1AI capex is forecast to grow 10% again in 2026 — last year that forecast missed by 60 points
The cleanest number of the day came from Gene Munster, on Excess Returns. AI capital spending was forecast to grow about 10% in 2025 and grew roughly 70%. The published forecast for 2026 is about 10% again; Munster's own estimate is 20-30%. The argument is not about whether the spending continues. It is about how much of it the published number has already counted.
The evidence he and Doug Clinton put behind it is bottom-up rather than thematic. Anthropic's revenue went from about $9bn to a mid-$40bn run rate in four months on Claude Code adoption, and the CTOs of ServiceNow and Uber each burned through a full year's inference budget inside the first quarter. Clinton, who called an AI bubble larger than dot-com back in 2023, now dates the market to 1995-96 rather than 1998 — three to five years before a peak rather than roughly two. Their power corollary is that turning electricity into computation makes demand for it, in his phrase, "seemingly infinite," which is how nuclear and small modular reactors, natural gas and grid storage enter the same thesis.
Brian, on Business with Brian, sizes the spending from the top down: he attributes 92% of US GDP growth in the first half of 2025 to AI data-centre investment, with growth excluding that spending running at 0.1%. That is one channel's figure and uncorroborated, but it is the scale the whole argument turns on.
The money to fund the build-out is still arriving. Cerebras listed the day the Excess Returns episode was recorded and closed its first session up 108%, which Clinton and Munster read as the IPO window opening for the private AI names behind it — SpaceX, Anthropic, OpenAI — with SpaceX expected to be the largest listing ever, ahead of Saudi Aramco. They attach a mechanical caveat that cuts against the incumbents: a listing that size enters the Nasdaq-100 under the 15-day early-inclusion rule, and index funds have to sell something to buy it. The Aramco precedent they cite is a stock up about 30% from issue two weeks in, then fading.
Hold the source problem alongside all of it. Munster and Clinton run AI-themed funds — Deep Water Asset Management and Intelligent Alpha — so a framing in which the Anthropic-linked names are winning suits their book, and they appeared on one episode together, which makes this one conversation rather than two independent reads. The forecast gap is checkable regardless: if 2026 capex grows anywhere near 20%, the published number will have been wrong in the same direction twice running.
2Three AI names beat their quarter and sold off anyway
Brian, on Business with Brian, built a seven-name list on a single filter — the company beat its last quarter and the stock is still below its recent high. Three of the names carry that filter on their own. grew revenue 84.7% and beat on every line, and fell 6% the next day. grew earnings per share 241% year on year and beat on every metric, and fell 19% on the news. sits 23% below its $796 March high while revenue grew 33% year on year and operating margin expanded by eight percentage points.
The other four rest on backlog and price rather than on a beat. carries roughly a $13bn market value against $23bn of contracted lease backlog, including a $7.5bn fifteen-year hyperscaler lease signed in April, and is scaling a pipeline from 600MW toward 1,000MW. booked a record $9.4bn of orders, up 78% year on year, at a book-to-bill of 1.24, having closed its $10bn CommScope acquisition in January. grew revenue 201% year on year at a PEG of 0.9 and guides optical revenue above $500m for fiscal 2027 after buying DustPhotonics for $750m. has doubled data-centre royalties year on year at 97% margins and has now announced its own processor, with Meta and OpenAI named as committed customers.
Two things qualify the list. The entry is not a fresh idea — he notes he flagged the $32bn FAA air-traffic-control program a year before Palantir made the shortlist, so it is a position restated rather than initiated. And the basket is one video on one channel, with a paid sponsor disclosed mid-episode (a gold-exploration company, unrelated to the seven names) and metrics chosen to fit the case rather than a full look at each business.
A different lens on the same tape came out of the Excess Returns episode. Tesla traded down when it committed to more than $25bn of capital spending, and Meta did the same on a $175-185bn number, while Alphabet and Microsoft raised their capex guidance and snapped back after hours. Clinton and Munster's read is that spending gets paid for when a cloud revenue line arrives with it and discounted when it does not — which is also how they rank recent mega-cap performance, ahead of ahead of , with the Anthropic-linked names beating the OpenAI-linked one. The capex-light alternative, , carries its own overhang in the OpenAI litigation.
