The AI Selloff Is Getting Worse. Nvidia, Dell and Data Center Stocks Are Plunging Hard.

AI stocks fell for a second straight day Thursday, with Nvidia's and Arm's losses roughly doubling from Wednesday; the S&P 500 also declined about 1.5%. Investors are concerned that a Democratic Congress and new state leaders could slow data-center construction. UBS estimates that nearly one-third of U.S. capacity due in 2028 could be delayed if construction pauses spread.

Analysis

The key market mechanism is a possible mismatch between chip delivery schedules and the power-, permitting- and building-ready sites needed to deploy them. If projects slip rather than disappear, near-term orders and revenue recognition can move out, while inventory and utilization risk rises across the data-center supply chain. NVIDIA appears more exposed to deployment timing than Arm, but the article provides no company-specific order or revenue data to quantify that difference; avoid treating the selloff as proof of canceled demand.

The estimate for 2028 capacity is a scenario, not confirmed lost capacity. A pause could redirect spending toward retrofits, denser compute and better utilization of existing sites, cushioning chip demand but slowing demand for new-build infrastructure. Conversely, permitting and grid constraints that persist would make announced capex a weaker predictor of installed compute.

Days: political-risk repricing and broad risk-off flows can dominate fundamentals. Over 1–3 months, watch hyperscaler capex guidance, project approvals and evidence of delivery deferrals. Over 6–18 months, persistent delays would favor operators with usable power and permitted capacity over greenfield projects. Contrarian point: a delayed build is not necessarily a canceled GPU order, and the market may be extrapolating a contingent 2028 scenario too quickly. Thesis weakens if capex plans hold, sites progress on schedule and GPU availability remains tight.

Key Decisions for Investors

• Avoid adding to NVDA or ARM solely on the two-day decline. Treat the move as a risk-premium signal until company guidance or project-level data confirms a demand change.
• For near-term risk control, consider a defined-risk QQQ or semiconductor-sector hedge rather than an outright short in either name; broad-market weakness suggests the exposure is not limited to AI hardware.
• Watch hyperscaler capex guidance, data-center permitting and utility-power availability over the next 1–3 months. Escalate to a bearish NVDA view only if these coincide with deferred accelerator deliveries or weaker orders; the article alone does not establish that.
• Falsifiers: capex plans maintained or raised, construction approvals advancing, and continued tightness in deployed compute. These would support the view that spending is being shifted in time or location, not canceled.