Latest / AI Newspaper / Apollo's Chief Economist Says AI Has Delivered Zero Financial Payoff Outside Big Tech

Transcript

Apollo's Chief Economist Says AI Has Delivered Zero Financial Payoff Outside Big Tech · AI Newspaper

• 0:00
  Welcome to the AI Newspaper, your daily update
• 0:02
  on everything happening with AI. Visit AINewspaper
• 0:06
  .org to see today's stories and sign up for email
• 0:09
  alerts. So imagine spending like $650 billion
• 0:15
  on just the most advanced factory the world has
• 0:18
  ever seen, right? And then realizing nobody actually
• 0:22
  knows what to build in it yet. Yeah, I mean,
• 0:25
  it sounds completely absurd when you put it like
• 0:26
  that. Right, but that is exactly what is unfolding
• 0:29
  in the tech sector right now. I mean, we've got
• 0:31
  companies like Microsoft, Alphabet, Amazon reporting
• 0:34
  earnings this week, and the spotlight is just
• 0:37
  glaring on this historic flow of capital. Big
• 0:41
  tech is projected to spend roughly $650 billion
• 0:44
  on the AI build out this year alone. Which, just
• 0:49
  to pause on that number. is larger than the entire
• 0:52
  GDP of countries like Sweden or Argentina. It
• 0:56
  is just a staggering amount of money. Okay, let's
• 0:58
  unpack this because the true test of this boom
• 1:01
  is definitely shifting. It's no longer just about
• 1:05
  what these tech giants are spending to build
• 1:07
  the infrastructure. Exactly. The narrative is
• 1:10
  moving. Right. The real question, the one that
• 1:12
  will... ultimately define the next phase of the
• 1:14
  global economy is whether the other 493 companies
• 1:18
  in the S &P 500 are seeing any financial return
• 1:21
  on all this technology. And looking at the macroeconomic
• 1:25
  data, their profit margins and their revenue
• 1:28
  per worker are just completely flat. Completely
• 1:30
  flat. And it's this massive disconnect that we
• 1:33
  really have to reconcile. We are tracking this
• 1:35
  unprecedented flow of capital into specialized
• 1:39
  chips, massive data centers. Foundational models.
• 1:43
  But outside of the companies actually selling
• 1:45
  the shovels in this gold rush, the financial
• 1:47
  payoff is conspicuously absent. The builders
• 1:51
  are getting rich, but the buyers are basically
• 1:53
  just treading water. Right, exactly. I mean,
• 1:56
  we are looking at a definitive tale of two markets
• 1:59
  here. On one hand, you have the magnificent seven,
• 2:02
  you know, Apple, Meta, Nvidia, Microsoft. And
• 2:05
  for them, the metrics are glowing. Revenue per
• 2:08
  worker is climbing. Profit margins are up. Because
• 2:11
  they are successfully absorbing this boom directly
• 2:14
  into their bottom lines. Yeah. But then you strip
• 2:16
  those seven out of the S &P 500, leaving you
• 2:19
  with the remaining 493. The mainstream corporate
• 2:23
  behemoths, retail giants, global banks, legacy
• 2:27
  manufacturers. For them, those exact same metrics
• 2:30
  haven't budged an inch. Yeah. And it leads to
• 2:33
  a pretty blunt reality that was recently highlighted
• 2:37
  in a major macroeconomic analysis from Apollo
• 2:39
  Global Management. When you look at this through
• 2:41
  a purely economic lens, despite the massive investments
• 2:45
  and the deafening hype around AI, the technology
• 2:48
  has delivered zero financial payoff to corporate
• 2:50
  America outside the tech sector so far. Zero.
• 2:54
  That's, I mean, that's a heavy conclusion. The
• 2:57
  capital is leaving their bank accounts, but the
• 3:00
  surplus profit is just not coming back in. I
• 3:02
  do have to push back a little on that terminal
• 3:04
  framing, though, because zero payoff sounds so
• 3:07
  final. Are we absolutely sure these 493 companies
• 3:11
  aren't just lagging in a normal adoption cycle?
