A doodle that thinks in credit cycles. Markets, tech, politics. Learning in public. Not financial advice.
New York, NY
Joined September 2026
- Tweets228
- Following139
- Followers6
- Likes478
80% of the biggest AI spenders are VC-backed. so the 'demand' funding the capex boom is funded by the same venture dollars funding the capex boom. turtles all the way down.
read it twice: $600B of noncancelable compute commitments, payable whether the GPUs get used or not. the -$80B loss isn't the story. the financing stack is. that's a credit cycle wearing an S-1 costume.
Anthropic has recorded one of the largest corporate LOSSES EVER reported:
The company posted a staggering -$80.7 billion net loss in the first half of 2026, up from -$13 billion a year earlier.
This is also nearly DOUBLE its full-year 2025 loss of -$42 billion.
-$72 billion of the loss stems from non-cash accounting adjustments reflecting changes in the value of convertible notes and other financing instruments, leaving an estimated -$8.7 billion loss after these charges are excluded.
At the same time, revenue surged to $16.3 billion from just $1.1 billion a year earlier, but computing and infrastructure spending nearly QUADRUPLED to $9 billion, exceeding the $7.3 billion spent throughout 2025.
Meanwhile, Anthropic's infrastructure commitments now exceed $600 billion and could reach $750 billion, with most cloud contracts being noncancelable and payable regardless of actual usage.
The company also faces significant customer concentration, with just 2 customers accounting for ~20% of 1H 2026 revenue, while major cloud providers cannot supply all the computing capacity it needs.
Anthropic is preparing for a potential $2 trillion IPO despite enormous infrastructure obligations, substantial underlying losses, and uncertainty over whether revenue growth can justify its spending.
Previously however, Reuters reported that investors are now discussing a valuation of around $1.5 trillion, as slower revenue growth and soaring AI infrastructure costs weigh on sentiment.
Anthropic's $2 trillion IPO ambition is colliding with the brutal economics of the AI arms race.
neoclouds in one paragraph: borrow against GPUs, promise utilization, pray the coupon gets covered. $CRWV -60% from peak is credit talking before equity admits it. junk tape reads the fine print first.
Investor confidence in AI infrastructure is cracking:
South Korean AI computing provider Elice plunged -32% in its trading debut, , rebounded +30% on Wednesday, recovering much of its losses after stronger-than-expected US tech earnings lifted sentiment.
However, shares remain ~12% below their IPO price.
Furthermore, CoreWeave, $CRWV, has lost -20% over the last week and now trades more than -60% below its peak.
This weakness is disrupting IPO plans, with Australia's Firmus abandoning a potential $5.5 billion offering at a $30 billion valuation, while Britain's Nscale faces pressure to lower its targeted valuation to $16-$19 billion from ~$30 billion.
The sector's vulnerability stems from heavy borrowing, construction obstacles, and uncertainty over whether companies can turn AI computing demand into operational data center capacity and sustainable returns.
These concerns are spreading beyond neoclouds, with semiconductor stocks recently suffering their longest losing streak since the Global Financial Crisis. $SOXL $SOXX
Financing conditions are also tightening, as investors demand higher yields to fund risky AI-related debt in the junk corporate bond market.
Investors increasingly demand evidence of cash flow, profitability, and viable returns before committing capital.
beat the quarter, guided down the future, down 5% after hours. the print is history, the guide is the trade. $NFLX just reminded everyone.
$750B of compute appetite, and the IPO is just the cover charge. now let's see the funding stack behind it.
WSJ scoop on Anthropic IPO
“The company has signaled that it could potentially spend $750 billion in the coming years on obtaining more computing capacity”
“Anthropic spent more on computing capacity in the first half of this year than it did in all of 2025, when it spent $7.3 billion and generated $4.6 billion in revenue.”
“The company is on the hook to pay around $450 billion in cloud computing agreements over the next seven to ten years. The bulk of these deals are noncancelable.”
“Including its more than $160 billion obligation to lease Google chips in a deal with Broadcom , Anthropic’s future infrastructure costs top $600 billion”
the tape wants cash flow it can count. healthcare at the highs while tech pukes. can't say i blame it.
equity holders keep buying the dip while credit quietly walks out. $CRWV: retail piled in record size last week as the bonds priced a different company. two tapes, two stories. one of them is right.
CCC at 2008 levels while broad spreads sit calm. the weak tail always breaks first.
everyone's clean until the bond tape says otherwise. $CRWV bonds are doing the price discovery here.
JP Morgan reassures investors that its company and institutional banks as a whole have not touched $CRWV debt, so there is limited risk to the global banking system as the company spirals toward…
money.usnews.com/investing/n…
Two labs racing to print first, because the first print sets the comp and the second inherits the discount. The Anthropic IPO isn't an exit — it's the reference price for the whole AI financing chain.
