THIS WEEK · 12 PODCASTS · WEEK OF JUNE 19, 2026
This week we bring you twelve essential podcasts circling one question from every angle: how much of the AI boom is real, what the energy and inflation reality does to it, and which trades sit on the other side. Kevin Coharki and Jonathan Weil open with the cleanest version of the case — strip the cash cost of stock-based compensation out of reported free cash flow and Meta’s 2025 figure collapses from $46 billion to $5 billion, an optical illusion dressed up as a money-printing machine. Jim Chanos and Val Zlatev turn the same mismatch into a long/short blueprint — own the silicon, not the “landlords.” Julien Garran delivers the deepest bear case of the week: six structural reasons the large-language-model ecosystem cannot earn its cost of capital. Jim Bianco offers the measured bull’s mirror — it may still be 1997, not 2000, yet AI alone is now 49% of the S&P 500, the most concentrated the index has been since the railroads. Jim Grant zooms out to history, arguing this is one of the greatest bubbles of all time, with the railroad build-out and the deflation after the panic of 1873 the right rhyme. Kevin Muir maps the market plumbing — serial “rolling mini-bubbles,” a flip from stock scarcity to stock abundance, and a possible “token mirage.” David Hay supplies the missing input: this is the most severe energy crisis on record, roughly half of announced AI data centers are stalling for lack of power, and the summer’s ~$4.6 trillion IPO wave dwarfs every prior year combined. Jeff Currie sharpens the point — you can’t print molecules, so AI is a cyclical commodity business deserving 10–15x, not 30–40x. David Woo carries it to the dollar — AI is the one thing keeping the greenback riding high, and its first crack is the strongest case for gold. Charles Goodhart and Manoj Pradhan supply the deepest macro frame — the disinflation of the past forty years was a one-off supply shock now reversing, leaving central bankers “unanchored” between inflation and a bond-market crisis. Chris Whalen brings it to the portfolio: the bond market has already delivered its own rate hike, and the rotation into defensive income and hard assets is underway. And Steve Eisman and Tom Gallagher close with a forensic look at the credit plumbing beneath it all — whether private equity is hollowing out the life-insurance industry. 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 Optical Illusion in Big Tech’s Free Cash Flow — Kevin Coharki & Jonathan Weil — CAE Consulting / Purdue & WSJ “Heard on the Street” — Watch Full Video
EP 2 Own What the Chips Produce, Not Where They Reside — Jim Chanos & Val Zlatev — Chanos & Company & Analog Century Capital Management — Watch Full Video
EP 3 The Achilles Heel of the AI Economy — Julien Garran — MacroStrategy Partnership — Watch Full Video
EP 4 Two Stock Markets: AI and Everything Else — Jim Bianco — Bianco Research President & Founder — Watch Full Video
EP 5 The Next Crash Won’t Look Like 2008 — Jim Grant — Grant’s Interest Rate Observer Founder & Editor — Watch Full Video
EP 6 The Rolling Mini-Bubble Machine — Kevin Muir — The Macro Tourist Author — Watch Full Video
EP 7 The Energy Bottleneck Beneath the AI Boom — David Hay — Haymaker; Former CIO, Evergreen Gavekal — Watch Full Video
EP 8 You Can’t Print Molecules — Jeff Currie — Carlyle & Abaxx; Former Goldman Sachs Commodities Head — Watch Full Video
EP 9 Gold on the Other Side of American Hegemony — David Woo — David Woo Unbound Founder — Watch Full Video
EP 10 The Unanchored Central Banker — Charles Goodhart & Manoj Pradhan — LSE Emeritus & Talking Heads Macroeconomics — Watch Full Video
EP 11 The Bond Market Already Hiked — Chris Whalen — Whalen Global Advisors Chairman — Watch Full Video
EP 12 Is Private Equity Hollowing Out Life Insurance? — Steve Eisman & Tom Gallagher — The Real Eisman Playbook & Evercore — Watch Full Video
Full summaries with actionable insights and investment focus for each podcast follow on the pages below.
EP 1 - The Optical Illusion in Big Tech’s Free Cash Flow
Kevin Coharki & Jonathan Weil — CAE Consulting Principal & Purdue Professor; The Wall Street Journal “Heard on the Street”
Kevin Coharki and Jonathan Weil, in conversation with host Telis Demos on the WSJ’s Take On the Week, deliver the cleanest version of the week’s central case: the free cash flow that makes the hyperscalers look like money-printing machines is, in Weil’s phrase, “something of an optical illusion.” The reason is a cost that never appears in the standard free-cash-flow calculation — the cash that big tech spends buying back stock to offset the dilution from employee stock-based compensation. Strip it out, and Meta’s reported 2025 free cash flow falls by roughly 89%, from $46 billion to $5 billion. Coharki and Weil walk through why the accounting is perfectly legal, why EBITDA and leverage screens hide it, and why the balance sheet is now absorbing the strain through debt and equity issuance.
