THIS WEEK · 10 PODCASTS · WEEK OF JULY 20, 2026
This week's ten podcasts converge on a single fault line: the AI capital cycle that is either the greatest misallocation in market history or the engine of a global spending super-cycle — now colliding with a real oil shock radiating out of the Persian Gulf and a bond bear market that refuses to break. Jim Chanos delivers the definitive structural indictment: a boom in which nothing dependent on Nvidia should out-value Nvidia, long-duration data centers are underwritten on spot prices, and "construction in progress" is hiding the depreciation that will reverse the marks. George Noble — himself the guest this week — frames it as a bust seventeen times worse than dot-com, argues semiconductors are shipping stocks whose peak earnings must never be capitalized, and lays out the rotation: short tech, long energy, own gold. Ed Zitron zeroes in on the keystone, Oracle, downgraded to one notch above junk with S&P naming OpenAI as the reason, its core flatlining while its only growth destroys margin and Larry Ellison's margin loans threaten a doom loop. Fred Hickey maps the earnings bubble beneath the price bubble — strip the one-time markups and supplier gains and S&P growth is single-digit — as Chinese models take nearly half of usage and collapse pricing. Scott Galloway and Ed Elson catalogue OpenAI's hell week and predict a management coup within six months, while warning that the market's "broadening" is an illusion — just new ways to buy AI. Louis-Vincent Gave supplies the China thesis: Beijing is commoditizing both the models and the chips, has set the oil range at $65–100, and — the real lesson of the Iran war — has exposed that the United States no longer controls the world's sea lanes. Peter Boockvar anchors the macro backdrop: a global bond bear market driven by real rates and deficits, three-to-four percent as the new inflation normal, and a dollar now hostage to the tech trade. Philip Pilkington brings the boots-on-the-ground oil shock — Bab al-Mandab and the Strait together are a fifth of world supply, the SPR is draining toward the midterms, and physical rationing, not price, may clear the market. Robert Pape supplies the geopolitical backbone: the most wicked escalation trap he has modeled, with the Strait no more than a third open before January and oil inventories counting down toward exhaustion. And Tavi Costa closes with the monetary endgame — "none of us own enough cash" if the path of least resistance is to inflate our way out — prescribing gold, silver, copper, quality miners, and a rotation out of a dangerously concentrated US market. Each summary is designed to be immediately actionable — whether you are allocating capital, running a business, or simply trying to understand the forces reshaping the world around you.
THIS WEEK'S LINEUP
EP 1 The AI Bubble Is "Much Worse" Than Dot-Com — Jim Chanos — Chanos & Co Founder & President — Watch Full Video
EP 2 17x Worse Than 2000: The Biggest Bubble Unwinds — George Noble — Noble Capital Advisors Managing Partner — Watch Full Video
EP 3 Oracle at One Notch Above Junk — Ed Zitron — EZPR Founder & CEO, "Better Offline" Host — Watch Full Video
EP 4 A.I.-Mageddon: The Earnings Bubble Nobody Is Pricing — Fred Hickey — "The High-Tech Strategist" Editor — Watch Full Video
EP 5 OpenAI Is Spinning Out of Sam Altman's Control — Scott Galloway & Ed Elson — "Prof G Markets" Hosts — Watch Full Video
EP 6 How China's AI Strategy Kills U.S. Tech Margins — Louis-Vincent Gave — Gavekal Research Co-Founder & CEO — Watch Full Video
EP 7 A Bond Bear Market Meets the Greatest Financial Bubble — Peter Boockvar — One Point BFG Wealth Partners CIO, "The Boock Report" Author — Watch Full Video
EP 8 Bab al-Mandab and the SPR Countdown — Philip Pilkington — Macroeconomist & Author — Watch Full Video
EP 9 The Escalation Trap: Hormuz Stays Half-Shut to January — Robert Pape — University of Chicago Professor, "The Escalation Trap" Author — Watch Full Video
EP 10 You Don't Own Enough Cash for What's About to Happen — Tavi Costa — Azuria Capital Founder & CEO — Watch Full Video
Full summaries with actionable insights and investment focus for each podcast follow on the pages below.
EP 1 - The AI Bubble Is "Much Worse" Than Dot-Com
Jim Chanos — Chanos & Co Founder & President
Jim Chanos, in conversation with Dan Nathan and Guy Adami on the Risk Reversal podcast, argues the AI buildout is a far larger and more dangerous version of the late-1990s telecom bubble — not because the technology fails, but because the capital being committed dwarfs anything from that era and is being underwritten on the wrong economics. Running a hedged model portfolio of forty names, Chanos says the short side has generated real alpha through violent single-stock dispersion, and that the arrival of heavy issuance now completes the classic top: rich valuations, retail speculation, and a torrent of supply. His unifying frame is that everything in AI is being priced as if it has already worked.
