00:00Where are we in terms of the AI narrative in your view and what's the reality?
00:05Well, I think investors have to ask a question right now. What am I getting into when I invest
00:09in Microsoft, Google, and Amazon? So UBS estimates that 27% of Google Cloud's revenue this year
00:16will be OpenAI and Anthropic, increasing to over 48% next year. That is a remarkable amount of
00:22money. It's going to be over $124 billion next year. Everyone is buying into these stocks because
00:27they believe all of that CapEx is going towards diverse and spread out AI demand, when in fact
00:33what it's actually doing is helping create infrastructure for two unprofitable, unsustainable
00:38companies. So the other one would be Anthropic, you argue. So give us more data because you're
00:45citing Microsoft, but what about AWS? Well, that was what I was saying. So Barclays actually says
00:49that this year, 13% of AWS revenue will be both OpenAI and Anthropic, and next year will be 18%.
00:56AWS, much bigger business than Google Cloud. Now, just to be clear, when I was saying that 27%
01:01this year and 48% next year for Google Cloud, I meant both Anthropic and OpenAI. Most people don't
01:07know that OpenAI is a large customer of Google Cloud. It's not a well-known fact, but this was
01:14actually mentioned by UBS's Stephen Jew. So where would those companies be right now without Anthropic
01:19and without OpenAI? Well, I have serious questions about that. So in calendar year 2025, according to my own
01:25reporting about OpenAI's numbers, 69% of the year-over-year growth of Microsoft Intelligent Cloud
01:32segment was actually from OpenAI. Without that, it would have only grown 8% year-over-year, which is barely
01:37beating inflation. And so everyone is being sold what I consider kind of a lie. It's honestly kind of a
01:43scandal. So this goes back to, I feel like, we have companies, the circular financing, the circularity of it all,
01:51and kind of creating demand for their products. So when does it start? When does the, I asked this
02:00earlier with the guests, when does the party end in your view? Like how, yeah.
02:04So with OpenAI's IPO, I think that could be one of the flashpoints. Remember, this company was meant
02:08to go public this year. They filed about a month or two ago. And now the New York Times has
02:12reported
02:13that they're considering, they are delaying until 2027. That's lethal for a number of people,
02:18but OpenAI and Anthropic need continual flows of capital. They do not pay their bills out of
02:24existing cash flow. So when anything happens to that cash, I think that's the first kind of domino
02:30to fall. But then again, there's also the overall problem of data centers just not getting built very
02:34fast, taking about 12 to 36 months, depending on how small or large a data center is actually being
02:40built at. And the problem is, is that everyone believes that AI is coming out of cash flow, that AI
02:46is
02:46coming out of just this diverse revenue base when it's really not. It's extremely narrow. The
02:51information reported a few months ago that 89% of the largest AI companies, well, their revenue comes
02:57just from OpenAI and Anthropic. It's heavily centralized.
03:00Doesn't it have to be centralized to some extent? This is expensive to do or no, in terms of data
03:05center build out and so on and so forth. And what's going to make AI, generative AI, the ability for
03:12it to
03:12be really, really good is having access to lots of information. So doesn't it have to be to some
03:17extent, Ed, concentrated? Well, when I say concentration, I mean, concentration of revenue
03:21in these two companies. No, I understand. But to make it good. So doesn't it make sense that those
03:26who are exposed the most, it's going to be concentrated to some extent? Well, I mean, when
03:30we're talking about, so Sightline Climate said that they saw back in February, about 190 gigawatts
03:35worth of data center capacity being built in the next few years. It was built or under planning.
03:40Now, if you work that out with a PUE, so just the efficiency rating of 1.3, you're coming
03:45out to 12 million a megawatt, over $1.6 trillion of annual revenue needed to satiate those data
03:52centers. Having two customers is not going to do that. Even at their most spendy, Anthropic
03:58and OpenAI, well, they can't afford anything. They need venture capital, but they're only going
04:01to spend $400 billion a year. And that's if they get that far, which I don't believe they
04:05will. How much do we know about their balance sheets, really? Really? Well, I'd say from
04:10personal experience, a great deal about OpenAI, because I reported their audited financials
04:14with the Financial Times. Right. And it's a company just burning cash. They lost $20.9
04:19billion in 2025. And things are only getting worse. And what's crazy as well was over $800
04:25million of OpenAI's revenue came from SoftBank for their Cristal Intelligence, and yes, that's
04:30really what it's called, their Cristal Intelligence program, which I can find no evidence of actually
04:36anything happening. And SoftBank, a large shareholder of OpenAI with no board seats.
