News vs. Noise: AI Capex Rises as Markets Hesitate

Iran Blockade Reignites Oil — Even After the Toll Got Scrapped

President Trump notified Congress on July 13 (in a letter dated July 10) that the U.S. had resumed military action against Iran and reinstated a naval blockade on Iranian shipping through the Strait of Hormuz, effective July 14 [1]. Brent, the international benchmark price against which roughly two-thirds of the world’s traded oil is priced, jumped above $83 a barrel within hours and traded up more than 10% on the week as the U.S. and Iran continued exchanging strikes [2]. Trump also floated a 20% “reimbursement fee” on all cargo transiting the Strait — then dropped it the very next day, July 14, after Gulf leaders lobbied against it, saying he’d pursue trade and investment deals with Gulf states instead [3][4].

The toll’s reversal doesn’t change the underlying picture. The blockade is still in effect, strikes are ongoing, and oil is pricing a live supply risk, not a one-day headline. We’re watching Brent and actual Hormuz shipping volumes here, not toll headlines that can reverse in 24 hours.

Fed Chair Warsh’s First Testimony Kept the Rate Path Deliberately Unclear

Fed Chair Kevin Warsh delivered his first semiannual Monetary Policy Report to Congress on July 14–15, telling lawmakers that cooling June inflation was not “mission accomplished” and that the committee has “no tolerance for persistently elevated inflation” [5][6]. He outlined five internal task forces — communications, balance sheet policy, economic data, productivity and jobs, and inflation frameworks — but declined to signal the Fed’s next move, consistent with his stated preference for giving markets less forward guidance than his predecessors, public signaling about where interest rates are likely headed [7].

Less guidance means every incoming data point carries more weight than it used to, not less. We’re treating this testimony as the baseline for reading the rest of the summer’s inflation prints, rather than assuming the Fed’s reaction function is what it used to be under prior chairs.

ASML and TSMC Raise Their Numbers on the AI Buildout

ASML — the Dutch company that makes roughly 90% of the world’s chip-lithography machines and is the sole maker of the extreme ultraviolet (EUV) systems needed for the most advanced chips — raised its 2026 sales outlook to €43–45 billion, up from €36–40 billion guided just one quarter earlier [8]. Taiwanese chipmaker TSMC raised full-year capex guidance 15% to $60–64 billion, lifted its revenue growth outlook from 30% to “over 40%” year-over-year, and committed an additional $100 billion to its Arizona operations, bringing total U.S. investment to $265 billion [9][10][11]. Capex, short for capital expenditures, is the money a company spends to build or upgrade physical assets, such as factories and equipment.

Credit Markets Are Pricing AI Risk More Cautiously Than Earnings Headlines Suggest

S&P Global cut Oracle’s credit rating to BBB- — one notch above junk — on July 9, citing a $250 billion data-center buildout burning cash faster than revenue replaces it [13][14]. Capex rose 162% to $55.7 billion in fiscal 2026, leaving free cash flow, cash left over after a company pays for operations and capital spending, negative by $23.7 billion, with a further roughly $42 billion cash-flow deficit forecast for fiscal 2027 [13][14]. Separately, Bank of America’s July survey of 210 portfolio managers overseeing a combined $555 billion found 82% now call “long global semiconductors” the most crowded trade on record, up from 73% in May, meaning more investors have piled into the same position than BofA has ever recorded, raising the risk of a rapid, self-reinforcing unwind if sentiment shifts [15][16]. Another 45% cited an AI bubble as the top tail risk [15][16]. The semiconductor index sat roughly 18% below its June peak even after TSMC confirmed a 77% profit jump [11][17].

None of this says the buildout is failing — the point above says otherwise. It says the market’s more sophisticated participants are more skeptical of this quarter’s headline numbers than retail sentiment is. The distinction that matters here is between the bottleneck and the proxy: TSMC and ASML’s own numbers confirm real demand, while the sharp drawdown in chip stocks reflects that same crowded trade unwinding, not weakening demand [2]. We’re watching credit spreads and positioning surveys as a cross-check against equity headlines. Credit spreads are the extra interest rate investors demand to hold a company’s debt over safer government bonds, so a widening spread signals rising perceived risk; positioning surveys poll professional investors on which trades they currently hold.

Record Bank Earnings Look Like Broad Strength — Mostly Two One-Time Items

All five major U.S. banks beat earnings estimates last week [18][19][20]. JPMorgan posted a $21.2 billion quarterly profit, the highest in company history, on $7.70 per share [18][19][20]. Goldman Sachs’ earnings per share (EPS), a company’s profit divided by its number of outstanding shares, came in at $20.98, far above the $14.48 estimate [18][19][20]. Wells Fargo posted $2.00 per share against a $1.72 estimate, on revenue up 8.6% year-over-year [18][19][20]. Taken together, the results read like unambiguous evidence of a resilient economy heading into the second half of the year.

Most of the outperformance traces to two nonrecurring events landing in the same quarter: fee income from SpaceX’s $85.7 billion initial public offering, the largest in history, which alone generated roughly $500 million in underwriting fees across the banks, and a spike in trading volume tied to the same Iran-conflict volatility described above [4][21]. Strip those two items out, and the “historic quarter” looks a lot more ordinary. One-time deal fees and volatility-driven trading revenue don’t repeat every quarter, and we wouldn’t extrapolate a broader economic or market signal from a quarter this loaded with one-offs.

Markets Priced Out a July Hike Off Two Soft Prints — That Outran a Divided Fed

June’s CPI (the Consumer Price Index, which tracks what households pay for goods and services) and July’s PPI (the Producer Price Index, which tracks prices businesses receive for their output) both came in cooler than expected [22][23]. Traders cut the implied odds of a July rate hike to roughly 11%, down sharply from earlier in the month [22][23]. On the surface, that looks like a market that has decided the Fed’s tightening cycle is over.

It also outran what the Fed itself was actually saying. Chair Warsh’s testimony offered no forward guidance by design, leaving both a hike and a hold on the table for the Fed’s next meeting [6]. Dallas Fed President Lorie Logan became the first of his colleagues to publicly call for higher rates last week, arguing that current policy isn’t restraining the economy and that AI’s demand effects on jobs and prices are “already here” [24]. Add the Hormuz-driven oil spike described above, and the disinflation reflected in these two prints has a fresh headwind the market hasn’t fully priced in. Two cooler-than-expected prints don’t settle a policy debate the Fed chair himself pointedly declined to settle.

IBM’s Historic Crash Wasn’t a Software-Sector Verdict — It Was a Capex Story

IBM fell 25% on July 14, its worst single-day drop on record, after missing on both earnings and revenue estimates [25]. The decline surpassed even Black Monday in 1987 [26]. Peers with little direct connection to IBM’s specific business, including Microsoft, ServiceNow, Salesforce, and Intuit, fell 3–5% the same day in apparent sympathy, as if IBM’s miss was a read on enterprise software demand broadly.

CEO Arvind Krishna was explicit about the actual cause: clients are redirecting spending away from IBM’s software and mainframe business toward AI hardware — memory, servers, and storage — to lock in capacity and pricing before both get more expensive [25]. Mainframes are large, high-capacity computer systems that many big enterprises still rely on to run core operations. That pattern is being described elsewhere as a broader “memory tax” reshaping corporate IT budgets, as companies across the industry prioritize memory and AI infrastructure spending over other technology investments [27]. That’s a company-specific capex-reallocation story, not evidence of weakening enterprise software demand overall — the peers that sold off alongside IBM have little direct exposure to its mainframe transition, and most of that intraday loss reversed by the close.