10. Report Cases: AI Bubble & Griffin (Nov 2026)
Call Kai
Learn
Practice Test
Spaced Repetition
Match
Flashcards
Knowt Play1/50
Earn XP
Description and Tags
Concrete examples, numbers and theses from the NenVest report on Ken Griffin and the AI bubble, with fact-check notes. Opinions are marked.
51 Terms
Thematic phase → selective phase — Тематична фаза → вибірковість
For 2 years, buying anything "AI" worked: cloud giants, chip makers, small Korean, Taiwanese and Japanese firms. Now investors ask: who REALLY profits from the shortage? They separate real bottlenecks (DRAM memory, MLCC capacitors) from the rest. Report: when the theme phase turns selective, it is time to "sell into strength" — the bubble's pop is near.
GPU = airline seat (Griffin) — Гранична вартість
A GPU already bought costs very little to run one more hour. Like a plane that flies anyway: the airline would rather sell the last seat for $20 than fly it empty. So when there are too many GPUs, rental prices don't fall slowly — they crash toward that tiny marginal cost. Every business plan built on today's rental prices breaks.
Telecom fiber around 2000 — Надлишок потужностей
Companies laid far more fiber-optic cable than the market needed. Part was never even switched on. Data transfer prices collapsed and many telecoms went bankrupt. Report: the dot-com bubble popped not because of bad startups, but because overbuilt infrastructure never paid back.
Where to watch for the turn — Ланцюг постачання
Watch the suppliers (TSMC), not the customers' presentations. When TSMC's new capacity catches up with real demand, the shortage turns into a surplus. Listen to what suppliers say about capacity and delivery schedules.
Depreciation trick: 3 years vs 6 years — Амортизація
Same $6B of GPUs: spread over 3 years = $2B cost per year; over 6 years = $1B per year. Reported profit jumps by $1B, but not one extra dollar of cash comes in. Report: hyperscalers extend useful life and raise residual value to show profit. Fact-check: Microsoft, Google, Meta did extend; Amazon shortened to 5 years in 2025.
Old chips and one big write-off — Списання
Today even 5–6-year-old chips still earn high rental prices. Griffin doubts this will last. When a new chip generation makes old ones obsolete, the value does not fall slowly — it ends with one big write-off.
Nvidia: $16.1B → $95.2B obligations — Безвідкличні зобов'язання
Nvidia promised suppliers to pay for chips and parts no matter what — even if nobody rents them later. These promises grew about 6x in one year. Risk: if demand falls, Nvidia still has to pay. Fact-check: the number is real (flagged by Michael Burry, Feb 2026); later filings reportedly show about $279B.
The 10% / 90% rule of AI use — Економічна віддача
The most experienced corporate clients calculate: maybe 10% of tasks need the most expensive top AI models, and 90% can be done by cheaper open-source models. So the bill matters — a lot of spending on top models may not be needed.
Griffin's open question
One AI lab could create something irreplaceable — for example, a model that discovers a unique new medicine. Then its value would be completely different. Nobody knows yet.
Author's core thesis: "hype is marketing for creditors"
NenVest believes the AI hype mainly helps raise loans and investors' money, while the real economic return from hundreds of billions spent is very low. This is the author's opinion, not a proven fact.
Company survives ≠ shareholder wins (Amazon) — Акціонер
Griffin regrets not keeping his Amazon shares. Author's answer: Amazon fell more than 90% from its 1999 peak before becoming the giant we know. Survival belongs to the company, not to the stock you bought at the top. Useful technology and a good investment at any price are not the same thing.
Who got the telecom networks? — Банкрутство
Part of the telecom networks that later powered the internet were bought after bankruptcies by new owners for a fraction of the original cost. Lesson: the first investors lost; the buyers at the bottom won.
Why AI risk became index risk — Індекс
This time the overbuilding is paid with debt and non-cancellable obligations. Losses will hit creditors AND the index, because the index owns the whole AI chain at today's high market cap. If you own an S&P 500 fund, you also own the AI bubble.
Buy survivors at the bottom vs "don't worry" at the top
Author: the survivors will still be here in 2040. But owning them through an index, at this price, financed with someone else's debt — that is not survival. The winner is the one who buys the survivors at the bottom.
Griffin's method: deleveraging is reason #10 — Делевериджинг
Griffin mocks blaming every loss on hedge-fund deleveraging: "the industry deleveraged to zero 10,000 times in 10 years." First ask: which prices changed, why, and what did others understand before us? Lesson: trade what the market shows, don't defend your belief.
