323 | Breaking Analysis | Did Jensen just make the AI buildout too big to fail?

This is Breaking Analysis with Dave Volante. >> Nvidia is no longer just selling technology. It's helping create a financial asset class around AI compute. In our last breaking analysis, we argued that AI can be technologically transformative and still produce a capital bubble. Our thesis was simply that the AI bubble pops if deployable supply grows faster than monetizable demand and financing stops bridging that gap. Jensen Wong has just attacked that weak link directly. Nvidia announced partnerships with Apollo, Black Rockck, Blackstone, Brookfield, Goldman Sachs, and KKR to establish financing platforms designed to mobilize more than $500 billion for AI infrastructure. This is not a funded $500 billion pool today. The final agreements still have to be completed, but the goal is pretty clear. Specifically, Nvidia is trying to turn AI compute into collateral and the AI factory into a repeatable financable infrastructure asset. And that makes AI much more than a chip story. If theou turns into solid agreements, it intertwines AI with credit leverage, customer contracts, productive monetization, cash flows, and the residual value of aging silicon. And if this market scales, as we believe it will, the same assumptions about AI demand are going to connect semiconductor suppliers, neoclouds, data center developers, utilities, private credit funds, infrastructure investors, and of course, governments. A failure in one part of that system may no longer stay contained. Did Jensen just make the AI buildout too big to fail? Not yet, but he may be making it too interconnected to fail quietly. Welcome to this breaking analysis number 322, and we've titled it, did Jensen just make the AI buildout too big to fail? In this episode, we'll briefly explain how computebacked credit works, why Nvidia's residual value support is an important tell sign, what Coreweave and Nibbius's earnings prints reveal about the current demand and economics picture, and whether this new capital market reduces the AI bubble risk or simply pushes it downstream because independent capital can extend the buildout, but independent capital is not independent demand. Let's start with exactly what Nvidia did and did not announce. Nvidia signed memorandums of understanding with those companies that we mentioned up top, Apollo, Black Rockck, Blackstone, Brookfield, Goldman, and KKR. And the intent is to establish independent financing platforms designed to mobilize more than half a trillion dollars of third party capital over time. This is obviously a major announcement, but it's not $500 billion of Nvidia revenue. It is not one funded pool. It's not an immediate commitment to specific customers or projects and the final agreements have not yet been completed. So there's a headline number and that's exciting, but the mechanics are much more important to understand. Specifically, today many AI factories are financed one company and one project at a time. Builders use some combination of corporate debt, customer prepayments, asset back loans, maybe [clears throat] equity, and sometimes vendor financing. Nvidia is essentially trying to make that process repeatable. And the idea is to bring longduration institutional capital into the market and underwrite AI factories against customer commitments against utilization cash flows and the expected residual value importantly of the installed compute. Now the most important phrase in the announcement we think came from Goldman Sachs. It's to quote create a market for credit backed by Nvidia compute. That is the transition that we need to better understand. Nvidia wants its systems to be treated as more than technology equipment. It wants the compute to serve as collateral and the AI factory to become an investable infrastructure asset class. Now, if this works, capital can move away from individual company balance sheets and into infrastructure funds or private credit or insurance capital or other institutional pools. And that could potentially reduce the cost of capital and broaden access to AI infrastructure. But it does not eliminate the risk. It does change who holds the risk, how the risk is financed, and how widely the exposure is distributed. So don't think of this as a program designed to sell more GPUs. I mean, it is that, but it's much much more. Nvidia is attempting to build a capital market around its architecture by making AI infrastructure an investable asset. And this is the key to understanding this perspective deal. The AI chip, the AI chip cycle is becoming a credit cycle. So the next question is how an AI factory actually becomes a financable financable asset and what investors are being asked to underwrite. And to understand what Nvidia is building, let's put our banker hats on and think like a finance lender. So this proposed structure is similar to the financing used for things like power plants or aircraft fleets or large infrastructure projects. Institutional investors provide debt and equity to a dedicated financing vehicle, often called a a special purpose vehicle or SPV. You've heard that term thrown around. The SPV uses the capital to buy or lease the NVIDIA systems, secure the site and the power, build the AI factory, but the physical infrastructure is only one part of the asset. The complete asset includes the NVIDIA platform, the customer contract, the site, the power connection, the expected monetization profile, and importantly, the residual value of the equipment after the first contract ends. So, the customer agreement or what's called an offtake contract is super important because the lender wants to know four things. One, who's obligated to pay? Two, how long is the commitment? Three, is the contract a take or pay, meaning the buyer either takes a minimum amount of product or pays for the shortfall if they don't take the product? And four, can the customer cancel or they can they delay acceptance or renegotiate the price? Once the factory is operating, usage