Actionable Bullet Points
Reported Free Cash Flow at the Hyperscalers Is an Optical Illusion: The standard free-cash-flow calculation — cash from operations minus capital expenditures — ignores the cash cost of stock-based compensation, namely tax withholding plus the buybacks used to offset dilution. Adjust for it and Meta’s 2025 free cash flow collapses from $46 billion to roughly $5 billion (13:28), an ~89% reduction (12:47); Alphabet’s falls from $73 billion to $24 billion (13:36). Do not value these businesses off the headline free-cash-flow number.
The Buybacks Are Compensation, Not Capital Return: Coharki estimates roughly 90% of Meta’s buybacks last year existed purely to offset dilution from employee stock grants (6:48). His test: take Meta private and tell the engineers there is no more stock — to keep them you pay cash, which lands in SG&A and craters both net income and operating cash flow (9:11). Treat dilution-offsetting buybacks as the operating expense they are.
The Accounting Gap Is Being Funded With Debt: Because cash is going to buy back stock, the balance sheet has to carry the build-out. Alphabet borrowed $30 billion in Q1 and issued $85 billion in shares (16:24); companies that were asset-light with zero debt five years ago are now levering into a data-center land rush (17:24). Watch rising debt and equity issuance as the tell that legacy cash flows no longer cover CapEx.
EBITDA and Leverage Screens Hide the Cost: Adjusted EBITDA adds stock-based compensation back, so EV/EBITDA understates the true cost and net-debt/EBITDA covenants understate leverage — a credit-rating and interest-rate risk sitting in plain sight (18:32). Discount the multiples and the covenant math on these names.
The Sell-Side Recovery Is a Leap of Faith: Analysts model a “checkmark-shaped” free-cash-flow rebound to 2025 levels and a near-doubling by 2029 (20:30), even as Nvidia’s CFO floats $3–4 trillion in annual data-center spend by 2030 — roughly a 50% increase in all global IT spending (22:24). And the S&P earnings lift is partly a timing artifact: sellers book profits now while buyers defer depreciation that has not yet begun to toll (23:17). Be skeptical of the money-printing narrative until the depreciation actually lands.
Investment Focus
This is the cleanest accounting deconstruction of the AI profit boom available this week, and it underwrites everything that follows in the edition. The investment template: (1) strip the cash cost of stock-based compensation out of reported free cash flow before you value any hyperscaler; (2) treat dilution-offsetting buybacks as an operating expense, not shareholder return; (3) watch the balance sheet — rising debt and equity issuance signal that legacy cash flows no longer cover the build-out; (4) discount EV/EBITDA and covenant math, because the true leverage is higher than the screen shows; (5) read the S&P’s record earnings partly as a timing illusion — sellers recognizing revenue now, buyers deferring depreciation — and expect the marks to compress when the depreciation cycle finally tolls.
EP 2 - Own What the Chips Produce, Not Where They Reside
Jim Chanos & Val Zlatev — Chanos & Company; Analog Century Capital Management
Jim Chanos and Val Zlatev, on a Macro Minds panel hosted by Jack Farley, take the same accounting mismatch — chipmakers booking revenue immediately while hyperscalers capitalize their spend — and turn it into a long/short framework. Chanos, net long AI against his shorts, is not betting against the silicon; he is short the unprofitable business models attached to the AI ecosystem, from Bitcoin-miners-turned-data-center-developers to the neoclouds. Zlatev, who runs a long/short book across more than 500 semiconductor and hardware names, maps the memory super-cycle, the spot GPU-price spike, and why this is emphatically not 1999 in valuation terms. Their shared discipline: own the technology, not the real estate.
Actionable Bullet Points
The Profit Mismatch Is the Whole Trade: Chip and equipment sellers — Nvidia, GE Vernova, Vertiv — book revenue and profit immediately, while the hyperscalers spending those dollars capitalize them and defer the expense (7:16). Chanos’s rule cuts through it: “own what the chips produce, not where the chips reside” (9:46). Position long the silicon, short the real estate.
Neoclouds Are Leveraged Equipment-Leasing Companies, Not Tech: CoreWeave and its peers are run by ex-Magnetar finance operators, and Blackstone has now entered the space via a REIT — these are finance companies making a bet on the life of the chips (14:43). Even on heroic assumptions and a generous 10-year GPU life, the deals pencil out to just 5–8% pre-tax ROIC (16:31), returns Chanos expects to erode as capital floods in. If single-digit returns are the best the peak offers, avoid the middleman.
