‘Devil really is in the Details’ with AI. OpenAI, Jane Street & YouTube. ARD #143
‘The Devil is in the Details.’ The old proverb, and today it truly fits a number of things going on in AI land: three events where the story lives in the fine print, plus a sneak peek at Apple’s AirPods with ‘cameras’ that are likely not really cameras.
Today’s ARD is about those details, in this AI Tech Wave: the mega data center deal behind OpenAI’s mega-IPO run, the Leopold crash reaching Jane Street, and YouTube recalibrating for the AI content era. With my takes on each.
(1) OpenAI and Nvidia Close the Ohio Mega-Deal: the Fine Print
OpenAI signed a 20-year, 10-gigawatt data center lease in southern Ohio with SB Energy, SoftBank’s energy arm, on a former uranium-enrichment site owned by the Department of Energy, with a government gas plant funded by Japan under the recent trade deal. All in, the project could top half a trillion dollars.
July’s headline was a $250 billion Nvidia backstop, and Nvidia’s stock fell 5 percent on that report. The closed deal is a backstop of up to $105 billion, first phase only, and it kicks in only through a waterfall of mitigation steps: SB Energy would first re-lease the site at the same price, then sell it, and only then does Nvidia pay the difference in value, on completed data centers only. Nvidia is backing the asset, not OpenAI’s lease payments. In return: exclusive chips for half the site, potentially $600 billion of revenue through 2030, and $1.5 billion of equity in SB Energy ahead of its own IPO, as soon as next month. Japan recoups its $33 billion first; the US government then takes 90 percent of the power revenue.
My take: a complex Lego set of detailed pieces, intricately put together. The transaction is calibrated to optimize risk and reward for every party involved, with a relatively fair sharing of burdens allotted to each. The Street calls this structure a ‘credit wrapper’, and Google is already doing a version of it for Anthropic. These details are new and illuminating, and they will be repeated by almost everyone doing similar deals around the AI industry, in the trillions of dollars. And remember: Anthropic and now OpenAI are prepping their IPO stories for investors. These sorts of details are the grist for that mill. Nvidia’s ‘Kingmaker’ role and the ‘race to zero’ economics underneath have not changed.
Sources:
-
WSJ: “OpenAI Locks In Lease for Huge Data Center in Ohio With Backing From Nvidia”: the closed deal, the mitigation waterfall, SB Energy’s IPO.
-
WSJ: “Nvidia in Talks With OpenAI to Guarantee $250 Billion Financing for Data Center”: July’s original structure; the ‘credit wrapper’ framing.
-
Bloomberg: “Nvidia Will Back First Phase of OpenAI Project With as Much as $105 Billion”: phased 2028-2030 over 20 years; Jensen: ‘OpenAI will pay the lease.’
For longtime readers:
-
AI-RTZ #1182: OpenAI pre-markets its IPO through the churn (today’s companion)
-
AI-RTZ #1171: Nvidia the ‘Kingmaker’, now uber AI investor and banker
(2) Leopold’s Fund Crash Reaches Jane Street: the Leverage Math
Leopold Aschenbrenner, 24 years old, took a few hundred million to a $45 billion fund, up to $100 billion with leverage, on the back of a 165-page essay. He borrowed three to four dollars against every dollar of capital, longs and shorts amplifying the same AI theme. Wall Street calls that a ‘Texas hedge’.
Then July turned. Cheaper Chinese open-source models spooked the AI trade, and rivals did not need inside information: prime brokers publish aggregate leverage reports to clients daily, traders matched the drop against his known filings, and the shorts circled. The fund finished July down 67 percent, about $30 billion, selling the bulk of its stock book to Ken Griffin’s Citadel at a roughly 10 percent discount, signed twenty minutes before the market open, as his wedding guests arrived in Carmel. Still up 80 percent on the year.
The new detail that carries it to Wall Street’s core: Jane Street, the quant giant that rarely backs outside managers, lost about $15 billion in July, its worst month ever, roughly half tied to Situational Awareness and the rest to the AI selloff and its own de-risking. This at a firm running a record year, $40 billion-plus of revenue through July.
My take: the drama aside, these stories highlight that AI trends are jagged indeed, in both their technologies and their financial manifestations. The leverage turned the jagged into near-fatal cuts. Leopold’s own investor letter said it plainly: these dynamics are essentially similar to a bank run, vulnerability begetting more vulnerability. Other funds saw the wound and came in for the kill. We covered the first act on ARD #132 and on the Lumida podcast. We are at the beginning of the beginning, year four post-ChatGPT. These stories are but the start of these tales, and their details to come.
Sources:
-
WSJ: “How Wall Street Sussed Out That Situational Awareness Was On the Ropes”: the six-day unwind, tick by tick.
-
WSJ: “Jane Street Suffers Loss of About $15 Billion Following Troubles at Situational Awareness”: worst month ever, in a record year.
-
WSJ: “His Wedding Guests Were Arriving, Just as His $45 Billion Fund Was Falling Apart”: the rise, the leverage, the Carmel weekend.