revenue has to cover power, cooling, maintenance, all the operating costs. It's got to service the debt and it's got a return that that that that is required by the equity investors. It's got to hit that return. And then there's the important residual value question. When the first customer contract ends, can the cluster be leased to another customer? In other words, does it have enough value to be redeployed to say inference or a different workload? And what is that value? This is why Nvidia emphasizes that its systems are fungeable, transferable, and improved over time through its software, namely CUDA. Now, those salient characteristics are intended to support a longer economic life and give lenders confidence that the equipment still has value even if the original customer leaves. So, from the underwriters perspective, you they really don't care about the AI hype. What they care about is if this the specific AI factory generates enough predictable cash flow and retains enough recovery value to support the capital structure. Is it a good investment? This is how compute becomes collateral. Capital funds the factory. Customers rent the output. Lenders underwrite the utilization, the cash flow, and the recovery value. And this framework also tells us exactly where the risk moves if demand, pricing or residual value fails to live up to expectations. So, this may all sound like infrastructure finance, very much like the the leasing of IBM mainframe computers back in the 80s and 90s, but once these loans and leases begin to be pulled and distributed, the model starts to resemble something more like assetbacked credit and eventually perhaps even securization. And there's a lot of talk in the media about how this is like mortgage back securities. So, we but we need to be careful about the MBS analogy. It's kind of useful and instructive, but it can get ahead of the actual facts. So, let's look at those. What Nvidia really announced is not securization. Today, we're not seeing pools of AI factory loans being divided into tranches and raided and sold into broad secondary markets. Nvidia, what they've announced is financing platforms andou around financing platforms and dedicated pools of institutional capital and the final agreements of course are still still pending. So the point is this is not yet mortgageback securities. What we're seeing is a steady movement. The first stage is project financing. A lender finances a specific AI factory against a customer contract, a site, available power, and the forecasted cash flow. The lender then underwrites that individual project. The next stage is equipment leasing and securing the debt. Here the computing systems and the customer contracts contracts help support that borrowing. We have clear evidence by the way that this is already happening. Cororeeave has financed high performance computing infrastructure through syndicated term loans including financing supported by shorter duration customer contracts. Nibbius completed a $775 million assetbacked facility secured by deployed GPUs and contracted cash flows from an investment grade customer. Now the third stage as we're showing here is portfolio finance. So instead of financing a one-off AI factory, investors pull multiple projects across customers, across operators and across geographies. Now that diversification in the operating data center or uh uh created over time can make the asset class easier to underwrite and that appears to be the direction of Nvidia's institutional platforms. Then potentially comes securization. So if transaction volume grows and the assets develop a reliable and proven performance history, loans or leases could eventually be pulled divided into senior and junior tranches and distributed to a broader investor base. But that is a possible future state and is not what was announced. The mortgage back securities analogy, it helps us understand pooling and trunching as remember the big short and of course the distribution but also gives us the warning. Financial diversification can sometimes hide economic concentration if every loan depends on the same demand assumptions and collateral values. Let's take some examples. frequent flyer securizations. That shows us how an unusual future cash flow structure can support borrowing when investors think those cash flows are durable. Aircraft le leasing is probably even a better analogy from an operating standpoint. In this situation, you have standardized assets, you got multiple potential customers, you have recurring lease revenue and strong residual values after the first contract ends. You can always, you know, resell the plane. Even that comparison however has limits. An aircraft can be flown to another customer. A complete AI factory is tied to its power infrastructure, its cooling infrastructure, networking software and the physical site and where that site is doiciled. So the key question is not whether Wall Street can package this risk. Wall Street can package almost anything. The more important question is whether the packaging and the financing actually divers diversifies truly diversifies the underlying economics. In other words, pooling projects does not diversify the risk. Let's say if every project depends on the same exact customers, the same NVIDIA architecture, the same utilization assumptions. And that brings us to the most revealing part of this announcement. Nvidia's willingness to provide residual value support. So, let's look at that. Nvidia CEO Jensen Wong announced that Nvidia has the option the option to backs stop up to $125 billion at 25% of a massive deal this half a trillion dollar deal of of AI infrastructure financing. Now does that backs stop validate confidence or does it show that the lenders are a little squishy and still require credit enhancement? And the answer is both. A lot of media reports interpreted the term backs stop in a really negative light. But what they failed to convey is that the backs stop is at Nvidia's option. In other words, if the financer feels the