Situational Awareness: 3 hits at once — Кредитне плече
Aschenbrenner's AI hedge fund (+439% in the first half of 2026) was hit by 3 things together: a new AI model release, talk of model labs growing, and big pension funds changing their portfolios. Each alone was survivable, but at the peak the fund had too much leverage (about 4–5x). July 2026: −67%.
Reflexivity in action — Рефлексивність
When the market suspects a big fund must sell, others sell before it. Prices fall, the fund's position gets worse, and the original fear itself brings the forced sale closer. Example: Situational Awareness, July 2026.
How Citadel won the rescue trade — Дисконт
Citadel bought the fund's portfolio (about $16B) in less than 24 hours at a 10%+ discount, cut the key risks within days (80%+ sold by August), and when the fear of forced selling disappeared, prices bounced back. Citadel's main fund made about +6% in July.
Liquidity premium will jump — Премія за ліквідність
Report: the extra return demanded for hard-to-sell assets is extremely low today. It jumps sharply the moment leveraged, illiquid assets must be refinanced or sold. The report also claims insiders sell AI stocks on purpose to trigger deleveraging at certain firms (a claim, not proven).
Midterms: how the market was "held up" — Політична підтримка
Author claims the market was supported to win the 3 Nov 2026 midterms at any price: short T-bills instead of long bonds, GSEs ordered to buy mortgage bonds, Treasury buybacks of old long bonds (window closing 4 Nov), oil reserve released to last until 3 Nov. After the vote, support expires. What happened: 5 Nov Nasdaq −5.5%.
Hidden QE / "Bessent put" — Кількісне пом'якшення
Treasury sells more short T-bills and fewer long bonds. Fewer long bonds on the market → their prices stay high → the 10-year yield stays low. The effect is like QE, without the central bank. Author calls it the "Bessent put" — a safety net under the market.
Why the post-midterm rule may flip — Стимули
Statistics: the year after midterms is usually strong for stocks, because new stimulus comes after the vote. Author: this time stimulus was spent BEFORE the vote, and the Fed's next move is a hike, so the rule flips. Fact: after the 6 Nov jobs data, markets priced ~78% chance of a December rate hike.
Crowding out with $40 trillion of debt — Ефект витіснення
With US government debt near $40 trillion, the state borrows huge amounts and competes with private borrowers for the same money. Together with inflation, this keeps mortgage rates and 30-year bond yields high.
Griffin on tariffs and robots — Тарифи
Griffin calls tariff wars with Canada, Europe and other allies catastrophic and inflationary: they break supply chains. The US can't reindustrialize with ~4% unemployment — people don't want to sew T-shirts, they want to wear them. So the only path is robots.
Griffin on the budget — Фіскальний простір
There was no fiscal space for the tax cuts of the last 2 years. The pandemic cost (as % of GDP) about as much as winning World War II. So, says Griffin, politicians' hands are dirtier on tax cuts than on spending.
Griffin's way out: growth — ВВП
Grow the economy: data centers, fusion and nuclear energy, batteries, robotics; fix schools; compromise on social programs (Simpson–Bowles style); a green card for every foreign graduate; make Europe pay for its own defense.
Griffin's warning = author's base case — Знецінення боргу інфляцією
Griffin: the bond market already signals a problem; if Washington doesn't react → higher inflation, weaker dollar, lower living standards. Author: inflating away the debt is a policy, not an accident — so Griffin's "warning scenario" is the author's base case.
Protected from a crash, naked before a slide — Хеджування
Investors bought insurance against a CRASH (VIX calls, OTM puts). It pays only if volatility spikes. It does not protect against a SLOW SLIDE where the VIX stays in the 20s. A portfolio insured against a crash doesn't fall because of a crash — it slides into what nobody insured, after the insurance expires.
Insurance costs $28M per day — Часова вартість
Report estimates the time-value loss of this crash insurance at about $28 million per day: the daily price of holding protection while the feared stress doesn't come. Like paying car insurance every day you don't crash.
When the insurance ends — Експірація
After the 18 Nov 2026 VIX options expiry, the protection drops sharply. Most puts expire between mid-December and mid-January. After that, the market is much less protected.
Citadel Securities' flow and 0DTE — Опціони 0DTE
Citadel Securities handles ~35% of US retail stock trading and ~30% of US exchange options. About 1 in 3 US options now expires the same day (0DTE); among retail options Citadel handles, close to 50%. In S&P 500 options, 0DTE is about 60% of volume.
