Friday, November 28, 2025 12:03:40 AMEST

- The deal is expected to be finalised in the coming weeks.

**Reminder Nov 7th: ORCL Reportedly banks lining up $18B loan for an Oracle related data center in New Mexico - press
11/24 ORCL CreditSights' senior analysts: Oracle could issue ~$65B more bonds over the next three years, but Oracle needs to maintain investment-grade ratings, because the amount of funding available to lower-rated companies simply isn't enough to support its needs

***TTN Research Alert: Oracle's still missing AI jumbo financing is the quiet candidate for the next DeepSeek style shock, a GFC era CDS scare (Oracle CDS rose into triple digits) hiding in plain corporate plumbing rather than in a new China model demo
- The shock that could blindside AI infrastructure and data center names as abruptly as the 2008 CDS crisis or DeepSeek’s January 27th 2025 rout is not another model release, but a moment when lenders quietly refuse to fund the next wave of Oracle-linked AI debt. DeepSeek’s R1 assistant erased about $593 billion of Nvidia’s market value in a single session and knocked roughly 3.1 percent off the Nasdaq, reminding markets that an apparently bullish AI milestone can instantly be reframed as a threat to incumbents’ economics. In today’s environment the analogous surprise would be credit markets balking at swallowing $30–40 billion chunks of concentrated Oracle data center risk just as AI enthusiasm has invited a flood of specialised issuance. Over recent weeks, reports have flagged a record $38 billion dual debt package for Oracle-anchored facilities in Texas and Wisconsin, yet what has actually materialised is an $18 billion public bond and an approximately $38 billion loan facility that leave the much-trailed jumbo capital markets deal conspicuously absent. That divergence is appearing just as Oracle’s 5-year CDS has climbed into the low-80-basis-point area, the highest in years, while CoreWeave’s 5-year CDS has blown out above 500 bps, assigning default odds that would be more familiar in project finance than in a supposedly derisked AI darling. By 2026, the testable outcome will be simple: either a full-scale Oracle AI data center package has cleared the market on benign terms and CoreWeave’s CDS has mean-reverted, or credit will have been confirmed as the first real choke point of the AI era rather than compute or power.
- DeepSeek’s January 27th 2025 shock offers the cleanest template for how an apparently esoteric development can knock trillions off AI-linked valuations almost overnight, and the numbers still look surreal in hindsight: Nvidia fell nearly 17 percent, losing $593 billion in market cap, while the Philadelphia Semiconductor Index dropped 9.2 percent and data center names like Vertiv slumped close to 30 percent in a single US session. That episode did not reflect a sudden change in realised AI demand so much as a violent repricing of expected economics once investors saw that a low-cost Chinese model could deliver similar quality on far cheaper hardware and at a fraction of the running cost. In the same way, a failed or sharply repriced Oracle funding push would not signal that GPUs or data centers had become useless; it would signal that capital was no longer willing to finance capacity expansions on the assumption that demand, pricing power and AI ROI will eventually catch up. DeepSeek was a reminder that AI can be disruptive to incumbents even when it looks like technical progress; an Oracle stumble would illustrate that AI infrastructure itself can be overbuilt relative to the real pace of adoption.