deal is too risky, Nvidia at its sole option can absorb up to 25% of that risk. But if Nvidia doesn't feel like the project is viable, it can walk. It can choose not to provide the backs stop and the deal blows up. And this underscores the most important stress test in the entire financing model. As we explained earlier, Nvidia's premise is that its compute is not disposable technology equipment. Rather, it's an investable asset. But lenders ask a different set of questions. If the original customer leaves, can another customer take the capacity quickly without costly migration or reconfiguration or export control issues or you got data gravity friction and sovereignty? Can CUDA improvements, this is important, can the improvements in software offset the performance and per per watt and power efficiency advantages of subsequent generations? If you can get more out of a software turn that makes the residual values potentially more attractive. Another question are the potential offtakers truly diverse or are many projects ultimately dependent on the same anthropic and and AI and hyperscalers and sovereign buyers or is it more diversified and most importantly does the capacity generate enough cash after the whole power cooling site expense maintenance operations debt service and refinancing costs you know does it produce monetizable uh cash flow Now, some early evidence supports part of Jensen's argument. Cororeweave says a typical 5-year contract can repay the asset level debt used to build the cluster. It also recently contracted, according to the company, A100 capacity through 2029 at what it described as an attractive price. Even though the A100 architecture was introduced in 2020, Cororeweave says its prior generation AER and Hopper fleets also remain largely sold out. So this is reasonable evidence that older Nvidia infrastructure can retain its commercial value, but it's not yet a full cycle stress test. Those residual values are being seen during a period when supply remains constrained and rental pricing is unusually high. The real test comes after a capacity surplus cycle when newer systems are broadly available, rental prices normalize and customers have more alternatives. That is the key distinction at the bottom of this slide. The functional life is not the same as economic residual value. In other words, a GPU can remain technically useful and still fail to earn enough future cash flow to support its carrying value or capital structure. The independent underwriting, it does create discipline if lenders are willing to reject projects that are of marginal value or too risky. Now, if Nvidia must provide residual value support that backs stop, that doesn't mean the asset thesis is wrong. It means the market maybe has not yet accepted the thesis without some credit enhancement. The next question is whe is what happens if capital becomes tighter and the upfront payments start to matter more than the lifetime total cost of ownership. In other words, if if I can't fund the initial capital outlay, I don't really care if Nvidia's perf per watt is better. So, let's test Nvidia's asset class thesis against some more evidence. If compute back credit is going to become a durable market, the Neoclouds are a reasonable proving ground. I mean, Nvidia's essentially, you know, seated them. And right now, that proving ground is flashing green on our dashboard, but mainly on the front half of the cycle. So, let's start with Cororeweave and explain what we mean. The company reported $2.6 6 billion of quarterly revenue, up 112% and ended the quarter with 104 billion of backlog. Now, that figure did not include more than 25 billion of additional customer commitments signed shortly after quarter's end. Management says near-term capacity is effectively sold out with multiple buyers competing for each GPU brought online. Pricing and expected contribution margins on recent contracts are also rising. More than half of Core Weave's backlog is already attached to contracts where delivery has begun and management expects that figure to exceed 2/3 by year end. This is important because backlog is beginning to convert into installed revenue producing capacity. Let's look at Nebius. They provide similar evidence from a different operating model. It signed four four AI cloud deals averaging more than a billion dollars each. Customer prepayments cover roughly 50 to 60% of those associated capex and management says it could sell its entire planned 2027 capacity today if it chose to do so. Its capacity auction also cleared 15% above its previous record price, showing that scarcity, not surplus, still clearly defines the current market. No shock there. We're also seeing preliminary support for ini Nvidia's residual value thesis. Coreweave recently signed an A100 contract extending into 2029. As we said that architecture was introduced in 2020 and airier and hopper lines as we say also remain sold out. So the financing market is definitely responding. core we've raised of 18 billion during the quarter and more than 32 billion cumulatively and it it it it's in its latest uh structure support shorter term duration customer contracts. Not that we haven't seen him work through this whole cycle and have these things be self-unding yet, but there's strong evidence that you know this model at least for now is working. Now, NEBI has completed a $775 million assets back assetbacked facility secured by its deployed GPUs and contracted cash flow. So, there's again more evidence that that you can monetize these lit up GPUs. Now, inference is also emerging as a second monetization vector. Goreweave's managed inference uh book uh booked ARR increased from a million dollars, get this, from a million dollars to more than a hundred [snorts] million within several months. And the company said it expects at least$ 250 million by the year end. So this current evidence validates four things. One, demand is real. Two, pricing power is is in place today. It's strong. Three, capacity in these examples is being