For longtime readers:
-
ARD #132: ‘As the AI World Turns.’ Oracle, OpenAI & Leopold (the first act)
-
MP on the Lumida Wealth podcast: ‘Ladies, Leverage and Leopold’
-
AI-RTZ #1174: The ‘Dwarkesh’ Compute Demand Bull Case (part 2)
(3) Google/YouTube Recalibrates Creator Pay and Views for the AI Content Era
Two changes in the same week that look uninteresting separately, until you put them together again like a Lego set. Detail one: the pay bar for millions of YouTube Creators goes up. From February 1, 2027, monetizing on YouTube requires 8,000 qualified watch hours over the past year, or 20 million qualified Shorts views over 90 days, both double today’s bar, with ongoing activity minimums to keep earning.
Detail two: the view count goes up. From August 24, a view counts the moment a video starts playing, the way TikTok, Instagram and YouTube’s own Shorts already count. The old, stricter metric survives in the analytics dashboard as ‘engaged views’, and payouts still run on the engaged metrics. But the public number every creator shows brands and sponsors will climb faster.
My take: YouTube is getting ready for a world with far more AI-generated content, be it ‘AI Slop’ or quality AI-assisted work from the millions of creators who make a living directly and through branded deals, an economy already in the hundreds of billions globally. A higher bar to get paid, a more liberally counted view to make the scale numbers pop: tightening the payouts while inflating the optics. And remember, one person’s AI ‘slop’ is another person’s AI content gold. YouTube’s prime-time TV ambitions and the AI content flood are now converging on the same platform economics.
Sources:
-
The Verge: “YouTube is making it harder to earn money on YouTube”: the YPP bar doubles Feb 1, 2027.
-
YouTube: “An update to how we count public views across YouTube”: first-frame views from 8/24; ‘engaged views’ preserved.
-
The Verge: “YouTube is changing how it counts views to give the numbers a boost”
For longtime readers:
My overall take: new details on complex AI deals and strategies are what matter, whether they are deals, transactions, or platform arrangements. A financing structure that looks circular until you read the mitigation waterfall. A fund collapse that looks sudden until you read the leverage ratios. A metrics change that looks technical until you read what is coming down the content pipeline. All of these developments are just getting started in this AI Tech Wave. A lot of devil discernment ahead, on details to come.
Gadget AI: Apple’s AirPods, Sensors, Not Cameras
The sneak peek is charmingly Apple: MacRumors found the demo video sitting inside the macOS 26.7 release candidate. A man holds up a book, and the voiceover says: ‘With Visual Intelligence, your world becomes savable. See something you like? Just ask me to save it for later.’ Codename B790, and possibly out as early as this fall, alongside the iPhone 18 lineup.
The key detail: these may not be ‘cameras’ in the traditional sense. Current reports point to tiny infrared sensors, Face ID-style, not photo or video cameras. No capturing pictures, no recognizing faces. Just ambient visual context feeding Apple’s Siri AI, Visual Intelligence, and related AI services, with a privacy light when visual data heads to the cloud.
My take: I have written about ‘AirPods with cameras’ for two years, and the mislabel is the key detail potentially. Apple is doing it the Apple way, leveraging one of its biggest platforms and its hardware-software ecosystem across iPhone, Mac and Watch, uniquely versus its peers. The other side of that coin is the risk that society shuns cameras on our bodies, as Google Glass learned a decade ago. I wish they weren’t called cameras. I dare say they are going to be AirPods with sensors.
Sources:
-
MacRumors: “Apple’s Camera-Equipped AirPods Confirmed: See Them in Action”: the macOS 26.7 RC demo video; Visual Intelligence integration; September timing per Gurman.
For longtime readers:
Q&A
Q1: What is the most interesting aspect of Apple’s AirPods with cameras? ANSWER: The privacy protections likely built into this manner of non-image sensing: no face recognition, just the non-visible side of the spectrum providing ambient context for Siri and Visual Intelligence. Very differentiated versus Meta, Google and other peers on AI devices with cameras. And Apple can seamlessly leverage its hardware and software ecosystem, with these AirPods working with iPhones, Macs, Watches and the rest. A very unique superpower relative to its peers.
Q2: What is the biggest concern over AirPods with cameras? ANSWER: The risk that even Apple succumbs to a privacy backlash, as Google did with its smart glasses over a decade ago, and as Meta is seeing in the initial concerns around its AI smart glasses. All of these devices ask society to get used to new ways of capturing information around us. Cameras on our body, as it were. It is early days for this class of devices.
Today’s companion AI-RTZ, #1182, is on OpenAI pre-marketing its IPO through the churn, versus Anthropic’s numbers.
Full Source Reading
OpenAI/Nvidia Ohio deal
Leopold and Jane Street
YouTube’s creator recalibration
Gadget AI: AirPods with sensors
Clips from today
Apple AirPods: Sensors, Not Cameras
YouTube Doubles the Bar to Get Paid
YouTube Preps for the AI Content Flood
AirPods Cameras: Apple’s Privacy Edge
An impact of the AI Tech Wave worth tracking. Stay tuned.
(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here.)