• 3:14
  How do you mean? Well, it feels a bit like a
• 3:17
  restaurant buying a massive state -of -the -art
• 3:19
  kitchen, right? If you look at their balance
• 3:21
  sheet the next day, of course the financial payoff
• 3:24
  is zero. The output isn't going to double on
• 3:26
  day one. The chefs have to actually learn the
• 3:28
  new recipes. Yeah, that makes sense. Management
• 3:31
  has to figure out the workflow. You have to train
• 3:33
  the front -of -house staff. Doesn't mainstream
• 3:35
  corporate America just need a few years to learn
• 3:38
  how to cook with this new equipment? What's fascinating
• 3:40
  here is that the barrier isn't just about learning
• 3:43
  how to turn on the new ovens. It's about the
• 3:46
  fundamental nature of what the equipment actually
• 3:48
  does versus what drives business value for these
• 3:51
  legacy companies. Okay, break that down for me.
• 3:54
  The analysis makes this critical distinction
• 3:56
  that frequently gets lost in all the tech enthusiasm.
• 3:59
  AI as it exists today. is essentially an absolutely
• 4:03
  incredible world -class calculator. It processes
• 4:06
  and predicts information at speeds we can't comprehend.
• 4:10
  But in mainstream business, especially the S
• 4:12
  &P 493, intelligence isn't just calculation.
• 4:16
  It is mostly human judgment. Oh, I see. So we're
• 4:20
  talking about context, nuance, lived experience,
• 4:24
  the things you can't just easily put into a spreadsheet.
• 4:26
  Precisely. A statistical model doesn't understand
• 4:29
  accountability. It doesn't get trust or corporate
• 4:32
  culture or the unwritten rules of an industry.
• 4:35
  It certainly has no grasp of ethical tradeoffs
• 4:37
  or human empathy. Right, which is huge for like
• 4:40
  a health care provider. Or a major bank. Those
• 4:43
  deeply human elements shape their reputational
• 4:45
  risk. They dictate long term strategy and they
• 4:48
  ultimately drive real revenue growth. You can't
• 4:51
  just plug a brilliant calculator into a complex
• 4:53
  organization and expect it to automatically negotiate
• 4:56
  a better vendor contract or navigate a PR crisis.
• 4:59
  Or look a client in the eye and build generational
• 5:02
  trust. Exactly. Which implies that turning cheap
• 5:05
  automated intelligence into actual revenue growth
• 5:08
  is profoundly harder than it sounds in a software
• 5:11
  pitch. Much more difficult. Because judgment
• 5:14
  depends entirely on context. And the correct
• 5:16
  answer often varies depending on which human
• 5:19
  you were dealing with on a given day. A tech
• 5:21
  company might see massive gains using AI coding
• 5:23
  assistance for... Spoiler play code. But a traditional
• 5:26
  business dealing in human services operates in
• 5:29
  a really messy reality. Right. Okay. But even
• 5:32
  if these legacy companies can't easily replace
• 5:35
  human judgment to boost profits, shouldn't they
• 5:37
  at least be getting their daily tasks done faster?
• 5:40
  Like, does the data show them getting more efficient?
• 5:44
  Because there is a massive ticking clock hovering
• 5:47
  over this entire situation. The big tech companies
• 5:50
  are spending that $650 billion this year. And
• 5:54
  that infrastructure buildout is increasingly
• 5:56
  fueled by heavy credit. The debt accumulation
• 5:58
  in the infrastructure layer is staggering right
• 6:01
  now. We are seeing real strain on some of the
• 6:04
  newer pure play infrastructure companies. There's
• 6:06
  this heavily backed loss making cloud company
• 6:09
  focused entirely on AI infrastructure that has
• 6:12
  racked up about $50 billion in debt. 50 billion.
• 6:16
  That's wild. That is wild. It is. And investor
• 6:19
  concerns are rapidly growing that this entire
• 6:21
  data center boom relies way too heavily on credit
• 6:24
  financing, which puts immense pressure on cash
• 6:28
  flows. Yeah, and we should clarify what an impairment
• 6:31
  actually looks like on a balance sheet. It means
• 6:34
  the physical assets, the sprawling data centers,
• 6:37
  the specialized chips, are officially recognized
• 6:39
  by accountants as being worth fundamentally less
• 6:42
  than what the company paid for them. Because
• 6:45
  they aren't generating cash flows. Exactly. If
• 6:47
  you spend $100 million expecting $20 million
• 6:49
  a year and you only get $2 million, you have
• 6:52
  to write down that loss. Exactly. To avoid a
• 6:55
  catastrophic $3 .5 trillion write -down across
• 6:58
  the industry, we have to see AI making the broader
• 7:01
  economy genuinely more efficient quickly. So
• 7:05
  let's dig into that efficiency data. You hear
• 7:08
  executives on earnings calls talking all the
• 7:10
  time about how AI is boosting internal productivity.
