Home Affordability Crisis, Palantir's Advantage, Big Short on AI, H-1B Abuse, Solar Storm Hits Earth

The episode opens with Michael Burry, the investor known from the big short, who the hosts say has just deregistered his firm with the SEC. He had disclosed short positions against Palantir and made a broader bet against AI. According to the discussion, CNBC reported the value of the Palantir shorts as roughly 900,000,000, and Burry said the real figure was 9,000,000. One participant mentions that Sacks could not join because he had been up late at the White House.

how do you confuse 9,000,000 and 900,000,000?

The hosts first suggest the gap might come from confusing the cost of the options with the value of the stock behind them. They settle on a more specific cause: each option covers 100 shares, and the multiplier was applied wrongly. One participant says the error points to a newsroom with little personal experience of asset markets and no fact checking. In that participant's view, the inflated figure drove a whole news cycle and helped Burry's short. The argument is that a $9,000,000 bet would have drawn no notice, while a headline describing a bet two orders of magnitude larger created a headwind for AI stocks. One host jokes that the AI bubble was already deflating after Brad Gerstner "popped it."

Under the label "conspiracy corner," one participant speculates openly that the writer might have understood markets perfectly well, published the error on purpose knowing the editors would not catch it, and held a short position of their own. The participant suggests CNBC should check whether the writer had a trade on. This is offered as speculation, and no evidence for it is given. The participant presents two possibilities: the writer either does not understand financial markets and should not be covering them, or understood them and inflated the headline to move the market. The participant adds that a billion-dollar short would have caught their attention.

The more serious charge discussed is Burry's claim that Meta and Oracle have been "cooking the books with a 176,000,000,000 in hidden depreciation to inflate earnings by over 20% in 2028." The question is how long AI hardware, such as an NVIDIA H100 or a full NVIDIA server, stays useful: three years, or four to six. The hosts say CoreWeave may be using six years. Given the scale of the data center build-out, that choice changes reported expenses a great deal. Spreading a $100,000,000,000 data center over six years instead of three halves the yearly charge.

After a joke about "accounting corner," one participant explains the GAAP mechanics. Capital expenditure is not deducted from operating profit in the year it is spent. It is depreciated over an estimated useful life, and companies run periodic internal reviews to adjust schedules that do not match reality. As an illustration, the participant says Google made $120,000,000,000 in operating profit over the last twelve months and assumes $70,000,000,000 of CapEx this year. Spread over three years, that is about $24,000,000,000 a year. Spread over six years, it is $12,000,000,000 a year. The participant puts the difference in operating profit at roughly 10%.

The participant summarizes Burry's claim as follows: hyperscalers have roughly doubled the useful life of their data center assets, which halves the reported operating cost and inflates earnings. The participant then gives Google's history. In 2021, Google extended servers from three to four years and networking equipment from three to five years. In 2023, it extended networking equipment from five to six years. The participant notes that the chart shown on the show combines servers and networking equipment.

The participant gives a technical reason for these changes. Data centers used to be mainly storage and data-transfer facilities. Their hard drives were thrown away every eighteen or thirty-six months, solid-state memory lasted two to six years depending on type, and RAM lifetimes varied. With the AI boom, most spending and most energy now go to processors. The shift in the hardware mix is presented as the context for the longer schedules.