- Credit markets rather than equity screens are now where the most revealing signals reside, and the numbers around Oracle are already stretching what used to be considered comfortable for a mature software incumbent. One more potential shock sits inside SoftBank’s $5.8 billion exit from Nvidia to fund a $30 billion OpenAI bet, which turns out to be a near-perfect diagram of the AI capex prisoner’s dilemma. SoftBank has now sold its entire Nvidia stake for roughly $5.8 billion, plus about $9.2 billion of T-Mobile stock, explicitly to recycle that cash into OpenAI and the Stargate buildout, even as Nvidia itself plans to invest up to $100 billion in OpenAI and supply at least 10 GW of systems, and Oracle has signed a roughly $300 billion five year cloud pact to host OpenAI workloads on Nvidia hardware. In balance sheet terms the same dollars are chasing themselves around a tight loop: SoftBank raises cash to fund OpenAI, OpenAI uses it to pay Oracle and Nvidia, Nvidia reinvests into OpenAI, Oracle raises debt to buy Nvidia chips for OpenAI, and Stargate sits on top promising up to $500 billion of US AI infrastructure by 2029 with SoftBank and Oracle both as owners and customers. What could really frighten markets, in a way that feels as abrupt as the GFC CDS shock or DeepSeek, is not this circularity itself but a forensic moment when regulators, rating agencies or an unexpected disclosure make clear that key Stargate and OpenAI cash flows have been pledged multiple times across these entities, or that part of Nvidia’s and Oracle’s apparent AI demand is effectively financed by SoftBank-levered equity at the other end of the loop; if that happens, investors who thought they were looking at diversified, independently underwritten AI growth would suddenly see a tightly coupled, levered ring of self-referential bets, and the re-rating across data center, GPU and AI platform names could be as fast as any model-driven shock.
- CoreWeave’s balance sheet, though smaller, illustrates the same structural tension even more starkly. Over the past 18 months the company has stitched together more than $25 billion in debt and equity commitments, including a $2.6 billion secured facility and a $1.75 billion senior notes deal, to fund an aggressive AI cloud build-out anchored by long-term contracts with OpenAI and Nvidia. Reuters and others now put CoreWeave’s net debt around $11–13 billion, with leverage near 7x EBITDA, levels that in any other sector would be treated as firmly speculative-grade. Yet the marketing story has emphasised a backlog north of $50 billion and take-or-pay contracts covering more than 90 percent of capacity, inviting investors to treat those commitments as effectively cash-like. The recent blow-out in CoreWeave’s CDS to around 510 bps, from roughly 360 bps in early October, shows that credit desks have stopped buying that narrative at face value and are instead demanding compensation consistent with a genuine chance that not all of that backlog will be realised on time or on budget.
- Numbers put that repricing in context. A 5-year CDS spread of 510 bps with a standard 60 percent recovery assumption implies an annual hazard rate of about 8.5 percent (0.051 divided by 0.60), which in turn produces a cumulative 5-year default probability of roughly 35% using the simple formula 1 - exp(-0.085 × 5). That is project-finance territory, not what one would normally assign to a company touted as a core pillar of the OpenAI–Nvidia ecosystem, and it sits uneasily alongside talk of 70 percent gross margins on owned infrastructure once capex settles. Oracle’s move is gentler but still notable: a 5-year CDS in the low-80-basis-point range is hardly distressed, yet it marks the highest level in several years and comes at a time when many other large investment-grade tech names have seen their CDS drift tighter, not wider. Against that backdrop, the absence so far of a clean, bond-market-tested $30–40 billion Oracle data center package starts to look less like mere timing noise and more like a sign that underwriters and buy-side credit committees are haggling over terms in a way that equity investors have not yet fully appreciated.
- Oracle’s role as linchpin makes the missing jumbo all the more important. In late September Oracle closed a conventional $18 billion multi-tranche bond deal, and separate reporting indicates that banks have since extended an additional roughly $38 billion of loans tied to data center projects, including a roughly $18 billion project-finance facility for a campus in New Mexico. By late October, however, financial press was describing a different instrument entirely: a record $38 billion dual debt offering linked to Vantage-owned sites that would be leased to Oracle, designed to test public investors’ appetite for securitised AI data center cash flows and flagged as potentially launching “as soon as next week”. As of mid-November 2025 in New York, that flagship dual package has not appeared in the bond tape in the form originally trailed, even as Oracle CDS has edged wider and the broader AI debt complex has swelled toward an estimated $200 billion of issuance this year. In a world where DeepSeek demonstrated that AI equity valuations can move on a single day’s software headlines, the first failed or deeply discounted attempt to securitise Oracle’s AI build-out would be the credit analogue: a sudden public signal that the marginal buyer is no longer willing to underwrite the story at any price.
