$125.53
24h
-14.01%
22 Nov, 03:09
Bearish on Nvidia amid AI infrastructure bubble concerns
Bearish0.00%
AI Summary
The author argues that the Piketon AI infrastructure project is a fraudulent bubble that will likely never materialize, triggering a severe revaluation of AI hardware stocks and explicitly warning against buying Nvidia even at a 50% discount.
The Trillion-Dollar "Chain Reaction" on Ohio's Empty Land: The True Cost of the Piketon Incident and the AI Bubble
I. Background: Three Giants Tied Together in a "Left Foot Stepping on Right Foot" Scheme on Demolition Land
In the second half of 2026, the most magical epitome of the AI infrastructure narrative fell upon an unfinished Cold War uranium enrichment plant site in Piketon, Ohio, USA.
SB Energy, under SoftBank, plans to build an unprecedented AI super park here: 8GW of computing power, accompanied by 9.2GW–10GW of natural gas power generation.
To assemble this grand chessboard, SoftBank, OpenAI, and Nvidia formed an extremely tight closed loop of circular financing:
Landlord (SoftBank / SB Energy): Holds about 13% equity in OpenAI, but itself is under high leverage pressure, with a single-quarter massive loss of approximately $7 billion (including a $3.7 billion write-down of US data center assets), causing credit default swaps (CDS) to soar.
Tenant (OpenAI): Signed a 20-year lease, but incurred operating losses exceeding tens of billions in a single quarter, unable to obtain investment-grade credit ratings, making it fundamentally unable to afford the rent.
Guarantor and Exclusive Seller (Nvidia): To sell chips, Nvidia personally stepped in to provide up to $105 billion in credit guarantees for OpenAI's lease, conditional on the exclusive deployment of Nvidia's full-stack hardware in the park. Meanwhile, OpenAI and Nvidia injected $500 million and $1.5 billion in equity into the "landlord" SB Energy, respectively.
The three lifted each other up, painting a magnificent blueprint of hundreds of billions of dollars on paper. But reality is starkly different: Masayoshi Son held the groundbreaking ceremony back in March, yet by summer, satellite images showed no signs of construction on this land (0 MW under construction). Environmental impact assessments are expected to drag until early 2027, with the first power supply not arriving until 2028–2029 at the earliest. More critically, the SEC has begun reviewing Nvidia's accounting compliance regarding computing power financing and guarantees.
II. Core Characterization: A Scam of "Zero Credit Default, but Zero Future Growth"
To see through the essence of projects like Piketon, you only need a few plain words:
This type of transaction is essentially a capital scam. That so-called trillion-dollar super park is highly unlikely to ever be built; market promises will ultimately be torn apart.
Growth? 0.
Future revenue? 0.
But most importantly, the actual credit risk behind this is extremely low—the land hasn't started construction, money is merely verbal promises, and no real funds have been disbursed. What truly disappears is just the "future profits" drawn on Wall Street's ledgers.
This distinction is the biggest cognitive misconception in the current market: many people conditionally associate "$105 billion in guarantees" with the 2008 subprime crisis. But you cannot default on a loan that was never disbursed. In project financing, banks disburse funds based on physical construction progress; if the project stalls at the demolition stage, Nvidia's trillion-dollar guarantee will naturally become void, and banks and the bond market won't have the chance to lose money.
III. Impact on the Bond Market: Localized Bleeding, Far from 2008
Piketon itself hardly harms the bond market: aside from the small amount of initial equity invested by SoftBank and Nvidia, and the widening spread of SoftBank's own bonds, syndicated loans have not been drawn down on a large scale, so it will not trigger a chain reaction of explosions in the banking system.
The real exposure of the entire AI bond market is controllable: Of the approximately $680 billion in AI-related debt issued across the market this year, the bulk belongs to cash-cow hyperscale cloud factories like Microsoft, Google, and Meta (delayed construction actually turned their free cash flow positive instantly).
Those facing default are merely highly leveraged new clouds (such as CoreWeave) and private credit SPVs holding approximately $150 billion–$250 billion in drawn-down debt. Accounting for residual value recovery of 40%–60% from powered-up land and second-hand hardware, the final net credit loss for the entire market is likely only $75 billion–$150 billion—a mere flesh wound for the global bond market, far from evolving into a systemic financial crisis.
IV. Impact on the Stock Market: Evaporation of Trillion-Dollar Expectations, Triggering a "Double Kill" Revaluation
The bond market can escape unscathed, but the stock market cannot. Bondholders only care whether principal repayment is possible, while the stock market trades on growth multiples (Multiple) for many years:
Nvidia's "Low Valuation Trap": Once paper parks like Piketon are debunked, Nvidia's potential hardware orders worth hundreds of billions over the next few years will directly hit zero, while losing pricing power from sellers. Even if the stock price halves and its dynamic P/E ratio drops to seemingly cheap levels of 13–14 times, in a semiconductor downturn cycle, once the earnings denominator (E) shrinks, a low P/E is often a classic trap marking the peak of a cycle.
A 50% drop in Nvidia's stock is not an opportunity.
Ultimately, the Piketon incident does not puncture the foundation of the US financial system, but rather the illusion of "perpetual high growth" in AI hardware.
It is more akin to the clearing out of the telecom infrastructure bubble in 2001: limited credit losses, but the stock market must pay for those "trillion-dollar empty lands" that will never be built through a long period of valuation regression.
How many more Piketons are there in this wave of AI infrastructure?