Latest / Deep Values: Value Investing Deep Dives / AI Supercycle Showdown: Is Trillion-Dollar AI Infrastructure Fraud?
Transcript
- 0:00Well, welcome to the debate. Today we're going to dive into,
- 0:03well, the pretty extraordinary financial engineering that's
- 0:06really driving the artificial intelligence sector right now.
- 0:10We're looking at these massive trillion dollar commitments
- 0:13defining the current capital expenditure boom.
- 0:16Yeah, the source material paints a picture of an ecosystem built
- 0:20not just on, you know, high valuations, but on these huge
- 0:24super contracts, deep financial dependencies and frankly
- 0:27unprecedented leverage. So that brings us to our main
- 0:32question today. Is this system fundamentally a,
- 0:36well, a necessary rational investment in what you could
- 0:39call a foundational super cycle? Or is it maybe better diagnosed
- 0:45as a debt fueled malinvestment super cycle, one that's just
- 0:49loaded with systemic fragility? My view is that these contracts,
- 0:53they're a necessary strategic move.
- 0:55We need them to rapidly build this foundational cognitive
- 1:01infrastructure, essentially securing the compute resources
- 1:04for what is, I think we agree, a genuinely transformative
- 1:08technology. And I, I take the opposing view
- 1:12there. I think what we're really
- 1:14witnessing is a structure that's defined by, well, financial
- 1:17circularity, extreme leverage, and frankly, manufactured
- 1:21demand. And all of that creates severe
- 1:24systemic risk that looks much more like a speculative bubble,
- 1:27a classic malinvestment cycle, than any kind of sustainable
- 1:31organic tech growth. OK, well the premise of the
- 1:35Super cycle for me is pretty simple the the sheer scale of
- 1:40compute demand coming from these next generation foundational AI
- 1:45models. It just necessitates these kinds
- 1:48of super contracts. This isn't your normal
- 1:50incremental CapEx. It's a it's a global building
- 1:54phase. We're establishing a new utility
- 1:57kind of like, you know, building out the railroads or the early
- 1:59Internet backbone. Just look at the deal flow.
- 2:02We're saying open AI is a landmark $300 billion five year
- 2:07commitment to Oracle. That seems like a powerful
- 2:09necessity to me, representing A strategic break from its its
- 2:14previous cloud exclusivity with Microsoft.
- 2:16And that move, it was clearly driven by the urgent need to
- 2:21overcome these critical compute bottlenecks that were, you know,
- 2:24actively holding back product development and deployment.
- 2:27And crucially, this investment isn't just speculative paper
- 2:31shuffling, it's building tangible physical
- 2:34infrastructure, real data centers, high speed networks,
- 2:37massive high value GPU clusters. These assets, they hold
- 2:42intrinsic long term economic value, which I think really
- 2:45differentiates this cycle from AST bubbles that were built
- 2:48mostly on, well, ephemeral software only valuations.
- 2:52Plus the fact that Open AI is diversifying now, engaging both
- 2:56NVIDIA and AMD for huge GPU supply, that just demonstrates A
- 3:00rational business strategy aimed squarely at mitigating supply
- 3:04chain concentration risk. It's strategic foresight and
- 3:07action. OK, I I hear that and I do
- 3:11acknowledge the underlying technology is transformative,
- 3:14but the financial reality of this architecture, the way it's
- 3:20built, it suggests profound malinvestment to me.
- 3:24I mean, the capital is too often deployed not based on proven,
- 3:28you know, scalable market profitability, but rather it's
- 3:31strategically deployed by vendors to essentially create
- 3:35its own demand. That that really is the
- 3:38definition of financial artifice, isn't it?
- 3:40The key red flag here, and it's it's pretty glaring, is this
- 3:45massive disconnect between the commitments being made and the
- 3:48current financials. The material shows open AI's
- 3:51potential compute obligations could easily exceed a trillion
- 3:55dollars over the lifetime of some agreements, yet its annual
- 3:59revenue is what, around $13 billion?