- Hidden transmission channels make such a credit shock more systemically relevant than a simple single-name story might suggest. Neocloud providers like CoreWeave and peers increasingly rely on overlapping pools of private credit funds, structured vehicles and infrastructure-focused CLO-like products that package loans to data center projects, power developers and GPU financiers into yieldy paper for institutional buyers; Reuters’ recent piece on neoclouds notes that CoreWeave alone sits on about $13 billion in net debt while relying heavily on take-or-pay contracts with a handful of hyperscale clients. At the same time, the Stargate concept and similar ventures enlist sovereign funds, corporate cash piles and banks in multi-hundred-billion-dollar AI infra plans that explicitly assume continued access to cheap project finance and securitisation markets. A sharp wobble in just one or two marquee Oracle deals would therefore ripple through portfolios that are notionally diversified but in practice concentrated in one theme: highly leveraged, power-hungry AI capacity whose alternative uses are limited if AI demand disappoints. Those structures recall not the diversified CDS indexes of the mid-2000s but the more brittle, correlation-sensitive CDOs that amplified the 2008 crisis once housing stopped going up in a straight line.
- Skeptics will argue that there is an important difference this time: AI demand is visibly real in cloud revenues and GPU shortages, whereas pre-crisis housing demand depended on ever-looser lending standards. Microsoft, Amazon and Google have all reported strong double-digit growth in AI-linked cloud workloads, and Nvidia’s datacenter revenue has exploded in the past two years even after the DeepSeek shock.(Reuters) Moreover, contracts like Nvidia’s $6.3 billion capacity-backstop deal with CoreWeave, under which Nvidia agrees to buy unused capacity through 2032, offer non-trivial protection to some neocloud operators by effectively transferring some utilisation risk back to the chip supplier. A robust counter-story therefore runs like this: Oracle and CoreWeave may look stretched, but the secular AI demand curve is so steep that time and utilisation will heal today’s leverage metrics, and the market is simply repricing early exuberance rather than foreshadowing a genuine funding crunch. The single datapoint that would most strongly validate that view would be a successful placement of a $30–40 billion Oracle-linked securitisation at spreads only modestly back of recent tech IG issuance, accompanied by stable or tightening CDS despite the added supply.
- Across the sector, the macro numbers look increasingly like a classic overbuild. The Financial Times recently estimated that AI-linked debt and equity deals for data centers, chips and related infrastructure will total around $200 billion in 2025 alone, a figure approaching the peak annual run-rate of US telecom and fibre issuance in the late 1990s when too much capital chased a technology story ahead of usage. MIT’s GenAI research meanwhile frames a demand side still in pilot mode: 95 percent of enterprises seeing no ROI, most custom tools failing to reach production and adoption heavily skewed toward low-stakes productivity use rather than core workflows. Combining those two pictures suggests an uncomfortable ratio: tens of billions in annual fixed-cost AI infra supply being financed against cash flows that, outside a narrow set of hyperscale and frontier-model customers, remain speculative. Normalising that to something concrete, if we assume Oracle ultimately shoulders roughly $100 billion in AI-related debt against an AI revenue run-rate of perhaps $25–30 billion by the late 2020s, leverage within that segment could easily sit above 3x on a business that still carries sub-30-percent margins, leaving little room for error if enterprises continue to delay production deployments.

- By 2026, the public record should offer a clean verdict on whether this emerging stress is an echo of past debt-fuelled manias or a passing bout of indigestion on the way to sustainable AI returns. The single most informative near-term test will be the fate of the next genuinely large Oracle-linked AI data center package: if a $30–40 billion deal either fails outright, is drastically downsized, or clears only at spreads that shove Oracle’s consolidated funding costs sharply higher, it will confirm that credit, not compute, has become the gating resource of the AI boom. Conversely, a smooth syndication on reasonable terms, coupled with a durable tightening in both Oracle and CoreWeave CDS, would argue that today’s anxiety was merely the market adjusting to a heavier AI issuance calendar. Until that test arrives, the combination of a missing jumbo, widening CDS and an AI ROI drought means that the most consequential AI shock now lurking in markets is not a model like DeepSeek at all, but the first visible sign that the financing ladder beneath the data center build-out is less solid than it looked.