AI Data Centers Under Fire: What Investors in NVDA, DLR and EQIX Need to Know

Opening hook: Local backlash collides with 10-100 MW compute campuses

Many AI data center campuses often require tens to hundreds of megawatts of continuous power — commonly in the ~10–100 MW range for large builds — and that scale is now running into organized local opposition. Residents and local candidates rallied in Wilkes-Barre this week against a proposed facility, crystallizing a broader pushback that can delay projects by months or years.

What happened: A Wilkes-Barre rally and a widening pattern of resistance

On a recent evening a group of residents, local officials and municipal candidates gathered in Wilkes-Barre to oppose a proposed data center development, citing concerns about electricity use, traffic and local environmental impacts. The protest echoed similar hearings in smaller towns where developers seek to site 20+ megawatt facilities or hyperscale campuses that top 50 megawatts.

Big tech players are the usual tenants: Amazon Web Services, Microsoft Azure and Google Cloud are the anchor demand drivers, with real estate firms such as Digital Realty (DLR) and Equinix (EQIX) or specialized operators like Switch (SWCH) supplying the buildings. GPU demand from Nvidia (NVDA) and AMD feeds the compute side, tying chip cycles to real estate and utilities.

Why it matters: Permitting, power and political risk can hit timelines and margins

Data center planning cycles already run long, often 12 to 24 months for major campuses from site selection to construction start, though timelines can vary and sometimes extend further. Local opposition can add another 6 to 18 months via hearings, zoning battles and additional environmental studies, pushing capital deployment and revenue recognition out further. For REITs like DLR or EQIX that price in steady lease-up, a multi-quarter delay at a 50 MW campus represents tens of millions in deferred cash flow.

Power is the hard constraint. A 50 MW facility draws roughly the same steady power as 40,000 average U.S. homes, straining distribution systems and forcing utilities to negotiate new supply or upgrades. Utilities and municipalities may demand infrastructure contributions or higher tariffs, raising operating costs. For hyperscalers, incremental power cost of $0.01 to $0.03 per kWh can erode margins on AI workloads that already cost millions per training run.

History matters. In the early 2010s, concentrated pushback against fossil-fuel infrastructure and large data center campuses in small markets forced multiple projects to relocate or scale down. The pattern repeats: local political mobilization can stall deployment long enough to affect competitive positioning, especially when demand growth is rapid and capacity tight.

The bull case: Demand shock for GPUs and colocation capacity stays intact

Despite local resistance, the underlying demand drivers remain exponential. Large language model training and inference have driven enterprise demand for specialized GPUs, and Nvidia's H100 and successor families are sold out for quarters in many cases. When compute is scarce, hyperscalers will pay premiums for site readiness and grid upgrades, supporting REIT pricing and chipmakers' revenue. If a single campus yields roughly $20–50 million in annualized contracted revenue in some cases, delays are painful but not typically existential.

The bear case: Local politics can reprice risk and shift returns to utilities

On the other hand, repeated permitting delays and costly grid upgrades can lengthen payback periods. Municipalities can extract higher connection fees or impose stricter environmental mitigations, effectively transferring value from landlords and operators to local governments and utilities. If a significant fraction of planned projects face 12–24 month delays, that could reduce near-term capacity additions and create a temporary bottleneck in regions, pressuring tenants' bargaining leverage and REIT valuations.

What this means for investors: Look beyond GPU demand to zoning maps and utility balance sheets

Actionable takeaways: 1) Buy-side focus should be twofold, not just on NVDA and AMD for GPU supply but also on DLR, EQIX and SWCH for real estate execution risk. These REITs trade on cadence of lease-up and new campus deliveries, metrics that delays directly hit. 2) Monitor municipal permitting timelines and utility capital plans; projects that require new substations or transmission upgrades carry higher execution risk and longer paybacks, often signaled by public utility filings and interconnection queue positions.

Specific tickers to watch: NVDA for GPU revenue momentum; DLR and EQIX for exposure to hyperscaler leases; SWCH for concentrated hyperscale campuses; AMZN, MSFT and GOOGL as demand-side anchors; utility partners like NextEra (NEE) or local regulated utilities where projects are proposed.

Investment stance: We remain constructive on the long-term secular story of AI compute, which supports NVDA and the hyperscalers, while remaining selective on data center real estate. Prioritize REITs with diversified geographies and strong municipal engagement, and size positions to account for 6–18 month permitting volatility. Monitor quarterly disclosures for project-by-project cadence, and treat regulatory headlines as force multipliers for short-term volatility rather than proof the market is broken.

Investor takeaway: Short-term protests and permitting fights can delay revenue and raise local costs, but they do not negate the multi-year demand surge for AI compute; position size and geographic diversification in data center names matter now more than ever.