- 4:01And it's still operating unprofitably.
- 4:04This huge gap, it's being bridged by what the analysis
- 4:08calls vendor finance ecosystems. You know, NVIDIA, for example,
- 4:12is is functionally acting like the AI economy central bank.
- 4:17It's underwriting demand for its own products through direct
- 4:20investment, like committing up to $100 billion to Open AI and
- 4:24providing financial guarantees across the sector.
- 4:27This feels like an engineered boom designed specifically to
- 4:31accelerate hardware deployment by, well, bypassing traditional
- 4:34market validation. It doesn't feel like organic
- 4:37growth supported by sustainable end user revenue.
- 4:41And you mentioned strategic foresight, which is an
- 4:43interesting point, but I'd frame that financial interdependence
- 4:45very differently. Let's look really closely at
- 4:48that, that now infamous NVIDIA Open AI Oracle triangle, because
- 4:53the structure itself, it's inherently circular.
- 4:56And that circularity what fundamentally dictates valuation
- 5:00and crucially ensures there's no real price discovery.
- 5:03Think about it. The $100 billion NVIDIA
- 5:05investment in Open AI provides the core capital that enables
- 5:10Open A is massive $300 billion compute purchase from Oracle,
- 5:14right? Then Oracle turns around and
- 5:16uses a huge chunk of that inflow to fund its own $40 billion
- 5:20CapEx purchase of, guess what, NVIDIA chips.
- 5:23Now, while this loop provides those financial backstops in the
- 5:27planning security you're championing, it fundamentally
- 5:30blurs the line between customer and supplier.
- 5:32I mean, when the hardware maker is underwriting the demand for
- 5:35its own product and financing its customers purchases from the
- 5:37cloud provider, well, demand is being manufactured by vendor
- 5:41financing, right? It's not being validated by
- 5:44genuine external market update. Doesn't that you know that lack
- 5:47of organic validation? Were you at all?
- 5:50I see why you'd characterize it as manufacture demand, but I'm,
- 5:54I'm just not convinced by that line of reasoning because this
- 5:57framework, this structure, it provides the essential financial
- 6:01backstops, they're necessary for infrastructure projects of the
- 6:04share scale to even launch at the speed required.
- 6:07And I think we agree the speed is required for this super cycle
- 6:10to actually happen. If we sat around waiting for
- 6:13purely organic sustainable end user revenue to justify say $300
- 6:18billion in CapEx upfront, well, this foundational infrastructure
- 6:22would simply never get built or at least not nearly fast enough.
- 6:26The commitment structure, it's kind of analogous to a utility
- 6:29needing long term power purchase agreements just to secure the
- 6:32financing for a new power plant. Nvidia's strategic ecosystem
- 6:36support, yes, it's vendor financing is designed
- 6:39specifically to de risk the aggressive capital investment
- 6:42that these neo cloud providers need to make.
- 6:45Look at the the $6.3 billion agreement they made to purchase
- 6:49Corrie Weaves unsold cloud capacity through 2032.
- 6:53That's a crucial financial backstop.
- 6:56It allows a specialized cloud provider and Neo cloud to
- 6:59aggressively build capacity based on the assumption, a
- 7:01reasonable one, I think that the demand will materialize quickly.
- 7:06It's about maximizing resource deployment to accelerate what
- 7:09seems like an inevitable future, not about disguising week long
- 7:12term fundamentals. OK.
- 7:14I understand the stability argument you're making.
- 7:17Those backstops do provide some temporary planning security.
- 7:21But focusing on agreements like the one with core Weave, I think
- 7:24it distracts us from the the massive counterparty risk that's
- 7:28inherent in that central Oracle deal.
- 7:31The resilience you talk about is is fundamentally undermined by
- 7:34the sheer scale of these staggering commitments.
- 7:38Moving away from Microsoft exclusivity only really helps if
- 7:41the other counterparties are financially sound, right?
- 7:45And in this system, failure isn't isolated.