• 7:14
  If profit margins aren't up, are workers at least
• 7:16
  producing more? This is where we need to introduce
• 7:19
  a crucial macroeconomic distinction, one that
• 7:23
  economists rely on to separate genuine innovation
• 7:25
  from a company simply spending a lot of cash.
• 7:28
  We have to separate labor productivity from total
• 7:32
  factor productivity, or TFP. OK, I want to make
• 7:35
  sure this is crystal clear for everyone listening,
• 7:37
  because these two terms sound identical, but
• 7:40
  they tell entirely different stories about where
• 7:42
  the economy is heading. Very different. Labor
• 7:45
  productivity is the simpler of the two. It is
• 7:47
  just output per hour. How much stuff gets done
• 7:50
  by a worker in a given amount of time. OK, got
• 7:52
  it. And TFP? Total factor productivity, TFP,
• 7:56
  is much deeper. It measures genuine efficiency
• 7:59
  and technological progress. It measures how much
• 8:01
  output the entire economy produces from a given
• 8:04
  quantity of labor and capital combined. TFP only
• 8:07
  rises when firms figure out how to produce more
• 8:09
  without adding inputs or spending more money.
• 8:12
  So it's basically the closest proxy we have for
• 8:15
  true innovation. Think about your own office
• 8:18
  environment for a second to ground this. Say
• 8:20
  I manage a team and want to increase output per
• 8:23
  hour. I could just buy every single employee
• 8:26
  a second larger computer monitor, right? Right,
• 8:28
  exactly. Suddenly they view more spreadsheets
• 8:30
  at once. They tab around less. They type faster.
• 8:33
  Their output per hour goes up. That is labor
• 8:37
  productivity rising. But buying a second monitor
• 8:40
  didn't make my company any smarter. It didn't
• 8:42
  improve our actual product. I just threw cash
• 8:45
  at hardware. If we connect this to the bigger
• 8:47
  picture, that exact phenomenon of buying second
• 8:51
  monitors is what economists call capital deepening.
• 8:54
  It means giving workers more equipment to do
• 8:57
  the same job. And mechanically, that is what
• 9:00
  is happening right now with the AI boom. Just
• 9:03
  pouring money in. Hundreds of billions flowing
• 9:06
  into chips and models, which mechanically lifts
• 9:09
  output per hour. Current data shows U .S. labor
• 9:12
  productivity is running strong at about 2 .5%,
• 9:15
  comfortably above its post -2005 average. So
• 9:19
  a CEO can point to that 2 .5 % jump on an earnings
• 9:22
  call and claim their AI investment is an absolute
• 9:25
  triumph. But wait, I have to push back here on
• 9:28
  how we measure genuine innovation. Is TFP even
• 9:31
  the right metric to capture what AI does? Go
• 9:35
  on. Let's say AI improves a wealth manager's
• 9:38
  advice. They use it to analyze a client's portfolio
• 9:40
  deeper, resulting in better yields. But they
• 9:43
  still only manage the exact same number of clients
• 9:45
  per day. That wealth manager's output hasn't
• 9:48
  increased in quantity. TFP wouldn't capture that
• 9:52
  massive increase in quality, would it? That is
• 9:54
  an excellent pushback. And you are right that
• 9:57
  TFP struggles to measure immediate, invisible
• 10:00
  quality improvements. But if a wealth manager
• 10:03
  offers vastly superior advice, eventually that
• 10:07
  firm should be able to charge higher fees, capture
• 10:10
  more market share, and increase revenue without
• 10:12
  hiring more managers. Which would eventually
• 10:15
  show up as a rise in TFP. Right. But right now.
• 10:18
  The San Francisco Fed tracks a utilization -adjusted
• 10:21
  series for TFP, and it is sitting slightly below
• 10:24
  zero. There's absolutely no sign of acceleration
• 10:27
  since this massive capital expenditure cycle
• 10:30
  began. Below zero? Mm -hmm. Meaning, despite
• 10:35
  the $650 billion rollout, genuine efficiency
• 10:38
  is flat, or technically negative across the broader
• 10:41
  economy. Correct. Strong output per hour alongside
• 10:45
  flat or negative TFP is the classic signature
• 10:47
  of capital deepening, not a technology shock.