Actionable Bullet Points
The Rotation Under the Surface Is Violent, and Issuance Is the New Third Leg: Chanos points to a single recording-day tape where the SOX fell 3% and a DRAM ETF was down 10% while the hyperscalers rose three to three-and-a-half percent, and to well-followed large caps moving 5–10% a day on flows and sentiment rather than news (3:31), (4:29). The missing ingredient of this cycle has now arrived: record IPOs, secondaries, and insider selling joining already-rich valuations and retail speculation to form a third leg of the stool (5:34). He likens it to the first half of 2021, when shorts began working mid-year after GameStop — and says 2026 feels the same in his numbers (6:13). Treat heavy supply, not valuation alone, as the signal that the regime has turned.
Nothing That Depends on Nvidia to Exist Should Trade Above Nvidia, Yet Many Do: Chanos's core valuation discipline is that a handful of companies are the gatekeepers and price-makers of this boom, Nvidia chief among them, and no company dependent on Nvidia's chips to exist should carry a higher multiple than Nvidia itself (20:57). He singles out Micron — a business with negative gross margins three years ago (19:10) that ran from roughly $100 to $1,250 and back to $950 — and warns against awarding market cap today for long-term agreements and backlog that assume 80% gross margins hold (19:25). Anchor valuations to the chokepoint, not the derivatives that depend on it.
Committing Twenty-Year Capital on Spot Prices Is the Central Financial Sin: The "finance 101 crime" Chanos sees everywhere is people making decisions on long-term projects based on near-term spot pricing — digging a hole, signing a one-or-two-year rental at a 20–25% return, and building a twenty-year asset against it (16:09). He reaches for the shale patch and nineteenth-century railroads, where tremendous incentives to build double and triple capacity collapsed when demand ebbed and rates fell, and freight lines went bankrupt (17:22). Distrust any return underwritten on spot economics for a long-duration asset (18:13).
"Construction in Progress" Is Hiding the Depreciation, and the Marks Reverse When the Clock Starts: Chanos has been harping on a subaccount of property, plant and equipment — construction in progress — that keeps swelling with chips not yet plugged in and capitalized interest and labor the accountants let you defer before any depreciable schedule begins (23:55). Because the clock may not start for eighteen months, a chip on a five-or-six-year schedule effectively runs far longer; Chanos uses ten years in his own models to be conservative and still cannot pencil out most data-center economics (26:00). Watch the depreciation ramp as the mechanism that turns paper earnings into real losses.
Incremental Returns Are Halving and the Builders Are Quietly Heading for the Exits: The metric Chanos watches most closely for clients is return on incremental invested capital for the hyperscalers, which has fallen from roughly 40% eighteen months ago to about 20% and, if spend keeps up, is heading toward 10% (34:56) — with Oracle already the worst on that measure (36:42). Meanwhile Nebius unveiled an "asset-light" pivot after spending $4.40 of capital per $1 of revenue, Coreweave is reportedly hedging asset prices, and the legacy data-center owners have all suddenly put portfolios up for sale (27:56), (29:40). Layer on Nvidia's circular investments that let investees raise more debt and equity, echoing Lucent and Nortel vendor financing (40:08), and track ROIIC and asset-shedding as the crack.
Investment Focus
Chanos delivers the week's most rigorous structural bear case, spanning valuation, accounting, and financing at once. The investment template: (1) treat issuance — IPOs, secondaries, insider selling, and off-balance-sheet debt — not multiples, as the timing signal that the top is forming (5:34); (2) anchor to Nvidia as the valuation ceiling and regard anything trading above it as suspect (20:57); (3) watch hyperscaler return on incremental invested capital falling toward 10%, with the neoclouds and Oracle as the fracture points (36:42); (4) read the sudden asset-light pivots and portfolio sales as builders quietly discounting their own maintenance-capex math (27:56); (5) frame the whole complex as promises-over-reality late-cycle — everything priced as if it has worked, valued as if demand is infinite — where power is not the scarce input (about 5–6% of data-center revenues) and funding appetite is (45:39), (32:56).
EP 2 - 17x Worse Than 2000: The Biggest Bubble Unwinds
George Noble — Noble Capital Advisors Managing Partner
George Noble, in conversation with David Lin, argues we are watching the unwinding of one of the biggest bubbles in history — a tech bust he believes is far worse than dot-com because it is more asset-intensive and larger relative to the economy. His central analytical move is to treat semiconductors as commodity shipping stocks whose peak earnings can never be capitalized, and to separate the tape from the positioning: prices have fallen 30–40% while flows have not, leaving stale bulls stuck. From there he lays out the actionable rotation — short tech, long energy, own gold — and a scathing read on the mechanics propping up SpaceX and Tesla.