The GPU-Rental Spike Is Real but Fragile: Six-to-eight-year-old GPUs are up 40–50%+ since January, reversing the normal 20–30% annual price decline, as exploding token demand from reasoning models, larger context windows and agents outruns supply (12:44). Zlatev’s caveat: contracted prices have not moved nearly as much, and newer, more efficient architectures should resume the decline. Use the spot spike as a sentiment indicator, not a thesis.
Memory’s Cycle May Be Higher and Longer Than the Screen Implies: Memory stocks trade at 6–7x forward earnings, pricing a downturn six-to-nine months out (38:43); Zlatev argues the plateau lasts two-to-four years because equipment makers like ASML and Applied Materials physically cannot grow output more than ~30–35% a year (40:01). DRAM and flash are already up 4–5x, pushing memory from 20% to 50% of a PC’s bill of materials and driving unit sales down mid-teens (42:53). Respect the supply constraint, but size it as a cyclical plateau, not a permanent re-rating.
Valuations Are Dispersed, Not Uniformly Insane: Unlike 1999 — when Cisco traded at 160x earnings (49:08) — Nvidia sits at ~15x 2027 EPS and Broadcom ~12x 2028, while networking names stretch to 50–60x forward (49:39). The discipline, in Chanos’s words: “don’t put magical valuations on mundane businesses” (51:53), with the mundane models exposed within 18–24 months. Separate the cheap oligopolists from the expensive hopefuls.
Investment Focus
Chanos and Zlatev convert the accounting mismatch into a tradeable map of the AI complex. The investment template: (1) favor the technology layer — the chips and the silicon inside the box — over the “landlords,” the neoclouds and legacy data centers where returns on capital are thin and competed away; (2) read single-digit ROIC at the current peak as a warning that the middlemen lose when capital floods in; (3) treat the spot GPU-price spike as sentiment, not signal, since contracted prices have barely moved and new architectures should resume the price decline; (4) on memory, respect a real and physical supply constraint but remember it is a cyclical business — size to a multi-year plateau, not a permanent re-rating; (5) watch the one tail risk that breaks the picks-and-shovels thesis — scaling laws are empirical, not physics, so a genuinely cheaper architecture that sticks would force an entirely different discussion.
EP 3 - The Achilles Heel of the AI Economy
Julien Garran — MacroStrategy Partnership Partner
Julien Garran, in episode 75 of his regular update series recorded on SpaceX listing day, delivers the deepest bear case of the week: the Achilles heel of the US economy and its markets is the failure of the large-language-model ecosystem to earn a return over its cost of capital. He calls it “history’s biggest Hail Mary” and lays out six structural reasons the ecosystem cannot make money — and walks through the cracks already emerging, from the end of token maxing to SoftBank failing to raise margin against its OpenAI stake. The throughline: this is not a profitable industry waiting to mature, but a capital stock that does not earn a return, which makes it, in his words, the opposite of progress.
Actionable Bullet Points
The Whole Stack Loses Money — and Subsidizes the Layer Below It: Garran’s core claim is that no layer earns its cost of capital: Nvidia funds its own customers, the data centers (independent and hyperscaler-embedded) run at a loss subsidizing the frontier labs, which subsidize the app layer — Perplexity, Replit, Lovable — which subsidizes the end user (0:56). On the latest pre-S1 data, he says OpenAI loses about $2 for every $1 it earns, with costs roughly 3x revenues (6:45). Treat the entire ecosystem as structurally unprofitable, not temporarily so.
The Depreciation Bomb Has Not Detonated Yet: Construction delays mean hyperscalers have not started depreciating their unfinished data centers, so the spend crushes cash flow but not yet reported income; as the delayed facilities open, Garran expects depreciation to “hit thick and fast” (5:46). Do not extrapolate current reported earnings — the cost is coming.
Token Maxing Is Dying in Real Time: Uber burned its entire annual token budget in four months and has now sharply limited usage, with Walmart following a day later (9:55); Garran cites engineers told that heavy usage made them “better employees,” one reportedly running routines to spend around a million in six months (11:25). Watch the collapse of token maxing as artificial demand drains out of the numbers.
A Castle With No Moat: Garran traces a brutal cost curve — roughly $50 million for GPT-3, $500 million for GPT-4 and $5 billion for GPT-5, which insiders said was not enough better to justify the 10x bill (24:54) — and notes DeepSeek and others copied it within four months at about a tenth of the cost to train and run (35:08). With no commercial moat and a scaling wall in both data and compute, the economics never close; don’t underwrite durable pricing power at the model layer.
Misallocated Capital Is Two-Thirds of GDP — and SpaceX Is the Tell: Garran’s Wicksell-spread work puts misallocated US capital (AI plus private equity, private credit and Sunbelt condos) at 67% of GDP, about 23x the dot-com peak (51:01). He flags SpaceX’s listing — roughly $86 billion raised at a $1.85 trillion valuation, near 98x annualized Q1 sales with no earnings — as a macro event, the Mars/moon/Kardashev ambitions filed under science fiction (44:13). Position out of LLM-AI exposure ahead of what he frames as a deflationary bust.