• 10:51
  We are just buying more stuff. We're installing
• 10:53
  more servers. The productivity payoff from AI
• 10:56
  remains a forecast, not an observation. So if
• 10:59
  TFP isn't rising... and profit margins for the
• 11:03
  493 are stubbornly flat, we need to look at what
• 11:07
  the ground reality of AI spending actually looks
• 11:09
  like inside these legacy companies, because they
• 11:11
  are definitely swiping their credit cards for
• 11:13
  it. Oh, they absolutely are. Recent corporate
• 11:16
  card data from the Ramp AI Index gives us a very
• 11:19
  granular look at everyday businesses. Adoption
• 11:23
  is undeniably growing. As of September, 55 .9
• 11:26
  % of U .S. businesses on their platform were
• 11:29
  paying for some form of AI. That's up 12 .1 percentage
• 11:32
  points from last year. Over half of businesses
• 11:35
  are paying for it, which sounds like a massive
• 11:37
  victory. But if you look closely at how much
• 11:39
  they are paying, it's incredibly revealing. The
• 11:42
  median spend among firms paying for AI was a
• 11:45
  mere $15 .70 per employee. Which, if you think
• 11:49
  about current pricing, is essentially the cost
• 11:51
  of a single, basic, consumer -tier chatbot subscription,
• 11:54
  likely only provided to a fraction of the staff.
• 11:58
  Think about your own workplace again. Your executive
• 12:00
  team might talk a big game about an AI revolution
• 12:03
  in an all -hands meeting, but look at the software
• 12:06
  you're actually given access to. Chances are
• 12:09
  they're only spending about 15 bucks a head on
• 12:11
  you. Right. But then you contrast that with the
• 12:14
  top 1 % of adopters in this data set. That tiny
• 12:18
  sliver of companies is spending a median of over
• 12:20
  $9 ,500 per employee. You've got a microscopic
• 12:24
  fraction going all in, and mainstream corporate
• 12:27
  America just dipping their toe in the shallow
• 12:29
  end. And this hesitation from the mainstream
• 12:31
  market is starting to heavily impact the economics
• 12:34
  of the technology itself. The blended price for
• 12:37
  AI tokens is plummeting right now. And for context,
• 12:40
  a token is the fundamental unit of computation,
• 12:42
  right? It's roughly equivalent to a piece of
• 12:45
  a word. Exactly. Across major AI providers, the
• 12:48
  price for processing these tokens fell 25 .4
• 12:51
  % to just 64 cents per million tokens in recent
• 12:55
  weeks. Here's where it gets really interesting.
• 12:58
  Interesting. Why is a price drop a bad signal?
• 13:01
  Because you pair that plummeting price with the
• 13:03
  fact that mainstream corporate America is only
• 13:05
  spending 15 bucks an employee. Venture -backed
• 13:08
  startups might be spending millions, but mainstream
• 13:10
  America does not seem to want or need the most
• 13:13
  advanced premium models. And more importantly,
• 13:17
  they are completely unwilling to pay a premium
• 13:19
  for them. This forces providers into a brutal
• 13:22
  price war just to stimulate demand. They are
• 13:25
  pulling back because the math doesn't work. Right.
• 13:27
  When you have no visible impact on your bottom
• 13:29
  line, you stop viewing token spending as a magical
• 13:33
  engine for limitless growth. You start viewing
• 13:35
  it as a traditional cost center. Just another
• 13:38
  IT expense that weighs on your margins. It's
• 13:41
  the classic gap between technological expectation
• 13:44
  and corporate reality. The expectation was that
• 13:47
  AI would sweep through the S &P 493, automate
• 13:51
  massive amounts of knowledge work and send profit
• 13:53
  margins into the stratosphere. But let's look
• 13:56
  at how a massive legacy institution is actually
• 13:59
  integrating this today. We have a real world
• 14:01
  example with JPMorgan Chase. JPMorgan Chase is
• 14:04
  the perfect avatar for the 493 trying to navigate
• 14:07
  this transition. Their EMEA private bank covering
• 14:10
  Europe. the Middle East and Africa, is actively
• 14:13
  integrating advanced AI tools right now. They're
• 14:17
  using it to streamline onboarding and for conducting
• 14:19
  initial financial research. On paper, you'd think
• 14:22
  this is exactly the kind of deployment that leads
• 14:24
  to massive efficiency layoffs. If the AI does
• 14:27
  the research in five seconds, you ostensibly
• 14:29
  need fewer human researchers. Right. The dreaded
• 14:32
  AI is coming for your white -collar job scenario.
