$nbis ms keeps ew, pt $140. wide base case ranges & bull/bear spreads. likes demand sigs & news 75%/85% of '27 40-45B capex covered by prepays & abf. watching execution, connected power progress & incremental financing decisions . . .
3
12
2,193
$nbis, $crwv, $iren,
Under-discussed implication of rising prepayments. Great to cover capex, but customer quality/liquidity matters. Scaling fast with leading ai-natives can work, so long as you have a diversified mix within and growing beyond your startup book.
But if the VC spigot starts closing, it will pressure the space and privilege the highest quality stacks . . .
Nebius expects $17–20B of customer prepayments in 2027. CoreWeave expects ~$9B over the next three months. Customers are increasingly financing the AI infrastructure they’re contracting to use.
I've seen people discuss the Bull read: Customers are committing cash to secure capacity. Prepayments reduce the provider’s funding requirement, while signed contracts support secured borrowing at better terms than unsecured corporate debt. CoreWeave says its contracted cash flows should cover costs and repay debt before counting renewals. If that underwriting holds, the initial customer contract pays for the infrastructure and subsequent bookings generate additional returns.
Seen less on the Bear read: Expensive unsecured funding is pushing more of the cash requirement onto customers. For AI labs, that means paying for compute before generating the revenue needed to support it. The supplier’s liquidity improves, while the customer commits more of its available cash upfront.
If demand disappoints, spreading customer payments over more years may also require lenders to extend repayment schedules. The GPUs keep depreciating during that negotiation too btw!
Customer quality becomes central. A contract supported by a cash-generating hyperscaler has a different funding profile from one supported by a lab dependent on further capital raises.
The market can reasonably read rising prepayments as evidence of scarce capacity and stronger supplier bargaining power. Sustaining those terms depends on customers continuing to earn enough from the compute, or continuing to attract capital to fund it.
1
7
1,243
$nbis
This was exactly my concern going into the print & why I wrote the long-dated debit put spreads w/collar. Doubt I’ll need them beyond peace of mind—a lot else went right, and the LT thesis is in tact.
That said, I was surprised and disappointed in the way both Arkady & Andrey blamed the 3rd party. They chose DO, and they’ve been involved throughout the build. Fair to say you’re taking on more self-builds for greater control (and margin). But they should have explicitly owned the delay alongside DO.
$nbis, alot to unpack from the call. but the focus on financing flexibilty & balance sheet protection stand out to me, as does the ambition to build dynamic price discovery & sales assist across all contract sizes.
the capex plans are undimmed by current credit market (~45B) and they're clear-eyed about optimizing their contract mix (price, duration, pre-pays) against attractiveness of financing terms. if/when debt is attractive, they'll grab it & divert more capacity to high priced spot; if terms too onerous, they'll shift capacity focus to >100% capex-covering pre-pay contracts (longer term).
nitter.gq/brandonjcarl/status/21…
Listening to the earnings call: this seems to be a reflection of the capital market conditions.
Nebius can fund through debt, equity or customer prepayments. Given the tightness in markets, they shifted prepayments from 52% of capex to more than 100%.
The AR isn’t a prepayment so much as the commitment to one.
All said: Nebius wants to prevent degradation of their balance sheet and minimize adverse impacts to cost of capital.
Of all of the neoclouds, I like this approach best.
7
1,216
$NBIS
waiting on call; some very good news in earnings report, but some hiccups & questions as well. the vineland/msft delay is unfortunately not a surprise (issue looks to be they can't run the gas gens in parallel through the bridging to the bloom fc's). other key question in the $325 one-time rev addition from the asset-lite model; excluding that drives the overall & ai cloud run rates down . . .
5
1,175
$nbis, $crwv, $iren, $orcl Great piece from Denis on Oracle's recent debt & equity financing, which may set a near-term floor for the neo's. Iren pointed to pre-payments & pacing . We'll learn more this week re: how others plan to handle current environment . . .
nitter.gq/DenBuivolov/status/211…
Oracle goes back for money $ORCL $NBIS
Remember I wrote that at its recent Investor Day, Oracle said its need for additional financing through fiscal 2030 is capped at $20B? The link to that post is in the comments if you missed it. Why am I bringing this up? In the current environment, Oracle decided not to wait and went straight to raising that $20B: as planned, $10B in debt and $10B in equity. Both deals closed on Monday, but things didn’t go exactly smoothly. Let’s go through the timeline.