- 7:48It cascades across those circular financial arrangements
- 7:52we just discussed. Let's really scrutinize that
- 7:55scenario. A you brought up what actually
- 7:57happens if Open AI, which is highly leveraged and still
- 8:01unprofitable, defaults on that $300 billion Oracle commitment.
- 8:05Oracle has already pushed its free cash flow into negative
- 8:07territory specifically to finance the required hardware
- 8:10build out for this team. So a default of that magnitude,
- 8:14it would be catastrophic for Oracle.
- 8:16It would leave them holding billions and sunk likely
- 8:18unrecoverable costs. It would wipe out a massive
- 8:21chunk of their projected cloud revenue, potentially $144
- 8:25billion by 2031, according to the analysis.
- 8:29And it would almost certainly trigger the cancellation of its
- 8:32its $40 billion GPU orders from NVIDIA.
- 8:35Bang. That failure instantly
- 8:37propagates the shock right across the value chain.
- 8:40So how can we possibly view this, you know, the stability of
- 8:43these physical assets as somehow mitigating the acute systemic
- 8:47counterparty risk created by these highly leveraged model
- 8:50developers whose future monetization path is still,
- 8:53let's be honest, entirely speculative?
- 8:56OK, that scenario definitely paints an acutely dangerous
- 8:59picture, a plausible one of short term financial shock, I
- 9:02grant you that. But the argument that the assets
- 9:05themselves become worthless if just one customer fails, I don't
- 9:09think that stands up when you consider the foundational nature
- 9:12of the technology itself. You're essentially assuming that
- 9:15the failure of open AI means the failure of the entire AI
- 9:18paradigm, and it it just doesn't.
- 9:20The intrinsic economic value that I'm arguing for resides in
- 9:23the physical hardware, those state-of-the-art GPU clusters,
- 9:26the high density data centers. If Oracle is left holding these
- 9:30assets after an open AI default, well, guess who's waiting in the
- 9:33wings? AWS, Google, Microsoft.
- 9:36They are absolutely ready to acquire them, probably at a
- 9:39discount. These giants possess the balance
- 9:42sheets necessary to absorb shocks like that and integrate
- 9:45those distressed assets into their already functioning cloud
- 9:48ecosystems. The long term trajectory is
- 9:50undeniably towards cognitive infrastructure.
- 9:54It's a utility needed by virtually every major industry,
- 9:58and that means the assets retain significant value even if the
- 10:01intermediate, highly leveraged financing mechanisms happen to
- 10:06fail. The physical capability itself
- 10:09is the real strategic prize here.
- 10:11Well, that that sounds a bit like we're justifying
- 10:13potentially massive malinvestment by arguing that,
- 10:16you know, the strong will eventually clean up the mess
- 10:18left by the week. And Speaking of the weak, the
- 10:21argument that the risk is merely transparent and credit based, It
- 10:25completely fails to account for some critical governance
- 10:28failures among these highly leveraged players, failures
- 10:31which historically often precede outright financial implosions.
- 10:35We really have to look at the specialized corner of the market
- 10:38here, the whole neo cloud sector.
- 10:41Take Core Weave's S1 filing for example.
- 10:43It contains some severe red flags that go way beyond just
- 10:46normal business risk. They disclosed $15 billion in
- 10:50off balance sheet lease commitments.
- 10:53Now for listeners, this is critical.
- 10:54Off balance sheet commitments essentially means they were
- 10:58intentionally obscuring the true extent of their financial
- 11:01leverage. They were making their balance
- 11:03sheet look significantly healthier and less indebted than
- 11:07it actually was. And what's more, they explicitly
- 11:10admitted to material weaknesses in our internal control over
- 11:15financial reporting. Now, this isn't just, you know,
- 11:17sloppy bookkeeping. This is a classic precursor to
- 11:21major accounting scandals, and it poses a massive threat to
- 11:25investor trust across the entire neocloud ecosystem.
- 11:29If a firm that's built on vendor financing is simultaneously
- 11:32obscuring its true leverage, well, how can any investor
- 11:35really trust the valuation or the solvency of its contractual
- 11:39partners? A localized failure, esecially
- 11:421, involving questionable governance.