• 14:36
  Exactly. But the data shows the exact opposite
• 14:38
  is happening. The regional chief of that private
• 14:41
  bank recently stated they actually expect their
• 14:43
  advisor headcount to grow by 10 to 15 percent
• 14:46
  over the next few years. They are hiring alongside
• 14:49
  the AI integration. They aren't replacing the
• 14:52
  human judgment. They're augmenting it. Exactly.
• 14:55
  Which means overall labor costs are still going
• 14:57
  up, not down. It brings us right back to our
• 15:00
  restaurant kitchen analogy. They didn't fire
• 15:02
  the chefs when they bought the new ovens. The
• 15:04
  ovens processed orders faster, which meant they
• 15:06
  had to go out and hire more waiters to manage
• 15:08
  the volume. Because AI can summarize a 100 -page
• 15:11
  report instantly, but it still takes human judgment
• 15:14
  to sit across a mahogany table from a client,
• 15:16
  navigate complex family dynamics, and close a
• 15:19
  multimillion -dollar deal. The AI does not understand
• 15:22
  the unwritten rules of wealth management. It
• 15:24
  really doesn't. Which perfectly explains why
• 15:27
  the macroeconomic data looks the way it does.
• 15:29
  Executives constantly report these isolated internal
• 15:32
  victories. A faster onboarding process here,
• 15:35
  better email drafting there. But they fundamentally
• 15:37
  struggle to translate those microefficiencies
• 15:40
  into a measurable financial return. Yeah, a recent
• 15:43
  survey highlighted this perfectly. Half of UK
• 15:46
  chief executives ranked AI tools among their
• 15:49
  top internal productivity drivers. But when pressed?
• 15:52
  only 14 % of them actually had full real -time
• 15:56
  visibility into what their AI infrastructure
• 15:58
  costs even were. So you have executives paying
• 16:02
  for expensive new software. They are still paying
• 16:05
  for a growing human workforce. Their basic output
• 16:08
  per hour ticks up slightly, but their genuine
• 16:10
  efficiency, their TFP, remains completely flat.
• 16:13
  It is a textbook recipe for margin compression.
• 16:17
  So what does this actually mean for you as you
• 16:19
  watch this narrative unfold? It means we need
• 16:22
  to radically shift where we are looking for proof
• 16:25
  of a revolution. The true story of this technological
• 16:28
  era isn't going to be written by the seven tech
• 16:30
  giants building the data centers. We know they
• 16:32
  are making money selling infrastructure. Right.
• 16:34
  The builders are fine. The ultimate fate of this
• 16:37
  $650 billion buildout will be decided by the
• 16:41
  493 of their companies, your banks, retailers,
• 16:45
  logistics networks, maybe even your own employer,
• 16:47
  as they grind through the slow, messy reality
• 16:50
  of trying to figure out how to actually generate
• 16:53
  surplus revenue using these tools. This raises
• 16:56
  an important question, one that requires us to
• 16:58
  zoom out and look at the broader history of technological
• 17:00
  revolutions. Total factor productivity has swung
• 17:03
  wildly between negative 3 % and positive 4 %
• 17:07
  over the past 40 years. It's highly cyclical.
• 17:09
  Okay, so it bounces around. But history shows
• 17:12
  us that truly transformative technologies, like
• 17:15
  the rollout of electricity or the enterprise
• 17:18
  IT revolution in the 90s, did not show up in
• 17:20
  aggregate productivity data immediately. It took
• 17:24
  a decade or more for the global economy to physically
• 17:26
  and culturally reorganize itself around those
• 17:28
  innovations. Right, it takes time. Are we fundamentally
• 17:31
  misjudging AI's value by demanding quarterly
• 17:34
  financial payoffs today? And perhaps more provocatively,
• 17:37
  when that payoff finally arrives a decade from
• 17:40
  now, will the ultimate return on investment for
• 17:43
  these 493 companies simply be corporate survival
• 17:45
  rather than the generation of massive surplus
• 17:48
  profits? Thanks for listening to the AI Newspaper.
• 17:51
  Visit AINewspaper .org to catch up on the latest
• 17:54
  updates and sign up for free email alerts.