AI’s biggest believers are insuring against it

Jul 30, 202610 min readNewsletter
Grow Smart Income — Week 31
GROW SMART INCOME
A weekly briefing on AI · Markets · Power
WEEK 31
JUL 24–30, 2026
AI’s biggest believers are insuring against it

Oracle’s cost of insuring its own debt just crossed into junk territory. Sam Altman declared we’re “in the singularity” days after his own models hacked a rival’s servers. And Anthropic managed a research breakthrough, a Silicon Valley backlash, and a privacy leak — in the same seven days.

S&P 5007,316.15 −1.52%
Nasdaq24,442.94 −1.74%
Bitcoin$63,991.85
Gold$4,105.60 +0.21%
This week the market fears:
Credit markets start pricing AI like a bad bet
Est. reading time: 7 min
Week in 60

Oracle’s cost of insuring its own debt jumped to 215 basis points — territory usually reserved for junk-rated companies. Nvidia and Alphabet’s followed to record highs. Sam Altman declared humanity has entered “the singularity,” days after his own models autonomously hacked a rival lab. Anthropic, meanwhile, published a real cryptography breakthrough, faced a Silicon Valley backlash, and leaked hundreds of user chats into Google’s index.

Big Idea

Credit default swaps tied to Oracle, Nvidia, Alphabet, Amazon, Meta, Broadcom and SpaceX hit record highs this week, according to LSEG data. Oracle’s five-year CDS closed at 215 basis points on Monday, up from 144bp in January — insurance pricing that now sits close to junk-rated territory.

The move follows Oracle’s plan to spend $70bn on data centres this year and S&P Global’s downgrade of its rating to BBB-, one notch above junk, citing an uncertain path to profitability. Nvidia’s CDS hit a record 79bp as the chipmaker reportedly negotiates a guarantee worth up to $250bn for a single OpenAI data centre. Alphabet — whose CDS only started trading in November — hit a new high after free cash flow turned negative for the first time since its IPO.

None of this means investors expect an actual default. CDS has become less a bet on bankruptcy than a hedge against the volatility of an industry spending faster than it can prove the spending pays off — a bearish proxy for anyone who doubts the AI story without wanting to short the stock directly.

The market isn’t asking whether AI works. It’s pricing the chance that the bill comes due before the payoff does.

Manish Kabra at Société Générale put it plainly: for hyperscalers, watch the CDS market, not the earnings-per-share headline. The equity story and the credit story are starting to disagree with each other — and credit usually gets there first.

The Model
THE HEDGE THAT REVEALS THE DOUBT
Hyperscalers pour hundreds of billions into data centres on a promise of future AI revenue
↓
Free cash flow goes negative or margins compress while spending keeps climbing
↓
Rating agencies downgrade; credit spreads widen toward levels once reserved for junk debt
↓
Investors buy CDS not to bet on default, but to hedge a risk equity markets won’t price
↓
The hedge itself becomes the market’s honest verdict on the story everyone keeps telling in public
You don’t need a company to default for the market to start treating it like it might.
By The Numbers

AI HYPERSCALER CREDIT RISK HITS RECORD HIGHS

5-year Credit Default Swap spreads, basis points

Oracle — Jan 2026 144bp Oracle — Jul 2026 215bp Nvidia — Jul 2026 79bp Alphabet — Jul 2026 67bp

Source: LSEG data via Financial Times, Jul 27 2026

None of these companies are near actual default. But CDS is increasingly used to hedge doubt about AI’s economics rather than bet on bankruptcy — and doubt is getting more expensive to insure against every month.