- 11:44It poisons the well for the entire financing structure.
- 11:47Right. And while those governance
- 11:49warnings are certainly serious red flags, no doubt about it,
- 11:53for investors in any specialized entity like Core Weave, I still
- 11:57maintain their fundamentally localized, operational and
- 12:01frankly management failures. The core technology itself and
- 12:04the massive underlying demand signal, those remain unchanged.
- 12:08We seem to be discussing the difference between a systemic
- 12:12structural flaw in the technologies valuation versus
- 12:15the failure of individual, perhaps over leveraged or poorly
- 12:19managed players. Core Weave specific issues don't
- 12:22fundamentally invalidate the necessary build out being
- 12:25undertaken by the true hyperscalers.
- 12:27The source material strongly suggests that giants like AWS,
- 12:31Google, and Microsoft, they have the mature internal controls,
- 12:34they have the massive balance sheets to avoid these kinds of
- 12:37governance pitfalls, and yes, they have the ability to acquire
- 12:41distressed assets at a discount when smaller, maybe less
- 12:44disciplined players inevitably fail.
- 12:47The pain in that scenario will likely be localized primarily to
- 12:51the unsecured creditors and the equity investors in Cor Weave.
- 12:56But the hardware itself, it won't sit idle for long.
- 12:59It will almost certainly be folded into the broader super
- 13:02cycle build out. So ultimately what we've mapped
- 13:05out here is an infrastructure boom built on well immense
- 13:09leverage and these incredibly complex, often self referential
- 13:13financial arrangements. And while the underlying
- 13:16technology is absolutely genuine, potentially highly
- 13:19valuable, the methods of financing these massive long
- 13:22term inflexible commitments based on highly speculative
- 13:25future revenue structures that clearly prioritize immediate
- 13:29demand creation over proven organic market profitability.
- 13:32All of this strongly suggests we are currently living through a
- 13:35highly speculative, debt fueled period of malinvestment.
- 13:39The entire system seems predicated on a future where
- 13:42just a few still largely unproven model developers
- 13:45managed to monetize their technology to an unprecedented
- 13:48degree in corporate history. And that core assumption, I
- 13:50think, is the fundamental systemic flaw that threatens to
- 13:53create significant financial volatility down the road.
- 13:56Yeah. I think the debate today really
- 13:58highlights this fundamental tension, doesn't it, between the
- 14:00technological necessity on one hand and the, let's say the
- 14:04financial execution on the other to build the required scale of
- 14:08cognitive infrastructure at the necessary seed, while
- 14:11unprecedented capital mobilization is mandatory and
- 14:14that inherently requires creative largescale financial
- 14:18structuring, which includes, yes, vendor financing.
- 14:21We fully acknowledge the acute risks created by that vendor
- 14:24financing and especially by the extreme concentration of these
- 14:28commitments, which indeed creates pockets of significant
- 14:32fragility, particularly among maybe less disciplined firms
- 14:35like Core Weave. As you pointed out.
- 14:37However, we still maintain that the intrinsic long term economic
- 14:43value of the physical hardware which will be needed regardless
- 14:47of which specific model developer ultimately succeeds
- 14:50and the what I see as the inevitability of the AI super
- 14:54cycle justifies this massive build out.
- 14:57Even if, and this is possible, a painful consolidation event or
- 15:02correction is necessary along the way to separate the
- 15:05sustainable infrastructure from the the purely speculative
- 15:09financial overreach. Investors really need to look
- 15:13beyond the hype here. Absolutely.
- 15:16The source material provides a very clear warning, I think that
- 15:20traditional financial metrics just aren't sufficient anymore.
- 15:23Investors really have to scrutinize the structural
- 15:26integrity of these incredibly complex contractual webs to try
- 15:30and distinguish between sustainable infrastructure
- 15:33investment and, well, purely speculative overreach.
- 15:37There is a clearly much more to unpack within this material.