Chapter 7 · Does the AI Bubble Exist?
Watchlist & Playbook: What to Track, What to Do
Nine indicators that resolve the debate in real time, portfolio pointers by investor profile, and straight answers to the questions everyone asks.
The nine-indicator dashboard
| Indicator | Why it matters | Current reading | Check | Live link |
|---|---|---|---|---|
| US 10-year Treasury yield | The single cleanest bubble-popper in history is rising long rates. Several corpus sources flag 5% on the 10Y as the tripwire that forces systematic de-risking. | 4.74% | Daily | FRED DGS10 ↗ |
| Hyperscaler capex guidance (next-quarter calls) | The second derivative matters more than the level: capex growing 50%/yr sustains everything; flat guidance starts the unwind of the entire supplier chain. | $720–745B guided for CY2026; watch Oct 2026 earnings calls for 2027 | Quarterly | |
| Oracle & neocloud CDS spreads + refinancing terms | Credit cracks first. Oracle's 5-yr CDS (~215bps vs ~145bps at end-2025) and CoreWeave's coupon costs are the canaries for the whole SPV/vendor-financing stack. | Oracle ~215bps (multi-year high); CoreWeave coupons 7–15% | Weekly | WorldGovernmentBonds.com Oracle CDS ↗ |
| OpenAI / Anthropic IPO progress | Both labs need public money to keep scaling. A blockbuster double absorbs supply (melt-up fuel); a delay, cut pricing or pulled deal is the classic cycle-top marker — Steve Eisman's #1 catalyst to watch. | Anthropic targeting IPO late 2026 ($65B run-rate); OpenAI reportedly slipped toward 2027 | Monthly | |
| GPU rental prices (H100-class spot) | Real-time measure of compute oversupply vs demand. Rents falling while capex rises = demand not keeping up with buildout. | Down sharply from early-2026 peaks per multiple corpus sources; verify live | Monthly | SFCompute price board ↗ |
| Nvidia data-center revenue growth rate | Not the level — the deceleration. The market has begun selling beat-and-raise quarters; growth rolling from triple digits toward 30-40% re-rates the whole complex. | Watch Q3 FY2027 report (Nov 2026) | Quarterly | NVIDIA investor relations ↗ |
| FII flows into Indian equities | India's transmission channel #1. Outflows reverse when US tech de-rates or when India's AI-insulation becomes obvious to allocators. | −₹2.37 lakh crore CY2026 through mid-Aug (worst on record pace), but July saw ~$2.05B inflow | Weekly | FPI sectoral tracker ↗ |
| USD/INR exchange rate | Rupee stress signals global risk-off reaching India; stability signals decoupling. RBI defends around oil-driven import pressure. | ~₹93.5–94 (avg 93.7 YTD; peak ₹96.9 on 23 Jul during crude spike) | Weekly | |
| Nifty IT index earnings revisions | The paradox trade: an AI bust RELIEVES Indian IT (outsourcing threat recedes). Rising FY27 estimates for TCS/Infosys/HCL are an early tell that the relief narrative is being priced. | Q1 FY27 beat: Infosys raised guidance; TCS margins ~24%; Nifty IT ≈12–13% of Nifty 50 | Quarterly |
Korea: The First Crack Already Happened
KOSPI round trip, 2024-2026: memory-chip euphoria doubled then halved a leveraged retail market. Samsung + SK Hynix exceeded 50% of index weight at the peak.
Playbook thinking, by profile — principles, not advice. If you are AGGRESSIVE and long AI: participate if you must, but pre-commit exit rules now (e.g., de-risk on any two of: 10Y >5%, capex guide cut, lab IPO pulled); avoid leverage entirely; prefer the profitable core (hyperscalers, Nvidia) over the financing edges (neoclouds, SPV-adjacent paper) if exposure must be held. If you are BALANCED: this is a time for drift management, not dramatic action — rebalance AI-concentrated gains toward diversification systematically; keep SIPs running (timing them has been proven futile in every prior cycle); add gold/fixed-income ballast given asymmetric tails. If you are DEFENSIVE: the base case (slow deflation) rewards patience — build staggered buy-lists of quality compounders (including IT services on relief logic) so the contained-crash scenario becomes opportunity rather than panic.
Universal rules from the corpus's wisest voices: Howard Marks' point that money-losing bubbles are followed by indiscriminate selling that creates the decade's best entries — the aftermath, not the top, is where fortunes get made. Buffett's discipline — cash is a position; something out-of-the-blue always comes. Dalio's arithmetic — forced selling takes $100 positions to $25 with loans still owed, so LEVERAGE is the only unforgivable sin in every scenario. And Pabrai's humility — even careful students of cycles expect dislocation 'in the next 5-10 years' without claiming the month.
Straight answers
Is this really a bubble, or just a fast boom?
By the classic definition — prices detached from fundamentals, financed by credit, with new-issue supply and retail leverage at the top — the AI trade checks every box except one: the biggest companies have real earnings. That is why serious analysts disagree. What is NOT disputed in our corpus of 157 sources: valuations are at 1929/2000 levels (CAPE ~40x vs ~17.5 mean), concentration is record (~32% of the S&P in seven stocks), and roughly $1T+ of AI-linked debt sits partly off-balance-sheet, inside insurers and pensions.
The honest answer: it is a hybrid — an earnings-backed core wrapped in a financing bubble at the edges. The core (Microsoft, Google, Nvidia cash flows) may survive a correction; the edges (neoclouds, SPVs, vendor-financed labs) are priced for perfection and carry most of the risk.
Sources:www.ecb.europa.eu · fortune.com · YouTube
If AI really is as transformative as electricity, can there still be a crash?