productively utilized and the assets are increasingly number four financable, but it does not yet validate the complete full economic cycle. As we say, this is largely still being funded on speculation. Cororeeve reported 9.4 4 billion of quarterly capex, 640 million of interest expense, and $626 million net loss. Nibbius is relying heavily on customer prepayments and asset back debt at continued external capital while guiding to $20 billion to $25 billion in annual capex. Neither company has yet demonstrated that it can fund a complete hardware replacement cycle from organic free cash flow. That's especially important after scarcity pricing normalizes. And as we've suggested, the Neoclouds, they need to diversify. Many are doing so. I mean, otherwise, we think there's simply going to be a low margin distribution channel for NVIDIA hardware. Corewave's acquisition of Weights and Biases to build out a software stack is a good example, and Cruso's move into diversified infrastructure like storage and networking are other examples of this diversification. We would expect that to continue over time as a clear hedge if and when supply and demand come into equilibrium. Nonetheless, the key test remains the following. Can the next generation of infrastructure be funded from the cash produced by the current generation without depending on another large debt raise or an equity issuance or customer prepayments or a vendor backs stop? Our conclusion is the quarter validates demand, pricing, utilization, and finance ability. It does not yet validate full cycle returns on invested capital. And that determines whether computebacked credit becomes a durable infrastructure asset, as Jensen says it will be or is, or it simply finances the next stage of the buildout before the market clearing event that we talked about last week arrives. And this brings us back to the AI bubble forecast that we published last week. You know, we don't think we should change a probability distribution simply because Nvidia announced a bunch ofus. The half a trillion is not yet funded capital and and the final agreements they have to be completed. So, you know, the right side of this chart is obviously conditional, but let's assume that these things happen. What happens if these deals actually close? They attract capital and begin financing AI factories at scale. The immediate effect is to reduce the probability of an early financing led break what we were talking about last week. So we previously signed a 10% probability to a broad break beginning in 2027. Under the conditional case, you know, we think that falls even lower. So let's call it 5%. The 2028 probability also declines from 25% to 20% in our view. And that's not because the underlying economics have suddenly been proven. It's because institutional capital can bridge the gap. While remember, we talked about this last week, high bandwidth memory and packaging and power sites remain constrained. That bridge is critical. And while customers continue absorbing available capacity, the recent core and nebious results support that delayed reckoning scenario. That's a good thing. Demand clearly remains strong. Pricing remains elevated. Capacity is being absorbed. And the financing market is becoming more willing to lend against contracted compute cash flows. But more available capital does not eliminate the market risk. It postpones that clearing event that we talked about last week. As more projects receive financing and more hardware gets ordered and more sites are built and and lit up and more capacity eventually becomes energized, that increases the amount of infrastructure that must ultimately find productive workloads and generate cash flows. So we think the risk shifts out later. So our 2029 probability falls a little bit from 30 35% 30% but it remains a major test year as more of the current buildout reaches productive deployment and the 2030 then probability rises. So we're pushing it out from 20% to 30%. In other words, the risk window becomes that 2029 through 2030 period rather than you know what we had last week in a specific year. That is when utilization and rental pricing and refinancing and residual values are more likely to face a genuine full cycle test. We'll hopefully see it by then. The probability of a soft land landing or a series of rolling segment level corrections after 2030 also rises modestly in our view because of the this announcement. But there is one important caution from Ben Thompson's recent analysis. If you don't know who Ben Thompson is, you should be reading his stuff. He's exceptional. Anyway, the financing cycle can turn before the operating cycle is a point that he made. In other words, clusters can still be sold out and rental pricing can remain strong while lenders begin to widen spreads, reducing advanced rates, requiring more equity or applying larger residual value haircuts. So, even this conditional distribution assumes the new platforms remain open and willing to finance projects on attractive terms. So here's the paradox. More capital makes an early break in in this this pop in this bubble less likely, but it can make the eventual utilization monetization and residual value test even more critical. So that changes the likely mechanism of a correction if it occurs. Instead of the buildout stopping because companies can't finance construction construction, the eventual break could come through weaker productive utilization, lower rental pricing, residual value markdowns, and refinancing pressures after more capacity reaches the market. So Nvidia may be reducing near-term financing risk, but it may also be increasing the stakes of the later market clearing event that we talked about last week. So let's bring this argument together. Nvidia is trying to solve the capital bottleneck. If the financing platforms in the announced come to fruition and work, more AI factories are going to be funded. More GPUs can be purchased, more sites can be built, and companies with real compute demand can gain access to capital