04:42Oh, go ahead, Carol.
04:43One more question, though. Like, you talk about, for Google Cloud, the exposure, right?
04:49And you said 48% next year in terms of these two customers. I have to say that there are
04:55smart people running these companies. And normally, you would say your exposure to just a handful
05:00of customers is not a great thing. Do you say that these companies that aren't doing their
05:05due diligence, be it Alphabet or, you know, pick your hyperscaler?
05:09I think they did their due diligence in the sense that they said, we are going to create our
05:13largest customers, and we're going to own large parts of them. And on top of that, we're going
05:18to own all of their infrastructure. Google has a nice, they have a nice thing going here.
05:23They buy TPUs from, well, sorry, Broadcom sells TPUs to Google. They are then sold to Anthropic
05:29and then rented back to Anthropic through Google. Google gets to double up on revenue. This sounds
05:34really good right up until you realize that Anthropic and OpenAI are unsustainable. So what
05:40they may be, and the problem is with saying these are smart people, is it immediately makes me think
05:44of Enron, the smartest guys in the room. Not saying anything like that's happening.
05:48But I'm just saying you have a fiduciary responsibility. And you're right. You go back to Enron or
05:53WorldCom. Sure. I think the point I'm making is, with Google, they probably thought there
05:58would be more customers. I imagine with Azure and with AWS, they thought it would be more
06:03large players. But the problem with Anthropic and OpenAI is they've raised $200, $300 billion
06:07of funding, but they've actually raised more because OpenAI and Anthropic got all of their
06:13infrastructure built for them by Microsoft, Google, and Amazon. They didn't have to pay, I think,
06:18in the Sam Altman, Elon Musk trial, one of the Microsoft executives said that they
06:22cost $100 billion. So call it like $70, $80 billion of infrastructure. So the problem is,
06:28is that nobody else can get as big as them. No one else can get that much compute. No one
06:32else
06:32could afford that compute and have the chance to do the pre-training runs necessary. Except now China's
06:37coming up behind them. And it's unclear how anyone really deals with any of the problems I've been
06:43listening for years, which is unsustainable, unprofitable, and also not really finding the ROI
06:48and AI. We're speaking with Ed Zitron, the CEO of Easy Primary Research. He joins us on set in the
06:54Bloomberg Interactive Brokers Studio. Ed, play this out for us because I think a lot of people think,
07:01okay, for there to be some sort of ROI on this, one thing has to happen. And like the best
07:09case scenario
07:09for all this money being spent is that productivity increases, fewer people are needed to do more
07:16things. There are some serious implications if that were to come true and to the labor force.
07:21And Dario Amadei of Anthropic has talked about this in the past. Maybe he's talking his book.
07:25I don't know. The other side of this is, well, if that doesn't come true, then what does it mean
07:30for these stocks that have gained so much on hopes that they would be responsible for some of this
07:37productivity increase? Like how does the shoe drop? What happens? Well, the thing is, if you think
07:41about what Amazon, Google, and Microsoft have done, and met to some extent, but they're not selling
07:45compute capacity yet, is they have gone from being these cash heavy, these cash machines,
07:50they just spill out money, low cash burn, high revenue, low assets, into these bulbous GPU-filled
07:58asset mongers who are just full of these semi-built data centers for two customers or three customers at best
08:05so that they can do what? Rent them out so that they can rent their models. And it isn't really
08:11clear what the plan is at this point. And the problem is, for me to be right, it doesn't even
08:16have to go that badly. OpenAI and Anthropic have to grow so large to be able to make all of
08:23this
08:23data center capacity good. I mean, Google's, I think the UBS estimate was like $76 billion in 2027 of
08:30Google Cloud's revenue will come from Anthropic. How's Anthropic going to afford that? They burn tens of
08:35billions of dollars. So it's not just that these companies are unprofitable and unsustainable,
08:39but they have to grow so very large to make AI pay off because otherwise there just isn't demand
08:46for compute at scale.
Comments