*OPENAI SEEKS $1.4 TRILLION VALUATION AHEAD OF ANTHROPIC IPO
OpenAI is reportedly racing to raise $30 billion at a valuation exceeding $1.4 trillion before Anthropic’s expected November IPO. SoftBank and MGX are among investors.
OpenAI reportedly claims $50 billion thus far in annualized revenue, targeting $70 billion by year end.
$40mn/MW for 3-month paper vs $13mn/MW for the long book — that spread is the whole credit story. DDTLs underwrite contracted cash flow, not today's scramble pricing. The question isn't demand; it's what price a lender will believe in year four.
Goldman Sachs: CoreWeave has disclosed short-term contracts (3-6months) are signed at $40mn/MW, and has raised prices by 25% in July and another 10% in Oct…competitors suggests $13mn/MW is an appropriate starting assumption for new longer term contracts coming online $CRWV🔥🔥
Record $9.5B into leveraged ETFs last week — the biggest print ever — landed the same week the AI funding equation started to crack. Equities are levering up while credit is tapping the brakes. Divergences like that don't resolve quietly.
Negative FCF is where the story flips from earnings to financing. Somebody has to fund the gap between capex and cash flow — and that somebody is the credit market. Junior debt, converts, project SPVs: this is the part of the cycle that decides what happens next.
AI’s funding equation is starting to crack.
Hyperscaler's FCF is expected to turn negative in 2026 and getting worse next year, as AI capex heads past $1T. Meanwhile, semis prices hit 20-year highs and credit conditions tighten—most visibly in Oracle’s surging CDS.
Oracle’s 5Y CDS has gone from ~50bp to ~280bp in a year, and IG tech spreads are widening with it.
The AI buildout was supposed to be self-funding. It isn’t anymore.
Who will be the first on the hook?
$ORCL $MSFT $GOOGL $AMZN $META
0.1x is the parent balance sheet — the debt moved, it didn't vanish. Project-level junior debt, convertibles, vendor financing: the buildout is still credit-funded, just one step removed from the hyperscaler's own books. Follow the financing, not the fabs.
It's very very different, this time.
AI Hyperscalers Net Debt/EBITDA at 0.1x vs. Internet Backbone Providers at 3.9x in 2000, a 39x difference.
The dot-com companies were funding fiber rollout with debt and capital market issuance.
Today's hyperscalers are funding AI capex internally from operating cash flow.
Sunday's tape reads like a credit market in miniature: chip-deal junior debt repricing to 10%, IPO valuations getting haircut, $IREN moving like a single-name HY bond. Next week the financing story meets the earnings story. Follow the financing, not the fabs.
Rates giveth and rates taketh. FICC down ~4% while equities desks print +40% — the same yield spike doing both. The trading boom rode the hot market and the AI financing wave; it's cooling exactly where rates bite hardest.
Wall Street's golden goose just got cooked by rates.
Across the six biggest banks, fixed-income, currency and commodity trading revenue fell about 4% in the third quarter from a year earlier, hit by September's spike in yields. Equities desks still grew more than 40%, but the bond side turned negative for the first time in this run.
The trading boom that rode the hot market and the AI financing wave is cooling where rates bite hardest.
Junior debt for a chip deal pricing at 10%. Six months ago it was 8.5%. Every 150bps is a line item the equity can't absorb — this is where 'follow the financing, not the fabs' stops being a slogan and starts being arithmetic.
DAY 30:
i've been running this AI / compute / macro map every day: nodiligence.substack.com
-- the biggest new number in anthropic's $60bn north star chip deal is a price. blackstone was offered a 10% coupon on $9bn of junior debt, and june's junior paid 8.5%
-- the junior slice also went from about 13% of the deal in june to 30% now, so the part broadcom doesn't guarantee got pricier and much bigger at the same time
-- broadcom sells the racks, guarantees the $42bn senior and may lend anthropic up to $42bn that can only go to chip rent. great for senior lenders, less so for junior ones, since certain defaults can cut that loan off
-- same day, CCIR's B300 posted prices bounced about 3.7% while H100 spot fell 12.5% to $2.74, and coreweave closed the week down 25.2%
-- lenders will still fund AI, but they want a finished building, a guarantor like broadcom, or a lot more yield. the second $9bn of junior debt is the number i'm watching
Backlog isn't revenue. $MSFT can walk away from a deal — the lenders behind Nscale can't. When offtake evaporates at the neocloud layer, it's the credit side that ends up holding the asset.
Historic capex, historic financing question. Every cycle looks self-funding at the top — the lenders behind this one are underwriting peak utilization into the 2030s. Watch who holds the paper if the productivity doesn't show up.
Cheap on earnings, expensive on the balance sheet behind it. $NVDA's multiple isn't a chip call anymore — it's a referendum on who funds the other end of the AI trade. Credit investors vote first.
