAKER_01: Which is a real problem for the economics of the model makers.
• 7:12
  SPEAKER_01: Their pitch to investors assumes that ordinary companies will pay more and more for better intelligence.
• 7:17
  SPEAKER_01: If those buyers treat AI as a cost to minimize, the revenue math changes a lot.
• 7:22
  SPEAKER_00: I keep hearing that the tech leaders inside these companies are feeling the same squeeze.
• 7:27
  SPEAKER_00: Is there any data on what the people actually running the systems are saying about value?
• 7:32
  SPEAKER_01: There is.
• 7:33
  SPEAKER_01: A Gartner survey found that only about 13% of chief information officers report significant value from their AI investments so far.
• 7:41
  SPEAKER_01: That is the people closest to the plumbing, and most of them are still waiting.
• 7:45
  SPEAKER_00: Let me bring in the executive surveys because Apollo is not the only one waving a yellow flag here.
• 7:50
  SPEAKER_00: Haven't we seen chief executives themselves admit the returns are not showing up yet?
• 7:55
  SPEAKER_01: We have.
• 10:27
  SPEAKER_01: Completely agree.
• 10:28
  SPEAKER_01: And that is my main critique of the word zero.
• 10:31
  SPEAKER_01: Averages across hundreds of companies can hide very real gains in specific functions, like customer support or software development, that are being swallowed by costs somewhere else.
• 10:41
  SPEAKER_00: There is also a timing issue that nobody likes to talk about.
• 10:45
  SPEAKER_00: Most big companies spent last year and the early part of this year on pilots, data cleanup, and security reviews, and all of that is pure cost.
• 10:53
  SPEAKER_01: Right.
• 10:54
  SPEAKER_01: And those costs hit the income statement immediately, while any benefit arrives later and spreads out slowly.
• 11:00
  SPEAKER_01: So a flat margin line in the middle of an adoption wave is almost exactly what you would expect to see, even if the eventual payoff is real.
• 11:08
  SPEAKER_00: So, what is the practical takeaway for someone who has to justify an AI budget this quarter?
• 11:12
  SPEAKER_00: Because I suspect a lot of people are going to get forwarded this note by a skeptical boss this week.
• 11:18
  SPEAKER_01: Measure the full cost, including the review and rework time, not just the drafting time.
• 11:23
  SPEAKER_01: Try cheaper or older models before paying for the frontier.
• 11:27
  SPEAKER_01: And tie every project to a business number, like revenue per worker or cycle time, instead of usage counts.
• 11:33
  SPEAKER_00: And I would add one thing: pick a single workflow and redesign it end-to-end around the tool, instead of sprinkling licenses across the whole company and hoping productivity shows up on its own.
• 11:43
  SPEAKER_01: That is the lesson from electricity and from computers.
• 11:47
  SPEAKER_01: The technology was never the bottleneck, the reorganization was.
• 11:51
  SPEAKER_01: The companies that redesign first will be the ones that finally move the 493 line.
• 11:56
  SPEAKER_00: Well said.
• 11:57
  SPEAKER_00: Slock may be early, and zero may be too strong a word, but the burden of proof has clearly shifted from the skeptics to the people selling the transformation.
• 12:06
  SPEAKER_00: That's all for today's episode of the DX Today podcast.
• 12:10
  SPEAKER_00: Thanks for listening, and we'll see you next time.