AI Capex on Trial: The Week That Separated Signal from Spend — And a Fed That Couldn't Agree

The final week of July 2026 will be remembered as the moment markets drew a line between AI spending and AI proof. Microsoft added $450 billion in market value in a single day — the largest one-day capitalization increase in stock market history — because it linked $255 billion in planned capex to $678 billion in signed backlog. Amazon became the first company ever to surpass $200 billion in quarterly revenue. Meta and Apple, by contrast, were penalized: Meta for spending more and earning less, Apple for a Services slowdown and a China miss that punctured a 44-times earnings multiple. Meanwhile, the Federal Reserve held rates in a 9-3 vote with three hawkish dissents, and Chair Kevin Warsh offered so little forward guidance that the bond market did the talking for him — pushing the 10-year Treasury to 4.67%.

The result was a week of sharp divergence beneath indices that finished modestly green. The S&P 500 gained 0.97%, the Dow 1.11%, and the Nasdaq 1.42%^[1]. But the Russell 2000 barely held flat at +0.22%, and the month of July was the S&P 500’s first July loss since 2014^[2]. Consumer Discretionary led all sectors with an 8.39% weekly surge — almost entirely Amazon — while Utilities fell 3.75% and Information Technology slipped 0.45%^[1]. The market was not rewarding “tech” broadly. It was rewarding specific companies that could demonstrate, with numbers, that their AI infrastructure spending was meeting real demand.

The Earnings Scorecard

Sources: TradingKey^[3], LPL Research^[1], FMP quotes (as of 16:00 ET, July 31)^[4].

Microsoft Changes Everything

Wednesday July 30 was the hinge of the week. Microsoft’s Q4 FY2026 report, released after the close on July 29, triggered a 16% stock spike — the company’s biggest single-day rally ever and the largest one-day market capitalization increase in stock market history^[3]. The catalyst was not revenue or EPS in isolation. It was the convergence of three demand signals:

• Azure grew 43%, beating the 39-40% guidance range^[3]
• Commercial remaining performance obligations (RPO) hit $678 billion, up 84% year over year^[3]
• Copilot reached 30 million paid seats, and management framed FY2027 capex guidance of $255-260 billion as fulfillment of demand that already exceeds supply^[3]

The message was unambiguous: Microsoft is not speculating on AI. It is building to meet contracted backlog. The market’s response — an 8% gain in the iShares Semiconductor ETF and a nearly 5% gain in the XLK tech sector ETF, the largest single-day IT sector move since mid-2025^[3] — confirmed that investors had been waiting for exactly this kind of proof.

Amazon Confirms — and Makes History

If Microsoft provided the backlog, Amazon provided the velocity. The company reported $200.61 billion in Q2 2026 revenue — the first time any company has exceeded $200 billion in a single quarter — representing 19.6% growth^[3]. Amazon Web Services grew 37%, its fastest growth rate since 2021^[3]. Net income reached $62 billion, though roughly $53.4 billion of that was bolstered by a non-cash revaluation of Amazon’s stake in Anthropic^[3].

CEO Andy Jassy framed the $220 billion annual capital expenditure as a response to customer demand with visibility through 2028^[3]. The stock responded with a 15.3% gain on July 31, closing at $271.58^[4]. The Consumer Discretionary sector’s 8.39% weekly lead was almost entirely an Amazon story^[1].

The Anthropic markup, however, is a recurring analytical concern. Alphabet’s EPS was similarly boosted by a non-cash revaluation of approximately $6-7^[3]. Analysts will increasingly strip these out, and the “beat” premium on future hyperscaler reports may narrow as the market learns to adjust.

Apple and Meta: The Other Side of the Test

The same standard that rewarded Microsoft and Amazon punished Meta and Apple.

Meta’s earnings miss was a 14% decline in EPS, with free cash flow collapsing to $784 million even as the company raised its capex range to $130-145 billion^[3]. The stock fell 9.64% in the immediate reaction before recovering to $556.71 by Friday’s close^[4]. The problem was not the spending itself — Microsoft is spending more. The problem was the absence of a demand narrative to justify it.

Apple reported record revenue of $109.42 billion and EPS of $2.02, up 16% and 29% respectively^[3]. But the stock dropped 7.4% to $308.91^[4]. Two misses drove the sell-off: Services revenue came in at $30.74 billion versus $31.22 billion expected — the first growth slowdown in Apple’s highest-margin segment — and Greater China revenue was $18.8 billion versus $19.6 billion forecast^[3]. At 44 times earnings entering the print, Apple had no room for a Services deceleration. Management also flagged a memory supply constraint, adding to the caution.

A Divided Fed: 9-3 and the Bond Market Speaks

On July 29, the FOMC voted 9-3 to maintain the federal funds rate at 3.50-3.75% — the fifth consecutive hold^[5]. Three regional presidents — Beth Hammack (Cleveland), Neel Kashkari (Minneapolis), and Lorie Logan (Dallas) — dissented in favor of a rate hike, the largest dissenting coalition since Warsh took office^[5].