Investment Focus
This is the most complete technical-and-commercial bear case in the edition — the argument that the AI ecosystem cannot earn its cost of capital by construction. The investment template: (1) treat every layer of the LLM stack as structurally lossmaking and reliant on the subsidy from the layer above; (2) expect the depreciation cycle, delayed by construction lags, to compress reported earnings as data centers come online; (3) read the end of token maxing — Uber, Walmart — as artificial demand draining out of the numbers; (4) discount pricing power at the model layer, where there is no moat and DeepSeek-style copies arrive within months at a tenth of the cost; (5) reduce LLM-AI exposure broadly and watch the Wicksell spread, the SoftBank/OpenAI margin stress, and the coming IPO disclosures as the sequence that turns a funding problem into a deflationary unwind.
EP 4 - Two Stock Markets: AI and Everything Else
Jim Bianco — Bianco Research President & Founder
Jim Bianco, in conversation with Soar Financially, offers the edition’s measured bull case — constructive on AI’s long-run promise, sober about what it does to inflation, the Fed and your return expectations. He argues the US market has effectively split into two asset classes, AI and everything else, with the cycle still closer to 1997 than 2000. The capex, in his telling, can be paid off if the hundreds of billions spent on software subscriptions get redirected into AI. But the same forces leave him expecting stickier inflation, no rate cut, and a Warsh-led communication reset — and he wants investors to dial their expectations down to a “4-5-6” world.
Actionable Bullet Points
The Market Has Become Two Asset Classes — AI and Everything Else: Bianco notes the entire market move is AI: the non-AI universe is up just 1% in five months, while AI stocks hit 49% of the S&P 500 at the early-June peak — the most concentrated the index has been since the railroads in the 19th century (6:11, 6:43). Treat AI as a high-risk, high-reward sleeve of its own — capable of +100–150% and then –50% — and the rest of the market as a separate, slower book (33:27). Don’t let one asset class masquerade as “the market.”
It’s Still 1997, Not 2000: Bianco puts the AI cycle around 1997–98, before the frothy blow-off, because fewer than 1% of enterprises run agentic tools today while ~80% are merely testing (8:10, 9:00). His bull case for the capex: the ~$300–400 billion a year corporates spend on SaaS gets redirected into AI as a single context window replaces Windows, Office, Outlook, Teams and CRM — the way the iPhone subsumed a dozen devices (9:20, 10:23). If that promise holds, the data-center build is not as oversized as it looks; size for a bubble later, not yet.
AI Is Initially Inflationary, Not Deflationary: Against the consensus that productivity gains must be deflationary, Bianco argues the near-term effect is inflationary: goods deflation has run its course as globalization ends, while AI makes the most productive workers more productive and pushes their wages — and the prices of their output — higher (15:34, 16:22). Robots replacing service jobs is years away (16:43). Underwrite stickier services inflation, not the disinflation everyone expects.
Expect a Warsh Communication Reset — Starting Wednesday: Bianco sees no rate move (CPI 4.2%, core near 3%, funds at 3.5% — too high to cut), but flags that Warsh may go “cold turkey”: no dot submitted, a shorter press conference, and an end to the cascade of “what-if” forward guidance Powell leaned on (21:05, 22:26). Because validating the dot plot once makes it hard to kill, the change has to come immediately. Watch the format of the meeting as much as the decision.
Dial Into a 4-5-6 World: Bianco’s framework for the non-AI market: cash returns ~4%, bonds ~5%, non-AI stocks ~6% over time — and 10-year yields near 4.45% on the way to 5% are a neutral, fair-value rate, not bond vigilantes (34:15, 30:00). Fair value, he argues, tracks nominal growth (inflation plus real growth, ~5%), so a drift higher is fine if growth follows; the danger is a Fed that cuts into an inflation problem and makes long yields soar (28:06). Reset return expectations down, and treat bonds at 5% as a legitimate holding.
Investment Focus
Bianco is the edition’s measured bull — constructive on AI’s long-run promise, sober on what it does to inflation and returns. The investment template: (1) treat the market as two asset classes and size AI as a volatile, separate sleeve rather than confusing it with “the market”; (2) accept that the cycle may still be early (his 1997–98 analogy), with the SaaS-spend redirect as the mechanism that could justify the capex — and a real bubble still ahead; (3) underwrite stickier services inflation as globalization’s goods-deflation tailwind fades; (4) expect a Warsh communication reset — fewer promises, no dot plot — and no cut while core inflation sits near 3%; (5) anchor expectations to a 4-5-6 world, with 10-year yields near 4.5–5% as a neutral rate, and remember that a Fed cutting into inflation is the fastest route to higher long yields.