at a lower cost. Great. That reduces the risk of builders running out of money before the infrastructure becomes productive. But it does not solve the full bubble problem. It moves the decisive event downstream. So the next constraint becomes creditw worthy customer demand, then productive utilization, then residual value, then cash flow. Remember, independent capital does not create independent demand. The lenders, you know, they may be different, but the projects may still depend on the same frontier labs and hyperscalers and sovereign buyers and [clears throat] those same assumptions about AI adoption. And that creates a systemic concern. If many institutional portfolios own loans backed by the same Nvidia systems, the same customer contracts and the same utilization forecast, the financing might look diversified while the underlying economic risk remains concentrated. So the key warning signal is not lower GPU rental prices in and of themselves. Lower prices could actually expand demand and create a healthy volume cycle. I.e. Jebans paradox. The alarm goes off if three things happen together. If rental prices fall, if productive monetization weakens, and if financing terms tighten, at that point, residual values will tank. Lenders would reduce advanced rates and borrowers would need more equity, and refinancing becomes harder. We could take a lesson from 1986 when Congress rescended the investment tax credit, the ITC. I remember I was at IDC at the time and we had this leasing planning service which created residual values for mainframes. At that time, mainframe residual values suffered a steep collapse when the tax advantages for leasing incentives dried up and it happened to coincide with a huge technological shift toward less expensive microprocessorbased systems and mark the downfall of IBM as the leading company in the technology industry. I'm not saying quantum is going to do that to to accelerated computing, but forecasting technology cycles is a risky business sometimes. The point is a financing cycle can turn before the GPUs go idle. And this is the Ben Thompson comment that we believe is most worth highlighting. When capital is abundant, buyers optimize around total total cost of ownership, i.e. Perf per watt. When capital becomes scarce, the upfront purchase price required to to the the the required check that has to be written and the time to cash flow become much much more important. In other words, if I can't write the initial check, I don't care about the total cost of ownership. So, this move by Jensen potentially addresses the funding question for now. And the critical point becomes, can the factory earn enough to justify the funding? All right, let's close with this move by Jensen and how it affects our current scorecard that we're showing here. The announcement is a profound validation of AI infrastructure as an emerging asset class is it's unbelievable and awesome. Nvidia has brought together six of the world's largest institutional capital providers to establish financing platforms designed to mobilize more than a half a trillion dollars over time. Just let that sink in. But the announcement also makes this dashboard more important, not less important. Why is that? Because the question shifts from how much capital is being committed, that's not the problem anymore, to does that capital convert into productive monetization, durable cash flows in an asset that retains its value through a complete cycle. Right now, the green signals are quite constructive. We're seeing demand broadening. Near-term capacity remains effectively sold out. pricing and contribution margins and earnings. They're solid. There's new capacities entering uh revenue producing workloads and the financing market is demonstrating that it's going to lead against Nvidia infrastructure or lead lend against Nvidia infrastructure and contracted compute cash flows. That's the thesis and it's playing out. That's why we believe the near-term bubble risk has moderated somewhat. But the yellow signals tell us that the difficult underwriting tests are still ahead. Backlog has to become lit up and energized. It's got to be accepted and that's got to become billable capacity. And older systems have to retain their value after scarcity scarce pricing begins to normalize. And credit markets have to remain open if the spreads widen, if the advance rates decline, or if lenders require larger equity checks. And if capital tightens, buyers may care less about lifetime TCO and more about the upfront acquisition cost and time to cash flows. Then we have to read the red signals. Can't forget about those. Neither Coreweave nor Nebius has yet demonstrated free cash flow after the full burden of capital expenditures and interest at the scale being contemplated. and neither has completed an entire hardware replacement cycle funded organically from the cash generated by the prior generation. That is a decisive test of whether this becomes a durable infrastructure asset class. By the way, just a side note, Microsoft is currently the only hyperscaler promising positive cash flow. So, our current take is lower probability of a broad 2027 break, a stronger delayed reckoning case, and a wider primary risk window in 2029 and 2030. The likely break path also moves downstream. In other words, it becomes less about an immediate inability to finance construction and more about productive utilization and GPU rental pricing and residual value of haircuts and refinancing once substantially more capacity hits the market. That is why we say the bubble is deferred. It's not disproven. Could the bubble mimic the sports franchise bubble where valuations have gone up perpetually? Maybe. Let's hope. But look, independent capital can fund more factories, but independent capital is not independent demand. Now, Jensen may not have made the AI buildout too big to fail, but he may be making it too interconnected to fail quietly. As always, we'll be watching. And thank you for watching this breaking analysis. I'm Dave Volante.