The statement acknowledged that “economic activity is expanding at a solid pace despite elevated uncertainty” partly attributable to geopolitical conflict^[5]. But it was the press conference that unsettled markets. Chair Warsh reiterated the need for lower inflation but provided little forward guidance, declined to specify how or when policymakers might act, and suggested a new preferred inflation metric may be selected^[1].

The market reaction was telling. Long-term yields rose meaningfully on the day while short-term rates fell — a divergence suggesting markets are becoming less convinced the Fed is serious about inflation, even as they price in the possibility that the long end may ultimately force the Fed’s hand^[1]. The 10-year Treasury settled at 4.67%, up 33 basis points year over year^[6]. The 10-2 yield curve steepened to +0.45%, up 15 basis points month over month^[6].

Going into the meeting, markets had assigned a 33% probability to a hike — an unusually elevated number by FOMC day, when the odds are typically near zero^[1]. The hold did not resolve that tension. It deferred it.

The Macro Backdrop: Growth Cools, Inflation Stays Sticky

The macro picture that frames the Fed’s dilemma is one of decelerating growth alongside above-target inflation — the classic late-cycle tension.

• Q2 GDP cooled to 1.5%, missing consensus estimates of 2.0%, as a surge in semiconductor imports weighed on the calculation^[1]. Real GDP on a year-over-year basis stands at 2.1%^[6].
• CPI inflation is 3.46% YoY — above the Fed’s 2% target for more than five years^[6].
• Unemployment is 4.2%, up 0.1 percentage point year over year but down 0.1 month over month^[6].
• Consumer sentiment stands at 49.5, down 18.45% year over year — a strikingly low reading that complicates the “solid pace” narrative^[6].
• VIX is at 20.66, up 29% year over year and 17% month over month, reflecting the elevated cross-asset volatility of July^[6].
• High-yield credit spreads are 2.84%, up 9 basis points month over month but down 5 basis points year over year — not yet flashing distress, but no longer tightening^[6].

The closest historical analogs in the FRED kNN search are mid-2006 — when unemployment was 4.6-4.7%, CPI was running near 4%, and the Fed was holding at 5.25%^[6]. That period did not immediately resolve into recession, but the tension between sticky inflation and a central bank reluctant to ease is a recognizable pattern.

KOSPI’s Record Day: The Global AI Demand Signal

Perhaps the most striking move of the week came not from the U.S. but from South Korea. The KOSPI surged 17.91% on July 31 — the largest single-day gain in the index’s history — closing at 6,595.45^[3]. SK Hynix hit the 30% daily limit and Samsung gained 27%, with 7 trillion won in foreign purchases on the day^[3].

The catalyst was twofold: Amazon’s earnings call explicitly referenced confidence in AI semiconductor sales through 2027, and South Korea’s June export data showed a 199.5% increase, reaching $44.8 billion^[3]. This came after a 22.4% decline over the month of July, making the rebound a classic short-squeeze fueled by fundamental demand confirmation^[3].

The KOSPI surge is not a sideshow. It is a parallel confirmation that the AI infrastructure buildout has a global supply chain, and that supply chain just received a demand signal from the hyperscalers.

What to Watch Next

The week of August 3-7 brings a data-heavy calendar that will test whether the Fed’s deferral was wise:

1. NFP on Friday, August 7. The July jobs report is the last major labor data point before the September FOMC. With unemployment at 4.2% and consumer sentiment at 49.5, a soft print would ease hike pressure; a hot one would amplify the three dissenters’ case.

2. ISM Manufacturing (Monday) and ISM Services (Wednesday). Industrial production is running at just 1.14% YoY^[6]. A manufacturing contraction print would reinforce the growth-cooling narrative.

3. Earnings from Palantir (Monday), AMD (Tuesday), and SanDisk (Wednesday). These are the first AI-adjacent reports after the hyperscaler gauntlet. They will test whether the “proof-of-demand” standard established this week extends beyond the Magnificent Five.

4. The 10-year yield. At 4.67%, it is the pressure point. If it continues rising on growth and inflation concerns, equity multiples — particularly high-momentum AI names — face a direct headwind.

5. Geopolitical risk. Oil fell 5.14% this week^[1] but remains up 21.9% for the month^[1] on U.S.-Iran tensions and Strait of Hormuz disruption reports. Any escalation feeds back into the inflation loop the Fed cannot seem to close.

The base case coming out of this week is that the AI bull market remains intact but has raised its admission price. The market will no longer accept capex on faith. Companies that can demonstrate signed backlog, accelerating cloud revenue, and forward visibility will be rewarded. Those that cannot will be discounted. The probability that September brings a rate hike rather than a cut is, by this writer’s estimate, closer to 40% than 20% — the three dissents, the sticky CPI, and the bond market’s own verdict all point in that direction. But the 60% case is another hold with continued ambiguity, and that is not a benign outcome for risk assets priced for a friendly Fed.

FN2 Research provides market commentary and education, not personalized investment advice. All performance figures are historical and sourced from the references cited.