The Wall Street Skinny

Go to groundnews.com/skinny⁠ for a better way to stay informed. Subscribe through our link for 40% off unlimited access to world-wide coverage.

Nathan Fielder’s You Can See Everything is the most talked-about documentary of the year, and after sitting through all three hours of it (there is exactly one acceptable bathroom break, and we tell you when), we came out more divided than when we went in. The film asks us to decide who Elizabeth Holmes actually is: a liar who knew the truth and deceived people anyway, a woman who lied to herself so thoroughly she no longer knows the difference, or a useful idiot who has been handed from one controlling man to the next. One of us walked in firmly in the second camp. The other changed her mind three times before breakfast. So we start by asking whether those are even the right three options.

To get there, we recap what Theranos actually was: the Enron connection, the Stanford dean who never bought it, Rupert Murdoch’s $125 million, the Pfizer logos pasted onto “independent” reports, and the diluted blood samples telling Walgreens customers they’d had miscarriages they hadn’t had. Then we walk through what Fielder does with it, and why the second half of the movie turns into something closer to a heist. Why does Amanda Seyfried go from hired reenactor to wanting to break up a relationship? What does Billy Evans have power of attorney over, and why does he carry a gun? Did Elizabeth know her boyfriend had negotiated a cut of the documentary when she looked into the camera and said she had no economic upside? And when two people watch the same scene of Nathan Fielder asking a convicted fraudster about the word “can,” how do they come away with opposite conclusions about whether she’s being literal or being clever?

That last question is where the two of us really split. One of us has known the other since age seven and has some things to say about how literal minds actually work, including a detour through medevac helicopters, SEC testimony, and whether a woman who quotes Yoda in her interviews gets to claim she doesn’t understand figurative language. The other wants to know why Adam Neumann, Elon Musk, and half of Silicon Valley never saw the inside of a courtroom for the same fake-it-till-you-make-it playbook. Is Elizabeth Holmes a mastermind, a product of the zero-rate era, or just the one who got caught? And after three hours of getting closer to her than anyone ever has, do we know more about her than we did before, or less?

Timestamps

0:00 We’re in costume2:00 The three ways to judge Elizabeth Holmes2:50 Where we each land going in5:30 Theranos recap: the Enron dad, Stanford, the pivot9:00 Murdoch, the Waltons, and the $9 billion valuation11:00 The voice, the turtleneck, and fundraising as a woman14:40 No diligence: fake Pfizer reports, diluted blood, Walgreens16:50 The whistleblowers and what she was actually convicted of20:20 The documentary, part one: the month before prison and Billy Evans24:30 The economic upside question: did she know?27:00 Amanda Seyfried joins and wants to break them up29:30 Confronting Billy: power of attorney, the X account, the rage30:30 The kids, and whether we’re being manipulated31:50 Dead man switches and a gun33:30 From The Rehearsal to the reenactment35:20 The prison visit: two questions37:00 Did Nathan fall for her?39:20 Three things at the end: the contract, the swim, the New York Times quote42:00 Why Billy? Rich versus wealthy45:40 Does she believe her own lies? Who’s on first50:30 Figurative language and the medevac helicopter testimony56:30 Why Nathan Fielder? The “you picked me because I’m incompetent” confession59:30 Will the public give her a second chance?1:02:30 Is she any different from the rest of Silicon Valley?1:04:50 Where we actually agree1:06:40 ZIRP, AI promises, and what everybody does1:08:00 Do we know less than when we started?

Go to groundnews.com/skinny for a better way to stay informed. Subscribe through our link for 40% off unlimited access to world-wide coverage.