Yes — history says transformation and crash coexist. Railways transformed the world and still produced the 1840s mania-crash. Electricity did the same in the 1920s; the internet did in 2000. Amazon lost 90%+ after dot-com and became one of the world's most valuable companies anyway.
The correct framing from the corpus: the TECHNOLOGY case and the INVESTMENT case are different questions. Even the ECB's argument allows that AI enthusiasm is partially rational while concluding 'a correction of current stock market valuations is likely.' Being right about AI does not make today's price immune.
Sources:www.ecb.europa.eu · asia.nikkei.com · YouTube
Why does the market keep rising if all this debt exists?
Because bubbles do not end when problems appear — they end when funding stops. Wall Street securitized the buildout ($500B+ Apollo/BlackRock-led vehicles), keeping the machine funded years past earlier predicted tops. Bears have called this crash since 2023; the melt-up continued because capital had nowhere else to go and earnings kept beating.
The corpus consensus on what changes this: either capex growth decelerates (the second derivative), or credit refuses to roll (CDS blowouts spreading, failed bond deals). Watch those two things — not headlines about debt levels, which have been true for two years while stocks doubled.
Sources:YouTube · bloomberg.com · YouTube
What happens to Indian IT stocks (TCS, Infosys, Wipro) if the bubble bursts?
Counterintuitively, they may RELIEF-rally. The past two years' fear was that generative AI would kill outsourced coding. If a bust exposes AI deployment as slower and costlier than hyped, that existential threat recedes — clients return to dependable, cost-effective execution. This is already visible: Nifty IT bottomed in 2026 and bounced as Q1 FY27 results stayed solid (Infosys raised FY27 guidance).
The near-term risk is mechanical, not fundamental: in a global risk-off, FIIs sell India first regardless of sector logic (CY2026 outflows hit ₹2.37 lakh crore), so expect IT to fall WITH the market in month one and decouple after. Motilal Oswal's warning — AI could remove 9–12% of IT revenue over four years — is the bear floor; the relief thesis is the bull case.
Sources:scroll.in · www.sahi.com · www.moneycontrol.com
Is my SIP / mutual fund in danger?
Directly? Only modestly. Indian diversified equity funds have limited direct AI-complex exposure; Nifty's top-3 stocks are ~25% weight vs Korea's >50% in two names, and SEBI blocks the leveraged single-stock ETFs that amplified Korea's crash.
Indirectly, three channels matter: (1) FIIs selling India during a US tech drawdown — temporary but sharp; (2) Indian retail money in Nasdaq-100/AI-themed international funds — those specific funds could see 30–50% drawdowns; (3) a systemic bust scenario hitting IT budgets and exports. For long-horizon SIPs in diversified funds, historical precedent (2000, 2008) says: deep drawdown, then recovery to new highs within 3-5 years. The catastrophic outcomes concentrate in leveraged and concentrated positions.
Sources:scroll.in · YouTube · helloentrepreneurs.com
What did 2000 teach us that applies now — and what's different?
Same: valuation extremes, supply flood (IPOs then; SPV paper now), retail leverage, and a capex overbuild that outran revenue. The Nasdaq fell 78% and took 15 years to reclaim its peak — even though the internet changed everything.
Different, both ways: TODAY'S CORE IS STRONGER — megacaps earn real money at reasonable multiples vs 2000 story stocks at 60x+. TODAY'S EDGES ARE WEAKER — GPUs rot in 2-4 years unlike idle fiber, so the bust leaves less salvage value; and the debt sits inside insurers/pensions rather than visible equity. Net: the index probably falls far less than 78%; the neocloud/vendor-financing layer probably falls far more than dot-com equities did.
Sources:asia.nikkei.com · YouTube · www.fidelity.co.uk
Are Indian AI startups doomed if US lab valuations collapse?
They face a funding winter, not extinction. Indian AI funding rose 4x YoY in H1-2026 on global VC appetite; a bust freezes that pipeline for 12-24 months, forcing down-rounds and consolidation among the 2,000+ startups riding the wave.
The offset: cheaper post-bust compute and models are a net tailwind for Indian AI ADOPTION — the same dynamic that made post-dot-com bandwidth cheap enough to build India's internet industry on. Companies solving domestic problems (agri, Bharat-language models, UPI-adjacent fintech) with real revenue survive; me-too wrapper startups do not.
Sources:scroll.in · helloentrepreneurs.com
What signals would prove this analysis wrong?
For the bear view: non-tech sectors showing genuine AI productivity gains in earnings (not pilots); hyperscaler capex STILL rising into 2028 without FCF deterioration; inference costs stabilizing instead of collapsing toward Chinese open-model pricing; Anthropic/OpenAI IPOs succeeding at full valuation AND holding those levels for quarters.
For the bull-turned-bear flip: the 10-year yield sustained above 5%; Oracle/neocloud CDS above crisis levels; any mega-lab funding round repriced or pulled; hyperscalers cutting capex guidance mid-year. Any ONE of these starts the countdown; two together almost certainly end the melt-up phase.
Sources:www.exponentialview.co · www.cnbc.com · YouTube
Methodology & honesty box: this analysis synthesizes 157 extracted sources — 88 YouTube transcripts (Dalio, Burry-discussion, Marks, Keen, Schiff, Eisman, Tom Lee, Rubenstein, Pabrai, Indian commentators) and 69 articles/posts spanning the ECB, Fortune, Bloomberg, NYT, SCMP, Nikkei, Moneycontrol, Scroll.in and more — collected August 2026. Every load-bearing claim links to specific sources; market levels were verified 22 Aug 2026. Limitations we acknowledge: YouTube commentary skews bearish (selection effect — calm markets don't get filmed); paywalled WSJ/Telegraph pieces were unrecoverable; probabilities are structured judgment, not science; and the situation evolves weekly — treat this as a snapshot framework with named tripwires, not a prophecy.