Actionable Bullet Points
This Tech Bust Is Seventeen Times Worse Than Dot-Com, and It's the E, Not the P: Citing Julian Garran of MacroStrategy Partners, Noble puts the malinvestment at seventeen times the dot-com era, more asset-intensive and larger relative to the economy, so the implosion carries a real economic impact (12:19), (12:39). The bubble, he insists, is in the earnings: gross margins have run from 25% toward 75% against a long-term average near 25%, and those excess returns get arbitraged away as capacity comes online (10:06). Do not capitalize a cyclical's peak earnings as if they are permanent.
Semiconductors Are Shipping Stocks, and You Never Capitalize a Cyclical's Peak: Noble's original analogy is a VLCC tanker company whose day-rate spikes from $30,000 to $100,000 on fixed costs, briefly printing enormous margins you would never put a price-to-earnings multiple on — you value it on asset and book value instead (8:02). Yet since the June semiconductor peak, roughly $25 billion has flowed into the three leading semi ETFs even as prices fell, leaving positioning wrong while the tape corrected (6:49). The simple institutional move, he says, is to swap SPY for RSP to cut the tech weighting (10:52); avoid capitalizing one-off earnings.
Oracle Is the Next Cisco, Chained to an OpenAI That Goes Bankrupt: Noble argues Oracle can go bankrupt at this rate given a gross misallocation of capital and the extraordinary debt it has taken on, inextricably linked to an OpenAI he believes will itself go bankrupt (36:24). It is a company he and his team called as a short last year, and his rule is blunt: whenever you see the combination of hubris and debt, run, don't walk (9:46). Treat the hubris-plus-debt signature as a reason to exit, not to bottom-fish.
SpaceX Is an AI Play With a Rocket Attached, Set Up to Fail Near $30: Noble strips the story to the TAM — roughly $29 trillion claimed, of which about $22 trillion is AI — and argues the rocket is almost irrelevant to the valuation (30:47). The float was listed at only about 5% before a cascade of unlocks toward 100%, and index inclusion (Footsie Russell changed its rules where the S&P did not) forced passive buying so that "grandma's 401k" now owns it; the stock has already broken its issue price, and Noble pegs fair value near $30 a share (31:59), (39:39). He is equally scathing on Tesla — roughly $40 fair value for a fourteen-times-sales auto company that has earned only about $38 billion cumulatively in its history (35:46), (34:40) — so treat the collapse in these names as a feature, not a bug.
Short Tech, Long Energy, and Own Gold for the Soft Default: Energy, Noble argues, has suffered years of underinvestment against a 5%-plus depletion rate while real-terms capex fell about 70% in a decade, the stocks are cheap, and the sector is only about 3.5% of the S&P (23:00), (27:14). The recent oil weakness is one-offs reversing — China's import cuts, SPR draws, and the Hormuz traffic rebound (25:39), (28:00) — while on gold the daily sentiment index has hit zero and Luke Groman's point holds that rising rates in a country that can't afford them are bullish for gold (16:22), (15:30). Own gold to preserve real wealth into a soft default that inflates the debt away.
Investment Focus
Noble — a rare case of Pod Street Week's publisher featured as the guest — supplies the week's clearest actionable rotation map. The template: (1) reprice semiconductors on asset value, not peak earnings, and read the roughly $25 billion of wrong-way ETF inflows as stale bulls yet to capitulate (6:49); (2) swap SPY for RSP to mechanically cut the AI concentration (10:52); (3) treat Oracle and its OpenAI dependency as a hubris-and-debt short, not a value opportunity (36:24); (4) fade SpaceX toward roughly $30 and Tesla toward roughly $40, recognizing the index-inclusion and low-float mechanics that inflated them (39:39); (5) go short tech and long energy — a cheap, under-owned 3.5% of the index — while accumulating gold and silver with sentiment washed out, positioning for a soft default rather than disinflation (27:14), (16:22).
EP 3 - Oracle at One Notch Above Junk
Ed Zitron — EZPR Founder & CEO, "Better Offline" Host
Ed Zitron, in conversation with Isaac on The Tech Report, argues that Oracle — not OpenAI — is the single most important company to watch in the AI bubble, and that S&P Global has effectively made his case for him by downgrading Oracle to one rung above junk and naming OpenAI as the reason. His thesis is a chain of dependencies: Oracle must build the data centers, OpenAI must pay for them, and neither is likely to happen on schedule. Around that core he maps the data-center owners cashing out at the top, private equity poised to hold the bag, and Larry Ellison's personal margin loans turning a stock slide into a doom loop.