0:00 Why Enron matters for the AI conversation right now
5:00 Origins: the gas bank, Lay, Skilling, and Fastow
9:00 Mark-to-market accounting and booking 20 years of profit on day one
13:00 Growth targets, exec comp, and regulated vs deregulated energy
15:00 SPEs, the 3% rule, and Fastow's LJM
19:00 Circular earnings: the dark fiber swap and disguised loans
24:00 Earnings games, leverage ratios, and why cash flow matters
28:00 Every deal was a stinker (and Netflix before Netflix)
29:00 The collapse: Chanos, debt puts, the commercial paper run, Dynegy
35:00 Broadband, Blockbuster, and the $8 business valued at $60 billion
38:00 The fake trading floor
40:00 Take-or-pay, leases, and off-balance-sheet debt in AI today
44:00 Meta's Project Hyperion and thinking about risk holistically
50:00 Accounting is imperfect: pensions, EPS maximizers, and the AI lesson
55:00 California, the Raptors, Arthur Andersen, and who called it

Why is Enron suddenly back in the conversation about AI? This week, we revisit The "Smartest Guys in the Room" to unpack what actually happened in the Enron scandal. While there are some similarities, a lot is different. Enron hid debt, manufactured earnings, and convinced Wall Street to buy a story that even many professional investors couldn’t explain. From a fake trading floor to financial structures backed by its own stock, we break down what actually happened behind one of the most infamous corporate collapses in history.

Then we turn to the financing behind today’s AI boom: off-balance-sheet obligations, circular financing, and the gap between reported earnings and actual cash flow. We dig into Meta’s Project Hyperion, the long-term commitments funding the data center build-out, and why a chipmaker booking revenue today while its customer spreads the cost over years matters when evaluating the boom. Along the way, we explain mark-to-market accounting, special purpose vehicles, and why “it’s not on the balance sheet” doesn’t mean “there’s no risk.”

To be clear, we’re not accusing today’s AI companies of Enron-style fraud. The point is understanding where the comparisons hold up, where they fall apart, and what investors should actually scrutinize. A technology can change the world without every company funding it earning a return. Revenue isn’t profit, profit isn’t cash flow, and a rising stock price isn’t proof that a business model works. Before you call something “the next Enron,” you need to understand the first one.

Also if you want to learn more about accounting, valuation, M&A etc. Kristen spent 13 years teaching this to analysts and associates at bulge bracket investment banks and business schools and then spend two years creating the most comprehensive course out there. https://www.thewallstreetskinny.com/investment-banking-private-equity-fundamentals

Use code: VIPCUSTOMER for 20% off

Go to groundnews.com/skinny for a better way to stay informed. Subscribe through our link for 40% off unlimited access to world-wide coverage.

How do Bitcoin miners end up running the infrastructure behind AI? This week we break down neoclouds like CoreWeave, Nebius, and Lambda. What makes a neocloud different from AWS, Google Cloud, Azure, and the hyperscalers? Why were crypto miners, sitting on warehouses, power contracts, and a business model with a built-in expiration date, so well placed to pivot? And if these companies don’t have the cash flows of a Google or a Microsoft, how are they paying for billions of dollars of chips? We dig into chip-backed financing. How can a junk-rated company borrow against Microsoft’s AAA credit? Why are lenders comfortable with 95% advance rates? What happens when the compute contracts are only three to five years long and customers like OpenAI aren’t profitable yet?

Then we get to the NVIDIA of it all. NVIDIA invests in the neoclouds, backstops their unused capacity, and sells them the chips. Is that savvy, or is it circular financing? How does it compare with Lucent in the dot-com bust? What about the take-or-pay obligations that never show up on anyone’s balance sheet? Plus: OpenAI’s revenue miss, how media bias shaped the coverage, and whether Starbucks and Chipotle should merge (pumpkin spice burrito, anyone?).

0:00 Market-Moving Headlines: OpenAI’s Revenue Miss
3:27 Starbucks & Chipotle M&A
6:29 What Are Neoclouds?
9:12 From Crypto Miners to Neoclouds
13:40 Hyperscalers vs. Neoclouds
17:21 Chip-Backed Financing
23:30 The NVIDIA of It All
29:56 Circular Financing Risks

Go to groundnews.com/skinny for a better way to stay informed. Subscribe through our link for 40% off unlimited access to world-wide coverage.

We're back at the intersection of AI and high finance!