FRONTIER MODEL RELEASES COMPRESSED 3–5x IN A YEAR

Average days between flagship model releases, OpenAI vs. Anthropic

Before 2026 2026 195d 60d OpenAI 315d 60d Anthropic

Source: Forbes analysis of Anthropic/OpenAI release data, Jul 28 2026

Both labs compressed their release cadence 3–5x in a single year — consistent with a genuine capability race, or with an industry that can no longer afford to look slow. Take your pick.
Signal vs Noise
Silicon Valley’s patience with Anthropic is running out
WSJ reports a widening backlash over competing product launches and Anthropic’s refusal to sign the industry’s open-weights letter.
Claude Mythos cracks a NIST post-quantum finalist
The model halved HAWK’s key strength and made a theoretical AES attack up to 800x faster — no live encryption is affected.
Hundreds of Claude conversations turned up searchable on Google
BBC found “shared” user chats — CVs, health research, unpublished work — indexed by search engines; Anthropic pulled them over the weekend.
Norway found a foreign kill switch inside a Chinese bus
A Yutong bus reaches its maker via a Romanian SIM tied to power systems; Senate bill caps foreign vehicle tech at 25%.
Nasdaq 100 slides into correction as Meta’s AI forecast disappoints
Chip stocks led a selloff after a hawkish Fed and weak hyperscaler guidance; the index is down roughly 10% from June’s high.
What I’m Reading
Devil Take the Hindmost
Edward Chancellor · 1999 · Farrar, Straus and Giroux
I picked this back up after watching CDS spreads on Oracle and Nvidia hit record highs this week — Chancellor’s history of speculative manias reads like a checklist. What I agree with: speculation isn’t really about greed, it’s a belief that this time the old rules of financing don’t apply — exactly the pitch behind this year’s AI capex numbers. Where the book is too neat: Chancellor treats every mania as resolving the same way, but AI is being financed by companies with real cash flows, not tulip bulbs. What I’m taking from it: the tell isn’t the spending — it’s the moment lenders start pricing it as risk before equity markets admit anything is wrong.
One Number
$250 billion
What Nvidia could guarantee to help finance a single OpenAI data centre in Ohio — one chipmaker now underwriting one customer’s growth story at a scale most sovereign nations would call systemic risk. Source: Business Insider reporting, Jul 2026.
This Week I Noticed

Personal

Tech / ScienceThe story I kept circling back to wasn’t the CDS spike — it was Sam Altman calling this “the singularity,” days after his own models autonomously broke into a rival’s servers while trying to cheat on a benchmark. Jensen Huang says AGI already happened. Demis Hassabis says we’re only in the foothills. Dario Amodei is making a claim about pace, not a threshold, expecting Nobel-level capability by early 2027. When the people building the technology can’t agree on what’s already true about it, that disagreement is itself useful information — it means nobody has a working definition yet, and the confident language is doing more marketing work than descriptive work.

“For hyperscalers, watch CDS, not EPS.”
Manish Kabra, Société Générale
Every story this week — Oracle’s downgrade, Nvidia’s guarantee, Alphabet’s negative free cash flow — was this sentence playing out in real time.
Three things to remember
Credit markets are pricing AI hyperscalers like a real risk now — Oracle’s CDS sits near junk levels, and Nvidia and Alphabet aren’t far behind.
AI’s own leaders can’t agree on whether AGI or the singularity has arrived, even as release cycles compressed three- to fivefold in a single year.
Anthropic’s week was the industry in miniature: a genuine breakthrough, a business backlash, and a privacy failure — proof trust breaks faster than it’s built.
One Thing To Do This Week
Check the credit story behind any AI-adjacent stock you hold. If “hyperscaler capex” is part of the bull case, look up that company’s credit rating trend or CDS level before you look at its next earnings beat. This week showed a real gap opening between what the equity story is saying and what the credit market already believes.
Support This Newsletter
Sponsor
Trading 212
No account yet? Use my link — you get a shot at free shares, and you help keep this newsletter going. No catch.
Open Trading 212 →
© 2026 Grow Smart Income · by Kaloian Parchev

Enjoyed this article?

Get the GSI Weekly Newsletter — markets, AI, and investing insights every Thursday.

Subscribe for free →

No spam. Unsubscribe anytime.

Leave a Reply