Stocks are wobbling after a sharp NASDAQ drop, but the core claim is that AI demand is still outrunning supply while the real risk sits in leverage, not in an AI demand collapse. The video splits AI debt into cash-rich hyperscalers, stretched names like ORCL, and highly levered builders like CRWV, then argues CRWV’s bond panic is a financing scare built on short-term rate fears, not a business scare. It backs that up with backlog, customer concentration, earnings expectations, and repeated supply-shortage checks across chips, memory, data centers, and power. A sponsored mining pitch then shifts to Greenland Mines’ GRML, framed as a Western-critical-minerals play on platinum, palladium, and rare earths in Greenland.
NASDAQ hit an all-time high on Oct. 6 at about 27,600, then closed Friday near 26,140, down 4.5% on the week and just over 5% from that high; the S&P 500 is about 2% off its high.
A hotter-than-expected inflation report, Brent crude touching $110, and fears of more Fed hikes are framed as the macro mix hitting growth stocks hardest because long-duration earnings get discounted more when rates rise.
Big hyperscalers — MSFT, GOOGL, AMZN, and META — funded early AI buildout mostly from cash, but the five biggest hyperscalers sold $121 billion of bonds in 2025 versus about $28 billion annually from 2020-2024, and by early Aug. 2026 they had already sold about $220 billion; Goldman estimated debt now funds roughly a third of their spending this year.
CRWV’s debt stack is about $35.6 billion in debt plus $16.3 billion in lease obligations, roughly $52 billion total; in the first half of this year it had $4.6 billion in revenue and a $1.4 billion net loss, and in Q1 its top two customers made up about 65% of revenue.
CRWV’s unsecured 2032 notes dropped below 80 cents on the dollar this week, its 2031 bonds yield almost 16%, credit default pricing hit a record, and Bloomberg-implied default odds are better than 60% within 5 years.
CRWV stock fell 12% Friday and about 25% on the week to 61.39, roughly 57% below its 52-week high of 141.74; Elon Musk responded “Yikes.”
CRWV says it has about $104 billion of signed backlog, with about 40% expected within 2 years and many contracts structured as take-or-pay; analysts still expect adjusted EBITDA to rise from 7.6 billion this year to almost 17 billion next year.
Jensen Huang told Bloomberg CRWV’s finances are improving, borrowing costs are coming down, and NVDA does not need to backstop the company; the original bonds still carry about a 9.5% coupon, while the market’s near-15% pricing reflects what new debt would cost today.
The AI debt stack is split into three tiers: tier one cash machines (MSFT rated AAA; MSFT, GOOGL, AMZN, META borrowing by choice), tier two the stretched (ORCL, BBB/negative outlook, Sept. force majeure notice on a data-center lease, 2055 bonds around 77 cents, stock down about 59% from last October’s high), and tier three the builders (CRWV, APLD, NBIS) that use debt to build faster and are most rate-sensitive.
The bearish Shiller-PE argument says the Nasdaq 100’s Shiller PE near 59 is above 98% of readings since 2002, but the counter is that the 10-year average earnings denominator lags when profits explode, especially for names like NVDA.
The bull case cites S&P 500 earnings up about 29% this quarter, the S&P 500 at about 19x next year’s earnings, and the Nasdaq 100 at about 21-23x next year’s earnings; Brad Gerstner is cited saying Wall Street’s NVDA earnings forecast rose 100%-200% in a year while the stock is only up about 20%, with NVDA at about 13-14x earnings.
Jensen Huang is quoted saying it is a mistake to think only a couple of companies fund AI infrastructure; he says big AI labs are only a quarter to a third of NVDA’s business, another third is regular computing, and the last third is thousands of startups renting NVDA chips.
The supply-shortage case runs through TSMC revenue up 41% year over year with AI demand still above supply, MSFT saying Azure demand exceeds capacity with no end date, MU saying most of its 2027 AI-memory supply is already spoken for, and one index showing H100 rental prices up about 19% since May.
Bears are called too early on AI because 2024 claims that AI had hit a wall, Jan. 2025 “deep seek” would end American AI, and NVDA would be done all failed; NVDA fell 17% in one day around that Jan. 2025 scare and is up about 81% since.
Google’s token usage is cited as rising from about 10 trillion tokens a month in May 2024 to 480 trillion a year later and over 3.2 quadrillion in this May, while Sam Altman said cheaper compute would expand use more than proportionally; agentic AI is said to raise compute per request because one query can spawn multiple AI workers.
The sponsored segment pitches GRML as a Greenland critical-minerals play: Kvanefjeld for platinum/palladium, Tanbreez for rare earths and NDPR magnets, backed by the Sept. 22 US-Denmark-Greenland agreement and framed around Western supply-chain security.
Skaergaard has about 15 million palladium-equivalent ounces indicated and 17.5 million inferred after 45,000 m of prior drilling plus 4,480 m more in 2026 and more than 100 tons of bulk material for metallurgical testing; Sarfartoq has 6.9 million tons indicated at 1.6% TREO and 5.3 million inferred at 0.96%, with a high-case initial assessment of about $2.05 billion pre-tax NPV at 8% and a 118.6% IRR, though both projects are resources, not reserves.
GRML recently raised more than $59 million at $12 a share, says the money funds more metallurgy, drilling, engineering, and development, but the pitch also flags dilution risk because Greenland mining is remote, expensive, and likely to require repeated equity raises before any production.