Actionable Bullet Points
S&P Named the Reason for Oracle's Downgrade Out Loud: OpenAI: Zitron notes that unlike most bubble analysis, he did not have to frame the data — S&P Global directly said the downgrade to one rung above junk was because of OpenAI, much as SoftBank drew a negative outlook (9:30). Oracle is building roughly 7.1 gigawatts of capacity for OpenAI, at an estimated $340–400 billion-plus, against a $300 billion five-year contract that only began a month ago (9:51). And Oracle's own annual report warns it may not get paid by OpenAI — language Zitron says makes lenders nervous (10:58). Treat OpenAI's ability to pay as Oracle's entire investment thesis.
Oracle's Core Is Flatlining While Its Only Growth Eats the Margin: The mythology of an ever-growing Oracle is wrong, Zitron argues — its database, hardware and licensing business has plateaued for roughly a decade, and the only growth is AI compute that is destroying profitability, with gross margins down about 15% over three or four quarters and free cash flow deeply negative (11:45), (12:59). Because Oracle cannot build 7.1 gigawatts in three years, it will spend the capex without getting paid, leaving projects eternally "in progress" (12:31). One more downgrade makes Oracle a "fallen angel," forcing investment funds to dump its bonds and roughly doubling its borrowing costs (13:22); the growth line is the problem, not the fix.
Larry Ellison's Margin Loans Turn Oracle's Slide Into a Doom Loop: Ellison, who owns about 40% of Oracle, funds his life through margin loans against his stock and has pledged hundreds of millions of shares; now he has offered roughly $20 billion of his own cash for the Paramount–Warner Bros. Discovery merger he does not liquidly have (17:29), (18:05). If Oracle's stock — already down from about $350 a year ago to roughly $125 — falls toward $70, margin calls force him to post more stock or sell, feeding the decline (20:09). And a Paramount delay past September 30 costs about $500 million a quarter to keep the deal alive (19:47); a controlling holder funding himself on the stock is itself a credit risk.
SoftBank Is Oracle's Twin: Illiquid, Levered, and All-In on OpenAI: Masayoshi Son dismissed bubble fears while simultaneously claiming AI would cost $5 trillion a year by 2040 and require fusion power (21:25). SoftBank has sold its T-Mobile and Nvidia stakes and owns about 80% of Arm, but is near the limit of using Arm as margin-loan collateral and cannot liquidate it without the market treating Arm as a bet on SoftBank — ruining both (23:07). Zitron's verdict is that OpenAI is the "load-bearing fail son" of the AI generation (24:49); recognize that the financing chain rests on one unprofitable keystone.
The Price War Dilutes Everyone, and No One Can Slow Down: Meta is reportedly moving to undercut OpenAI's prices by about 75% using ad revenue, and new entrants keep arriving — Kimi K3 reportedly scores better than Fable (27:20), (28:07). The danger is not that these models seize all the business overnight but that they disperse demand across more parties just as OpenAI and Anthropic sit on more than $1.1 trillion of compute commitments they cannot pause — Anthropic alone needing to pay for roughly $100 billion of compute within two years (28:37). With one CEO already saying prices must fall 90% next year, Zitron predicts the epitaph will be "was that it?" (29:53), (32:02); read dispersing demand as fatal when everything must grow at once.
Investment Focus
Zitron supplies the week's sharpest single-name credit case, and it points at Oracle. The template: (1) treat OpenAI's ability to pay, on time, as the entire Oracle thesis, with the annual-report "may not get paid" language as the tell (10:58); (2) watch the fallen-angel threshold — one more downgrade forces bond funds to dump Oracle and roughly doubles its borrowing costs into a gigawatt-scale build (13:22); (3) monitor Ellison's margin loans and an Oracle price near $70 as a personal-balance-sheet doom loop that is itself a credit risk (20:09); (4) view SoftBank as the same trade — illiquid, levered, and dependent on the OpenAI keystone (23:07); (5) track the price war (Meta's ~75% undercut, Kimi K3, a called-for 90% price drop) as the force that disperses demand against $1.1 trillion of un-pausable compute commitments (28:37).
