Dear readers,
Yesterday we sorted Big Tech into three camps — companies converting infrastructure spending into cloud revenue, companies asking for patience, and Apple, which is barely spending on infrastructure at all and returning $100 billion to shareholders instead. That framework was useful for earnings season. It is incomplete for what comes next.
The S&P 500 closed at a new record of 7,230.12 on Friday. The Nasdaq 100 finished April with a 15.6 percent gain — its best month since 2002. But the stocks driving the next leg of this rally are not writing code. They are pouring concrete, pulling copper cable, and bolting liquid-cooling manifolds onto GPU racks. The AI trade is migrating from the application layer to the physical layer, and the investment map is being redrawn accordingly.
The Bottleneck Is Watts, Not Weights
Tower Semiconductor (TSEM) has become the clearest illustration of where the money is moving. The stock has surged 329 percent since August 2025, pushing its market capitalization to $22 billion. The thesis is straightforward: AI data centers are extraordinarily inefficient at converting electricity into useful computation, and Tower, under CEO Russell Ellwanger, built a solution. A new high-speed optical switching technology, developed in partnership with Coherent Corp. (COHR) for silicon photonics, landed Tower a deal with Nvidia for 1.6-terabit data center modules. Revenue in that segment went from $28 million in 2023 to $238 million in 2025. Management is targeting $750 million in net income by 2028, which at current prices implies a forward price-to-earnings ratio of roughly 35.
The capital flowing into physical cooling tells the same story from the demand side. Blue Owl Capital, Chirisa Technology Parks, and PowerHouse Data Centers just closed a $750 million transaction to build new facilities for CoreWeave on a campus in Chesterfield, Virginia, with an initial capacity of 120 megawatts. The key specification: direct-on-chip liquid cooling that runs nearly twice as efficiently as conventional air systems. When cooling accounts for 30 to 40 percent of a data center’s electricity bill, halving that line item is not an engineering novelty. It is a margin event.
Gigawatt Campuses and Their Discontents
The scale of what is being planned has no precedent in commercial real estate. In Carlisle, Pennsylvania, a $15 billion hyperscale AI campus led by Pennsylvania Data Center Partners and PowerHouse Data Centers will draw 1.35 gigawatts of power, expandable to 1.8 gigawatts. PPL Electric Utilities is committed to delivering the first 300 megawatts by the second quarter of 2027. Nebius (NBIS) hit an all-time high this week after Missouri approved its first gigawatt-scale “GPU-as-a-Service” superfactory in Independence. Nebius revenue grew 479 percent in 2025 to $530 million, though the adjusted loss widened to $447 million — a price-to-sales ratio of 41 that prices in a future where renting GPUs by the hour becomes as routine as renting cloud storage.
The bill, however, is arriving at residential mailboxes. Electricity costs are climbing in at least 13 U.S. states. In Georgia, average household bills have risen from $150 to $225 per month over two years. Investors searching for alternatives are finding that alternatives carry their own risks. NuScale Power (SMR), the small modular reactor developer that was supposed to be the nuclear answer to AI’s energy appetite, has fallen 75 percent from its 52-week high. Regulatory approval and federal support have not translated into a single completed reactor sale.
Europe Wires Up
Germany is making its own infrastructure bet, though the mechanism is fiscal rather than entrepreneurial. The federal government is acquiring a 25.1 percent stake in Tennet Deutschland through KfW, the state development bank, for approximately €3.3 billion. Tennet operates the largest of Germany’s four high-voltage grids — roughly 14,000 kilometers of transmission lines — and the investment is designed to accelerate grid modernization for the energy transition. The transaction is expected to close in the third quarter of 2026, with the federal government guaranteeing risks and expecting dividend income in return.
On the efficiency side, a British pilot program offered a glimpse of what software can do before new power plants come online. A data center running Nvidia Blackwell Ultra GPU clusters reduced AI power consumption by 40 percent using “Emerald AI” software that dynamically adjusted computing loads, responding in real time to over 200 simulated grid events. If those results hold at scale, they represent a meaningful bridge between current grid capacity and the gigawatt demands now being planned.
Crypto: ETF Inflows Do the Heavy Lifting
Bitcoin broke through $78,000 on Friday, reaching approximately $78,324. The driver is institutional rather than speculative. U.S. spot Bitcoin ETFs recorded $2.44 billion in net inflows during April. On May 1 alone, inflows hit $629.8 million, with BlackRock contributing $284.4 million. The consolidation around $77,000 that we noted in yesterday’s macro frame has resolved to the upside — for now.
In traditional hardware, Qualcomm (QCOM) reported second-quarter earnings of $2.65 per share on revenue of $10.59 billion, beating the consensus estimate of $2.55. Third-quarter guidance of $9.2 billion to $10.0 billion came in below expectations, but the stock rose anyway — a market willing to forgive a soft guide when the beat is clean.
The Macro Frame: Hawkish Hold, Troop Withdrawals, and OPEC Fractures
The Fed held rates steady on Wednesday and struck a hawkish tone, reinforcing the message that cuts remain distant. Geopolitics added its own friction. U.S. Defense Secretary Pete Hegseth ordered the withdrawal of 5,000 troops from Germany — a move SPD defense spokesperson Siemtje Möller called “lacking strategy” and Defense Minister Boris Pistorius described as “foreseeable.” In oil markets, OPEC+ — now minus the United Arab Emirates, which exited on May 1 — agreed in principle to a modest production increase of 188,000 barrels per day, even as the Iran conflict continues to disrupt shipping through the Strait of Hormuz.
The Takeaway
The April jobs report and ISM services data land in the coming days, and both will matter for the rate path. But the deeper shift this week is structural, not cyclical. For three years, the AI trade has been a software story — model architectures, token economics, inference costs. It is now becoming a physical story. The companies capturing the next wave of investment are the ones solving for watts per rack, coolant flow per chip, and transmission capacity per corridor. Yesterday we noted that Apple’s asset-light model proved you don’t need to pour concrete to win the AI era. That may be true for consumer devices. For the infrastructure layer underneath them, concrete — and copper, and coolant, and gigawatts — is exactly what is required.
I hope you enjoy the rest of your weekend.
Best regards,
The StocksToday.com Editorial