Put the two videos side by side and the question being priced is not whether the quarter was good. It is whether the spending behind it has a revenue line attached.
3Stocks and gold fell together while the dollar firmed — the pressure is coming from the long end, not from fear
Yahoo Finance's market wrap had equities falling as bond yields rose, with the 10-year moving against stocks and the 30-year near 5% squeezing the rate-sensitive sectors — REITs, utilities, long-duration tech. On the Reuters wire carried through Finnhub, gold fell to a one-and-a-half-month low on those same higher US yields and a firm dollar, and the dollar held near six-week highs on rate-hike wagers and war uncertainty. Reuters ties part of those wagers to the US-Iran stalemate.
Equities and the classic hedge falling in the same session, with the currency bid, is a discount-rate move rather than a flight to safety. Bloomberg's read, carried by Yahoo, put the stress in the riskiest stocks specifically and tied it to rate uncertainty rather than to results.
It had not carried into the new session as of the pre-market: S&P futures were up 0.21% and Nasdaq futures up 0.55%, with the Dow down 0.03% — a mild positive bias, led by the Nasdaq. But if the pressure is coming from the long end rather than from company results, the week's earnings can be good and the rate-sensitive names still wear the discount.
4Crude priced a de-escalation the policy track has not delivered
Oil fell after President Trump said the US would end the Iran war very quickly, and Chinese tankers carrying about 4m barrels of crude left the Strait of Hormuz. Vice President Vance told a White House briefing that the talks had made a lot of progress. That is the de-escalation the tape traded.
The policy track ran the other way inside the same 24 hours. The Senate advanced a measure curbing the president's war powers. Treasury Secretary Bessent pressed for more disruption to Iran's financing and said the US sanctions list would be reviewed. The CFTC opened a probe — first reported by the Wall Street Journal, carried on the Reuters wire — into a spike in oil-futures trading that preceded Trump's postponement of the strikes. Reuters paired the UK's cooling CPI print with a warning that the Iran-driven relief inside it is set to be temporary. And Reuters/Ipsos put the president's approval at 35%, with Republican support softening, in the same week the Senate moved on his war powers.
Nothing in the day's reporting removes a sanction or signs a term. The escalation premium is coming out of crude on words, and the same wire carried the tankers leaving Hormuz and the war-powers vote within hours of each other — the kind of move a single headline can reverse.
What the sources recommended
Seven buy calls, no sells, and all seven out of one video on one channel — a single view stated seven times, not seven sources agreeing.
The buys. (Palantir) on the gap between an 84.7% revenue quarter and a 6% next-day fall, plus the $32bn FAA program — though that program was flagged on the same channel a year earlier, so it is a restatement rather than a new idea. (Applied Digital) on a market value under half its contracted lease backlog. (Credo Technology) on 201% revenue growth at a PEG of 0.9 and the silicon-photonics integration. (Amphenol) on a record order book and a 1.24 book-to-bill after the CommScope deal. (Meta Platforms) on a 23% drawdown against accelerating revenue and expanding margins, at a PEG of 0.93. (Arm Holdings) on the new proprietary processor and 97%-margin royalty growth. (Teradyne) on a 19% post-beat sell-off with AI now around 70% of revenue.
All seven are framed as months-to-years positions accumulated by averaging in rather than as trades, which is the horizon on which the beat-and-drawdown filter would have to work.
The week has no inflation print, so one earnings report is carrying the macro
Finnhub's calendar has no Fed meeting, no CPI and no payrolls this week, which leaves after the close tonight as the most consequential scheduled event on the board — estimates of about $80.2bn of revenue and $1.79 of earnings per share. Morgan Stanley reset its target into the print, Reuters' Morning Bid led on it, and the only fresh company-level item alongside it is the start of Vera CPU shipments. If the spending gap in the first section is real, and if what gets paid for is a revenue line rather than a beat, tonight's print is where both show up first.
Before the open: at about $3.00 expected, reporting into a housing-slowdown overhang, at $1.02 and at $2.93. at $12.82 joins Nvidia after the close. Thursday brings at $0.66 and at $5.87 before the open and at $2.56 after it; at $1.35 closes the week on Friday. The IPO calendar is empty for the next seven days, after Cerebras. G7 finance ministers are pressing for action on economic imbalances, aimed at China, and the Iran headlines stay live through the session.