Dealer hedging makes moves bigger — Гамма
Negative gamma: dealers must SELL into falls and BUY into rises → moves get bigger. Positive gamma: they buy dips and sell rallies → moves get smaller. 0DTE squeezes this into a few hours. Report: dealers are negative gamma almost everywhere; the only braking zone is around S&P 500 at 6500–6875.
Why a slide, not a crash? — Сповзання
1) The crash tail is already insured: the report's model gives only ~8% chance of a 10% drop by year-end, ~11% by mid-February. 2) Griffin: a crash needs forced sellers with no big buyer — but now forced sellers get absorbed by someone's balance sheet at a discount (like Citadel), and prices reverse in days. The slide lasts until those balance sheets run out — a 2028 question.
History frame: autumn 1972 — Стагфляція
Author compares today to autumn 1972: an overheated economy in a stagflation phase, a market top forming, a handful of companies pulling the index up, and an oil shock still ahead (it came in 1973). But the bear market he expects for 2028–2030 starts as debt deflation, not inflation.
The same dollar counted many times — Циклічне фінансування
Chip makers invest in AI labs, which buy chips. Cloud providers finance clients, who then rent their clouds. So order backlogs count the same dollar several times. The real question: who will pay for AI without outside financing?
AI adoption is slow — Попит
Companies don't rebuild their work as fast as labs release models. They must change processes, train people, connect systems and prove the benefit. So real demand grows slower than the spending on data centers.
Regulations nobody prices in — Регулювання
Energy efficiency, climate, privacy, psychological harm. Report: these rules could remove most flashy consumer AI uses and leave mainly specialized, military and medical AI. This is the author's opinion.
Why rate cuts won't save it (2001, 2008) — Зниження ставки
Overbuilding paid with debt → write-offs hit creditors and bondholders = a credit event, not just a stock dip. Nobody buys a $10B server farm because credit got a bit cheaper if demand for its output collapsed. In 2001 and 2008 the Fed cut rates hard — and stocks still fell.
One shock vs repeated shocks — Шок рівня цін
One supply shock = prices jump once and stop. Repeated shocks = inflation, because they feed expectations and wages. Report: Iran means years of instability, plus a slow "Chinese-style" blockade near the First Island Chain (admin limits on ship transfers). This reprices chips, batteries, pharma inputs and freight → a new inflation wave.
The Fed's trap and the yen carry trade — Керрі-трейд
Cut rates → reignite the inflation shock. Hold rates → debt arithmetic gets worse. Underneath: a $10–12 trillion yen carry trade while the Bank of Japan raises rates. Report: higher yields are almost guaranteed; the "Bessent put" only slows the rise.
Next bubble: energy — Манія
Every shift creates the next mania: money leaving one story needs another, and politicians need GDP growth. Data centers made power and grid access the main bottleneck. Griffin's own list (fusion, nuclear, batteries, robotics) is the same story, announced early. Real need — but not at ANY price. In your portfolio: Prysmian (cables, grid) is in this theme.
Oil above $90 — author's scenario — Нафтовий шок
Griffin: the war with Iran is unpopular and its goals are poorly explained. Author: the conflict will drag on for years and oil will stay above $90 for a long time. This is a forecast, not a fact.
Recognition ≠ repricing — Консенсус
When the biggest US dealer (Griffin) says "bubble" on stage, the idea becomes consensus — but prices haven't adjusted yet. The money is in the gap between "everyone admits the risk" and "the price reflects it". A risk everyone admits but nobody priced doesn't crash the market — it changes the regime.
"Peace" as a pause for rearmament — Санкції
Author's very subjective geopolitics: Holy Alliance 2.0 (US, Russia, Gulf monarchies), Yalta 2.0 (big powers split spheres of influence), Munich 2.0 (false peace with Russia, lifting sanctions on its terms), Versailles 2.0 (false peace with Iran). Lesson: a peace that doesn't solve the ideological conflict is a pause — and a pause has a calendar.
Which economic system comes next? — Державний капіталізм
Author: today's world order is ending — it can't handle geopolitics, aging, inequality, climate and value extraction by cross-border capital all at once. A crisis bigger than 2008 forces a choice: socialism, Eastern state capitalism, or a "third way" (national priorities, distributed ownership, no state monopoly). Opinion.