Dear readers,
On Thursday we wrote that the equity market’s thesis — AI monetization and geopolitical relief outweighing the gravitational pull of higher rates — was being tested in real time. We noted that the bond market had a different opinion and that both could not be right indefinitely. On Friday, the bond market collected.
The S&P 500 dropped 1.24 percent to 7,408.50. The Nasdaq Composite fell 1.54 percent to 26,225.14. Brent crude surged to nearly $109 a barrel. The 10-year Treasury yield climbed to 4.597 percent. U.S. consumer inflation printed at 3.8 percent for April. And the probability that Kevin Warsh’s Fed delivers zero rate cuts in 2026 now sits at 70 percent on the futures curve.
The record highs we reported Thursday? Gone in a single session. The market that wouldn’t flinch finally did.
But the selloff obscured something worth paying close attention to: the companies building the physical infrastructure for artificial intelligence are posting numbers that have nothing to do with sentiment and everything to do with demand.
Applied Materials Delivers the Proof
We flagged Applied Materials’ earnings as the key read on semiconductor capital expenditure trends on Thursday. The results exceeded even optimistic expectations. AMAT reported earnings per share of $2.86 against a consensus estimate of $2.68, on record revenue of $7.91 billion. Gross margins hit their highest level in 25 years. The company’s Semi Systems segment is projecting top-line growth above 30 percent for 2026. Management raised the quarterly dividend from $0.46 to $0.53.
Institutional money is not waiting for a pullback. Northwestern Mutual increased its AMAT position by more than 380 percent in the fourth quarter, bringing its stake to $648.2 million. Deutsche Bank lifted its price target to $550. The stock closed Friday at $432.79 — punished by the broad selloff, not by anything in its own earnings report.
The distinction matters. Chip designers capture headlines. The companies that build the machines those designers depend on are capturing cash flow.
$250 Billion in Hyperscaler Capital Has to Go Somewhere
The demand signal feeding Applied Materials and its peers originates in a capital spending cycle that has no modern precedent. Oracle’s annual CapEx has risen 431 percent since 2022, reaching $35.5 billion in 2025 — equivalent to 58.1 percent of total revenue. Microsoft increased its spending 236 percent over the same period to $83.1 billion. Amazon remains the largest single investor at $131.8 billion.
That capital is converting into physical assets at extraordinary speed. CBRE reports that 62 percent of data center investment is flowing into opportunistic new construction, driven by a projected $7 trillion AI infrastructure requirement through 2030. PowerHouse Data Centers secured 145 acres in Spotsylvania, Virginia, for an 800-megawatt campus this week and announced plans with Poe Companies for the first hyperscale data center in Kentucky at 400 megawatts. A separate $5 billion joint venture with Blue Owl and Chirisa Technology Parks will deliver an initial 120 megawatts in Richmond, Virginia — capacity already reserved exclusively for CoreWeave’s GPU cloud operation.
These are not speculative bets on future AI revenue. They are contractually committed facilities with named tenants and defined power allocations.
The Electricity Bill Nobody Budgeted For
Every megawatt of new data center capacity requires power that the existing grid was not designed to supply. Monitoring Analytics, the oversight body for the largest U.S. wholesale electricity market, published numbers this week that quantify the strain: wholesale power prices in that region jumped from $77.78 per megawatt-hour in the first quarter of 2025 to $136.53 in the first quarter of 2026 — a 75.5 percent increase that the agency attributes directly to new AI data center load. The regulator characterized the price increase as “not reversible” and called for technology companies to bear their own infrastructure costs.
Thermal management is compounding the problem. Rack densities in modern AI deployments now reach 30 kilowatts, a figure that overwhelms traditional air cooling. Vertiv responded this week with the global launch of its CoolLoop RDHx, a rear-door liquid cooling system rated for up to 80 kilowatts per rack. The product addresses a real bottleneck: without adequate cooling, the GPUs that justify these billion-dollar campuses cannot run at full capacity.
The energy constraint is not a theoretical concern. It is showing up in utility bills and regulatory filings right now.
Geopolitics Handed the Sellers Their Excuse
Friday’s selloff had a trigger beyond inflation data. Tensions around the Strait of Hormuz pushed crude to three-year highs. President Trump, returning from Beijing, stated that President Xi Jinping had agreed that Tehran must reopen the strait. Trump floated lifting U.S. sanctions on Chinese oil companies as a reciprocal gesture. The combination of $109 oil and 3.8 percent consumer inflation created a macro environment in which rate-cut expectations effectively collapsed.
The irony is sharp. The same geopolitical thaw with China that reopened the chip spigot — Nvidia H200 sales to Alibaba, ByteDance, Tencent, and JD.com, as we reported Thursday — is entangled with Middle Eastern energy politics that are raising the cost of powering the data centers those chips will fill.
What Comes Next
All eyes turn to Tuesday, May 20, when Nvidia reports first-quarter results for fiscal year 2027. Analysts expect revenue as high as $78.5 billion and detailed guidance on the mass production timeline for the Vera Rubin platform, scheduled to begin in June. Those numbers will either validate the hyperscaler spending cycle or raise questions about whether demand is peaking.
The broader investment case is becoming clearer by the week. Pure-play AI software companies face compression as the market demands proof of monetization. The physical layer — semiconductor equipment, power infrastructure, cooling systems, data center real estate — is generating revenue and margin expansion that show up in quarterly filings, not pitch decks. Applied Materials just demonstrated that. Vertiv is shipping product into a cooling crisis. PowerHouse is signing tenants before pouring foundations.
The selloff on Friday repriced risk across the board. It did not reprice the fact that someone has to build, power, and cool the facilities where artificial intelligence actually runs. That trade remains intact. Whether it can withstand $109 oil and a Fed that refuses to cut is the question heading into next week.
We hope you enjoy the rest of your weekend.
Best regards,
The StocksToday.com Editorial
