The AI spending scare is over. The supply scare is just starting.

For six weeks, the question hanging over the AI trade was whether the biggest cloud operators would keep spending. The 29% selloff in chip and data-center stocks into late July was the market's working answer: maybe not. Then CoreWeave and Nebius printed quarter after quarter of triple-digit revenue growth, the sector caught a bid, and Alger's Brad Neuman named what had actually changed. "The constraint is moving from demand to supply," he said. "It's moving from whether the hyperscalers will keep spending to whether the data centers, the grid, and the chip pipeline can keep up with them." The rebound is real. The bottleneck just moved.

The reset started in late June. From June 22 to July 29, the PHLX Semiconductor Sector index (SOX) dropped 29% as investors questioned whether Microsoft's, Amazon's, and Google's AI capex would translate into revenue. The selloff was a rational repricing of an open question: the cloud giants were committing tens of billions to data centers and AI chips, and there was no public evidence the spend was paying back.

Two earnings reports broke the slide. CoreWeave's Q2 2026 earnings call showed 112% year-over-year revenue growth on a large compute-power backlog, and the stock jumped more than 19% the next session. Nebius's Q2 print showed 450% year-over-year growth on AI cloud-storage demand, and the stock rose 32%. CoreWeave rents GPU clusters to AI labs and hyperscalers. Nebius builds and operates AI cloud infrastructure. Both report a backlog, not a pipeline fantasy.

Hyperscaler monetization was the signal investors had been waiting for. Microsoft's and Amazon's late-July Q2 calls showed AI revenue traction in their cloud segments, the kind of operating evidence that lets a capex line stop looking like a charity. JPMorgan responded by lifting its 2026 S&P 500 price target to 8,000. The demand side, in other words, answered its own question.

"Hyperscalers" is the term for the largest cloud operators, the Microsofts, Amazons, and Googles whose capex sets the pace for the whole AI buildout. Alger, where Neuman is a portfolio manager, holds CoreWeave, Nebius, and Nvidia, so his read is informed but not independent. He argues the constraint has physically relocated. Data-center siting, construction, grid interconnection, and chip supply are now the rate-limiting steps. The demand question was answered by the Q2 prints. The supply question is open.

Nvidia reports Q2 FY27 earnings on August 26, at a market cap of roughly $5.4 trillion. The stock is already down 4.9% from its mid-May high, so the print is positioned less as a victory lap and more as a referendum. If the bottleneck is now on the supply side, the question for Nvidia is not whether AI labs and cloud operators want more chips. They do. The question is whether the company can deliver them fast enough, and whether its own supply chain, foundry capacity, high-bandwidth memory, and advanced packaging, can scale with the data-center pipeline.

The data-center pipeline is one watch item: how many megawatts of new capacity CoreWeave, Nebius, and the hyperscalers can actually bring online in 2026 and 2027, and how fast the grid can connect it. The chip pipeline is the other: whether Nvidia's suppliers can keep delivering at the cadence the AI buildout now assumes. If either lags, the supply side becomes the dominant story of the second half. If both hold, the rebound has a foundation.

The 29% drawdown had a logic to it. The market was pricing in the possibility that the AI capex cycle was a bubble. Two earnings reports and a hyperscaler monetization signal have made that read less likely. The new read: the capex is real, the demand is real, and the question now is whether the physical infrastructure of AI, the buildings, the power, the chips, can be built fast enough to keep up with the spending. Nvidia prints on August 26. After that, the answer belongs to the supply chain.