Anthropic is reportedly targeting a $2 trillion IPO valuation. So what did Reuters find when it got an early look at the S-1? Kristen, our resident investment banker, walks through the numbers: $4.6 billion in 2025 revenue, a $42 billion loss (and how much of it was an accounting charge on a convert), $20 billion in cash against a reported $518 billion compute plan, and a quarter of revenue coming from just two customers. Why don't those compute contracts show up as debt when the neoclouds on the other side are borrowing against them? And with open-weight models closing the gap, is it time to get nervous about this IPO?

Then we turn to the off switch. After a summer of OpenAI agents breaking out of the sandbox, a paused model, and a scrapped GPT-6.1 Astra release, who is actually reaching for the brakes? We cover Dario Amodei's "We Must Pace the Frontier" essay and the cynical read on it, the AI Kill Switch Act and the AI Emergency Button Act stalling in Congress, the executive order rebranding AI as "super intelligence," and the White House Accord on Super Intelligence, typo and all. And why did the industry name its newest products after the villains in The Matrix?

Finally, NVIDIA swore off kill switches in 2025, citing a 1993 government fiasco called the Clipper Chip. So why did it just announce a two-part system called OpenShell and Sentry, and what would it catch inside a private equity data room or on a trading desk? Can any guardrail keep up with something built to be smarter than us? Plus the trolley problem, Iain Banks' Culture series, and whether we should just give up and let the robots win. Make sure you're subscribed to our Substack so you don't miss our more in-depth coverage on these complex topics: https://substack.com/@thewallstreetskinny

Go to groundnews.com/skinny for a better way to stay informed. Subscribe through our link for 40% off unlimited access to world-wide coverage.

NVIDIA just agreed to buy Hugging Face, and the deal won't add a cent to its earnings per share. So why do it? In this episode we dig into what NVIDIA really gets from owning the home of open-weight AI, and whether this is a threat to OpenAI and Anthropic that the market hasn't priced in yet. What's the actual difference between open source and open weight, and why does almost everyone (yes, even us included) mix them up? If Amazon, Google, and the frontier labs are all building their own inference chips, where does NVIDIA's next wave of customers come from? And if the free "Temu version" of AI gets good enough for most people, who keeps paying for the Prada bag?

Then we bring on Michelle Weaver, Head of US Thematic Equity Research at Morgan Stanley, to ask the questions everyone in finance is worried about. Which companies are actually seeing AI show up in their margins, and which are just burning tokens so they can say "AI" on the earnings call? Is AI already showing up in the jobs data for young workers? What happens to the traditional Wall Street analyst class if the org chart stops being a pyramid? Does the Jevons Paradox of economic theory help or hurt the frontier model companies? And why are American companies nervous about building on Chinese open-weight models?

We also get into the growing fight over AI data centers. Why has this become one of the few issues uniting voters on both the left and the right? What do power bills, water use, and old brownfield sites have to do with it? How big is a 57-gigawatt power shortfall, really? And what does Oracle declaring force majeure on its New Mexico Stargate site tell us about where the AI build-out is headed?

00:00 Intro: Revisiting our early episodes
03:00 Life updates: Flying, dogs & kids
06:10 Our Yahoo Finance Market Hang recap
07:50 Why NVIDIA bought Hugging Face
11:10 Training vs. inference: Where AI compute is headed
14:15 CUDA and NVIDIA's software moat
15:50 Open source vs. open weight, explained
19:20 Is NVIDIA a threat to OpenAI & Anthropic?
21:25 Meta's Muse and why easy AI wins
25:00 Why companies struggle with AI adoption
27:20 Oracle's Stargate force majeure (Ground News)
28:45 Michelle Weaver joins: What is thematic research?
30:15 AI enablers vs. AI adopters
35:05 Open vs. closed models and Jevons paradox
37:25 The risks of Chinese open-weight models
39:45 Pricing security risk in AI adoption
42:15 Token budgets and the end of token maxing
44:00 Is AI already costing jobs?
46:55 What AI means for Wall Street analysts
49:25 Data centers: People, power & politics
53:00 Tax breaks, brownfields & state moratoriums
55:35 The 57-gigawatt power shortfall
57:35 US vs. China: Chips, power & the AI race
59:20 Where to find Michelle's research