The $10B of debt was initially planned in six tranches, but the floating-rate tranche was eventually dropped, leaving five, maturing in 2029, 2031, 2033, 2037 and 2046. The most interesting question was what yield buyers would demand. Later on Monday the company published the final terms:
🔹 2029 notes: $2B at 7%
🔹 2031 notes: $1.75B at 7.45%
🔹 2033 notes: $2.25B at 7.85%
🔹 2037 notes: $2.5B at 8%
🔹 2046 notes: $1.5B at 8.65%
That means yields came in 2.4–3.35pp above Treasuries of matching maturities. Of course, that’s not the near-15% yields on $CRWV bonds, but Oracle’s terms have clearly worsened too. For comparison, in September 2025 Oracle sold $18B of bonds, including a 40-year tranche, and the premium on the longest tranche was 1.37pp.
The next deal was in February 2026: $25B of bonds in eight tranches. Back then, the spread on the longest bonds was 1.95pp. Comparing bonds with the same maturities, spreads have widened by 1.45–1.75pp since February. For example, the yield on the 2046 notes rose from 6.56% to 8.67%, of which about 1.65pp came from a wider spread and about 0.45pp from higher Treasury yields.
What does this rise in yields mean? At the very least, credit quality is deteriorating, and investors aren’t willing to buy new Oracle debt at the old rates. For the company itself, it means more expensive debt service: the new $10B will cost about $780M a year in interest, versus $540M if it had borrowed the same amount at February rates. And this comes as Oracle’s interest expense already rose 55% YoY to $1.4B in Q1 of fiscal 2027 (ended in August).
It’s not critical yet, but against this backdrop any negative news will trigger sharp selloffs in the stock.
The $10B equity offering didn’t go perfectly either. The preliminary prospectus assumed selling 76.5M shares at Friday’s closing price, which, as a reminder, was $130.67. But in the end, according to Bloomberg, the shares were sold at $116, 11% below Friday’s close and only slightly above the 52-week low reached in July (~$114.5). To raise $10B, the company had to sell 86.2M shares, about 2.9% of shares outstanding. Not critical, but the question is whether this is the last offering, and the current price levels don’t add anything positive to the Oracle story.
The bottom line
Overall, given the current conditions, the financing came out moderately negative. It’s still unclear what Oracle will do about the heavy concentration of its backlog in OpenAI. This offering shows once again that AI infrastructure requires massive investment, and financing it will keep getting more expensive. Oracle’s rates also set a floor for what Nebius $NBIS would have to pay on unsecured long-term debt. Its earnings, by the way, come out on November 12. Before that, neocloud $IREN reports on November 5, and perhaps the most important report will be $CRWV on November 10: important in terms of management commentary and results, given where its bonds currently yield.
Not investment advice.
1
4
895
$nbis, $iren, $crwv Menlo Ventures out w/ State of Enterprise AI report based on survey of ~400 US co's. Trend data & market maps. Tidbits: E/AI spend accelerating (up 5.4x vrs 3.9x last year); agents driving inflection in token usage; open source taking off in usage (chinese mods dominate) but trails frontier in spend; post-training ecosystem beginning to emerge.
menlovc.com/perspective/2026…
5
1,132
$nbis Bloomberg piece is a round-up of current neocloud pessimism & known delay signals: msft capex slippage; gartner & elon guppy-level doom predictions, site-specifc challenges (iren: cabling; nbis: temp. gas engine fine; nscale: cancellations). Framing is around msft internal debates re: outsourcing to neos in face of unmeetable demand.