EP 4 - A.I.-Mageddon: The Earnings Bubble Nobody Is Pricing
Fred Hickey — "The High-Tech Strategist" Editor
Fred Hickey, in conversation with Adam Taggart on Thoughtful Money, agrees with Jeremy Grantham that this is the greatest bubble in US history — but adds a wrinkle he has never seen in nearly five decades of covering tech: an earnings bubble sitting beneath the price bubble. His mechanism is a timing gap: component makers book revenue immediately while hyperscaler costs are deferred through depreciation, inflating today's earnings and hiding tomorrow's. As Chinese models collapse pricing and the economics disintegrate, Hickey is sitting in the most cash of his career, waiting for the wash-out to buy survivors and hard assets.
Actionable Bullet Points
It's an Earnings Bubble, Not Just a Price Bubble, and Stripped Down the Growth Is Single-Digit: Hickey's market ratios run richer than 2000 — the Buffett indicator at 241% of GDP versus 160%, price-to-sales at 3.7 versus 2.3 — and an adjusted CAPE around 67 (4:34), (9:06). But the danger is the E: strip the roughly 40% of Q1 S&P earnings growth that came from one-time mark-to-market gains on stakes in Anthropic, OpenAI and SpaceX, and growth falls to 16%; strip the supplier surge and it drops to single digits (8:01), (8:10). Micron, negative gross margins in 2023, now runs 85% gross margins with the stock up about 700% on the year (5:52); recognize that the earnings, not just the multiple, are the bubble.
The Depreciation Bill Is Deferred, Not Cancelled: Component makers — Nvidia, Micron, Western Digital — recognize revenue almost immediately, while the hyperscalers' costs show up only slowly through depreciation over five to six years, creating a gigantic gap between reported profits and true expenditure (6:44). That bill is now ramping: Google's depreciation expense is set to roughly double from 17% toward 35% by 2028, Oracle is spending 100% of revenue on data centers with debt at two-and-a-half times sales, and only about 5% of planned capacity has even been switched on (18:22), (20:36). Model the depreciation ramp, not today's flattering margins.
Chinese Models Are Collapsing the Price, at Nearly Half of Usage and Climbing: Open Router now shows about 46% market share for Chinese models, up from an 11% twelve-month average, with DeepSeek delivering roughly 90% of tasks at 1.5% of the cost and Zipu exceeding Gemini's latest (15:05), (16:09). The economics were always suspect — Chamath's CTO said token costs were doubling every 45 days for a 5% return — and the reversal is here: Microsoft has curbed most Claude Code licenses, Tesla set $200 weekly token limits, Coinbase halved its AI spend, and a Vinod-backed startup shrank Qwen 3.6 to run on an iPhone 17 Pro (11:33), (13:37), (41:42). Read falling model prices as gutting the revenue assumption that underwrites the entire buildout.
The Circular Backlogs Are the Fragility: Half Sits With OpenAI and Anthropic: Roughly 50% of the hyperscalers' backlog derives from the unprofitable frontier labs, and about 40% of Q1 earnings growth came from marking up stakes in those same labs — paper gains on illiquid shares (19:10), (8:01). The trigger, Hickey argues, is the first hyperscaler to cut capex, pointing to Meta halting its $85 billion metaverse spend, after which the stock rallied (26:55). With the Mag 7 already the "Lag 7" — Microsoft down about 20% and Oracle about 28% on the year — watch for the first capex blink (21:33).
Hickey Is in Maximum Cash, Waiting for the Fat Pitch, and It's Gold: Having sold precious metals into last year's froth, Hickey is holding the most cash he can remember in T-bills, and points to gold miners as the cheapest in over four decades on Scotia's roughly ten-times earnings (35:57), (57:49). The driver is central banks, not retail — China is accelerating purchases while cutting Treasuries from $1.3 trillion toward $630 billion, and central banks now hold more gold than Treasuries (1:00:04), (1:03:04). With physical limits (land, water, copper, permits) and backlash already cancelling about $300 billion of data centers and surveys showing 55% expect AI to do more harm than good, play defense now and deploy dry powder into hard assets on the wash-out (46:38).
Investment Focus
Hickey offers the week's most complete account of the earnings illusion beneath the price bubble. The template: (1) strip the one-time mark-to-market gains and supplier surge from headline S&P earnings to reveal single-digit underlying growth, and treat that as the true baseline (8:10); (2) model the deferred depreciation ramp — Google toward 35% by 2028, only 5% of capacity switched on — rather than current margins (20:36); (3) track the Chinese-model share (about 46% and rising) and the token-spend cuts at Microsoft, Tesla and Coinbase as the pricing collapse that undermines revenue (15:05); (4) watch for the first hyperscaler capex cut as the trigger, using Meta's metaverse retreat as the template (26:55); (5) hold heavy cash and prepare to buy survivors and hard assets — gold miners at forty-year-cheap valuations, backed by relentless central-bank buying — into the correction (57:49), (1:00:04).
