1Most AI revenue traces back to two loss-making labs, and Friday's sell-off hit the firms most exposed to them
The AI sell-off deepened on Friday. The Nasdaq Composite fell 2.14% and the S&P 500 1.09%, according to Yahoo Finance. The VIX volatility index jumped 21% to just under 20. The worst moves were in the firms that rent out AI computing on borrowed money. CoreWeave fell 11.9% and Nebius 12.2%. Palantir dropped 7.2%, and Yahoo Finance tallied $260 billion lost across software stocks.
Steve Eisman, on The Compound and Friends, explained why a few names can move the whole trade. He calls it a concentration problem. He cites Nvidia's latest quarterly filing: about 70% of 's accounts receivable come from five customers. Those customers are the cloud giants. He says about 70% of their AI revenue comes from OpenAI and Anthropic. He adds that half of Oracle's order backlog is OpenAI alone. Oracle fell 4.6% on Friday, to about $128.
His bear case is a price war, not a collapse in demand. Open-weight models, many of them Chinese, now cost about a tenth as much. He says companies are cutting back on unlimited AI use, and expects the two labs to cut prices. If revenue growth turns negative, funding dries up, and their commitments to Oracle and others shrink. He thinks an OpenAI failure would put the US into recession "immediately". He also stresses this is not 2008: "The banks are fine."
His hosts pushed back. Josh Brown argued demand is not the problem. In his telling, the investment banks he visited all plan bigger AI budgets for 2027. If the debt-funded neoclouds cannot raise money, he says, the work goes back to Amazon , Alphabet and Microsoft . Eisman agreed the hyperscalers are stronger. He replied that a price war at the two labs would still hit the whole chain.
Liaquat Ahamed, the financial historian, made the same structural point on Prof G Markets. He compared the build-out to the railroad bubble that ended in the panic of 1873. That crash began when the era's best-connected financier, Jay Cooke, could not finish raising money for the Northern Pacific line. Ahamed suggests OpenAI is the obvious modern candidate. He says "we have entered the manic phase". He also says timing cannot be read from history, because past private investment booms ran two to four years. This one is about two years old. At about 3–4% of GDP, it is below the 5% peak of past booms.
So the two disagree on timing, not on structure. Neither offers a date. Both name the same trigger: one major borrower failing to raise money. That shifts attention from the chipmakers' strong balance sheets to the two private labs' revenue figures.
2Two hot inflation prints moved the expected Fed hike to December, not October
September's consumer and producer price reports both came in hot this week. FRED's consumer price index rose 0.65% on the month. On Friday, industrial production for September printed 0.66% higher than August, and capacity use rose to 76.9% from 76.4%. Unemployment was 4.2% in September, up from 4.1%. Payrolls grew by only about 29,000.
That data argues for tighter policy. Bond prices moved the other way. The 10-year Treasury yield fell to 5.10% on 15 October, from 5.21% two days earlier, per FRED. The Compound hosts read it as fear. As the AI trade wobbled, investors bought Treasuries for safety, and they saw AI, not the deficit, as what had been driving yields up.
The prediction markets back the delay. Kalshi prices an October hike at 13%, down from 15% the day before, on about $1.4 million of open interest. A hold sits at 86%. The hike has moved, not vanished. Kalshi's December contract prices a 25-point hike at 75%. Polymarket gives "another Fed hike in 2026" 77.5%, on about $99,000 of liquidity.
Ahamed adds the risk that matters most to him. He argues booms bust without causing depressions unless policy makes a mistake. In 1873 that mistake was dropping silver from the money supply. His worries now are a stock market worth 250% of GDP, which ties consumer spending to share prices. He also points to the $1 trillion foreign investors put into US stocks in a year. If that money leaves and the dollar falls, he says, the Fed could not easily ease. The so-what is a conflict that remains unresolved. A December hike into a bursting AI trade is exactly the mismatch he describes.
3Oil is pricing a wider war while the ceasefire market still gives it four in five odds of holding through October
Reuters reported that oil touched $110 a barrel for the first time since May, on Middle East and Russia supply fears. US November crude settled at $99.02, up 2.6%, per Yahoo Finance. The oil fund rose 2.8%. Several events pushed it there. Reuters reported President Trump said the US "may need to strike" Iran's Pickaxe Mountain site. It also reported missiles hit Aden airport while a Saudi-led coalition intercepted one aimed at the Saudi oil port of Yanbu.
Polymarket's contract on WTI reaching $100 in October jumped 33.5 points in a day, to 94%. The $110 contract rose 11 points to 21.5%. Both books are moderate, with about $20,000–45,000 of daily volume. The supply response is small next to that. Reuters reported the US is offering an exchange of up to 17 million barrels from the strategic reserve. Late Friday, Reuters also reported Trump waived some Russia-related sanctions and tariffs before they took effect.
The war market has barely moved. Polymarket gives the US–Iran ceasefire 78.5% odds of lasting through 31 October, down 2 points, on about $516,000 of liquidity. Through 30 November it falls to 38.5%. So oil traders price an escalation premium now. The deepest war contract prices the breakdown for November. The two are not yet the same bet, and a strike on Iran would settle it.
What the sources recommended
Only two single-name buys came out of four distinct videos, both from one source in one paid-membership pitch. Ross Givens named Freeport-McMoRan and Kinross Gold as his two "commodity boom" picks. His copper case rests on an IEA estimate that mine supply will fall 25% short of demand by 2035. He also cites the IEA's 17-year lead time for a new mine. Freeport's Grasberg mine ran at about 67% of capacity in the third quarter, after a 2025 mud rush. Freeport expects it near full by the end of 2027, so he sees output growth without new discoveries. He put the stock "in the low 70s"; it closed Friday at $69.51, down 0.8%. For gold, he cites China's central bank buying about 23 tonnes in September, its 23rd month in a row. Kinross is his beaten-up leg. He says it sits about 40% below its high on a small production miss, while raising cash returns to 50% of free cash flow. It closed at $24.03, up 0.8%. He named the risks himself: a global recession for copper, and a Chinese property recovery cooling gold demand.
The one avoid call was on a company that is not listed. Asked if he would buy Anthropic when it goes public, Eisman said "No way." He suspects it needs to list before reporting its third quarter, because slowing usage will start to show in its numbers.
Givens also made a bullish case on SpaceX buying low-band spectrum for a Starlink phone service. He framed it as a story for existing holders rather than a buy call. closed at $146, below the ~$160 he cited.
Next week puts the AI trade's two biggest questions on a calendar
Earnings start the test. Netflix, Capital One and Intuitive Surgical report on Tuesday, along with defence names RTX and Northrop. Meta, Tesla, IBM, Lam Research and Texas Instruments follow on Wednesday. Meta is the read on whether hyperscaler capex holds up. Lam is the read on chip equipment. Lockheed, Honeywell and Newmont report on Thursday; Newmont tests Givens's gold case directly. The Fed decision lands on 28 October, the same day T-Mobile reports. That call is the first test of Givens's SpaceX spectrum thesis.
The private test is Anthropic's listing. Polymarket prices an IPO by 31 December at 70.5%, down 3 points, on about $43,000 of liquidity. The 15 December contract fell 7 points to 60.5%. If Eisman is right that the numbers are softening, the odds and the timing should keep slipping.