On the neo-side, they call out nbis as being clear-eyed & intrumental re: their HS relationships: "Nebius has been unusually upfront with its competitive strategy. It will scale up by inking contracts with big customers like Microsoft or Meta Platforms Inc. knowing that they’ll abandon neoclouds in the long run, Nebius CEO Arkady Volozh said during a March conference hosted by Morgan Stanley.
“We treat this business as temporary,” Volozh said, referring to his company’s deals with customers like Microsoft and Meta."
$nbis, $iren, $msft On this late day tempest that there's sat imagery suggesting Horizon4 and/or Vineland are responsible for the capacity delay msft announced at earnings . . . .
First there are no sat images re: $nbis/Vineland--that's just a headline conflation. There are sat images & report for Iren by a research firm called aterio that, they say, suggest an elevated delay risk (see below).
The included nebius reference as potentially contribting to the msft 2026 shortfall is simply a speculation based on timing & the lack of any confirmation from co/ that they have/will successfully begun the transition @ Vineland from temporary gas engines to the Bloom blocks. It's appears to be a guess based on no evidence.
I would take all of this w/ due skepticism . . .
aterio.io/blog/aterio-identi…
$nbis, $iren, $msft On this late day tempest that there's sat imagery suggesting Horizon4 and/or Vineland are responsible for the capacity delay msft announced at earnings . . . .
First there are no sat images re: $nbis/Vineland--that's just a headline conflation. There are sat images & report for Iren by a research firm called aterio that, they say, suggest an elevated delay risk (see below).
The included nebius reference as potentially contribting to the msft 2026 shortfall is simply a speculation based on timing & the lack of any confirmation from co/ that they have/will successfully begun the transition @ Vineland from temporary gas engines to the Bloom blocks. It's appears to be a guess based on no evidence.
I would take all of this w/ due skepticism . . .
aterio.io/blog/aterio-identi…
$nbis Comparatively de-risked domestic & international dc dev portfolio suggests election results (& implied slower reg. regime) likely to strengthen nebius's pricing hand. The co's US builds are far enough along in workable states to avoid moratorium threats (though site-level op, permitting & leg issues are still in play across cohort).
Of the 4 ground-up US builds, the site least far along is the PA Highbridge Business Park AI Factory, one of largest project campuses, spec'd for 1.2 GW. Highbridge had its Butler Township zoning expansion approved in August.
A quick dive shows PA's election results don't seriously imperil Highbridge, and may even strengthen the project over time. At the local level, both republican lawmakers representing the districts covering Butler Township & Schuylkill County were relelected: Tim Twardzik (State House District 123) and Dave Argall (State Senate District: 29).
Importantly, both are supporters of the Highbridge project, and dc's broadly, as important economic drivers for communities, so long as they are developed in close consultation with those communities (as Highbridge has been).
At the state leadership level, Gov. Shapiro won handily & enough state house seats flipped to give him a trifecta--which means he will seek to focus on codifying "GRID" standards statewide (House Bill 2650: GRID, Governor's Responsible Infrastructure Development).
That could have some tax status & power supply guarantee implications for Highbridge down the road. But a lot of what would be in a GRID statute already applies, and the bill would not stop construction or cancel permits. It states plainly that it doesn't affect municipal zoning.
Plus, clear rules have a protective benefit in that they are the administration's alternative to a moratorium. A codified GRID is a more likely outcome than a statewide pause.
20
2,129
$nbis "The reality is that life is a spiral." Not just a stack or a flywheel, but an interative, looped system that levels up. 🤔 Works for desrcibing the evolution of the co. itself, or the advent/escalation of ai broadly . . .
But my💰's on a shout-out to the integrated training/inference helix driving nebius's sovereign ai enterprise offering: inference → proprietary data → post-training → better inference → richer training data.
Each turn builds more specialized intelligence & sov. compute keeps the full learning loop & data driving it under the customer’s control.
Own--and compound--your intelligence. Build & extend your own frontier. That's the pitch & differentiating power of what it means to sell a spiralling "intelligence platform."
And why the term "neocloud" doesn't do them justice.
nitter.gq/romanchernin/status/21…
8
1,461


















