December 20, 2025 — Data center stocks have become one of the market’s most crowded (and consequential) intersections: artificial intelligence demand on one side, and real-world constraints—power, land, permitting, and financing—on the other. The investment story isn’t just about more GPUs and bigger server halls. It’s increasingly about who can secure megawatts, finance the build, and turn capex into durable cash flow.
This weekend’s headlines sharpen that tension. Regulators in Georgia approved a massive electricity-generation expansion to serve data centers. In the Mid-Atlantic, federal regulators pushed PJM to clarify rules for AI-driven large loads—especially when they’re colocated near power plants. And dealmaking across data center assets hit a record high in 2025 as investors keep chasing “digital infrastructure,” even while valuation and debt-funding concerns simmer. [1]
Below is the full, Dec. 20, 2025 snapshot of current news, forecasts, and market analysis shaping data center stocks heading into 2026.
The Big Picture: Data Center Stocks Are Now a Power Story
For most of the last two years, the market treated “data centers” as a clean, linear AI trade: chips → servers → networking → racks → leases. In late 2025, the storyline is messier—and more realistic.
Three structural forces are now driving day-to-day sentiment across the sector:
- Hyperscaler capex keeps rising (even after repeated underestimates), suggesting demand is real—but also raising questions about ROI and free-cash-flow pressure.
- Grid constraints and large-load politics are tightening, turning interconnection queues, tariffs, and rate structures into headline risk.
- Financing and valuation discipline is returning, especially for projects that rely heavily on debt or require “perfect execution” to work.
Goldman Sachs Research put it bluntly this week: investors aren’t rewarding all AI spenders equally anymore, and they’ve rotated away from parts of the AI infrastructure complex where operating-earnings pressure and debt-funded capex are concerns. [2]
Today’s Top News Driving Data Center Stocks
1) Georgia approves a $16.3B generation expansion tied to data centers
Georgia’s Public Service Commission unanimously approved Georgia Power’s plan to expand electricity generation by roughly 50%, a program priced at $16.3 billion and designed largely for data center growth (reported as 10,000 MW of additional generation with about 80% tied to data centers). Critics warned that if projected demand fails to materialize, ratepayers could be left holding the bag; supporters argue large users will pay most costs and that spreading fixed costs could ultimately help bills. [3]
Why it matters for stocks:
This is a reminder that the AI buildout is now colliding with utility regulation. It can benefit grid-exposed names (generation owners, T&D buildout, equipment suppliers), but it also increases political scrutiny over who pays—and how fast projects can connect.
2) FERC orders PJM to clarify rules for AI data center connections near power plants
The U.S. Federal Energy Regulatory Commission directed PJM (the nation’s largest grid operator) to implement clearer rules for connecting AI-driven data centers and other large loads, particularly when those loads are located near generating facilities. FERC cited concerns about reliability, consumer costs, and ambiguity in PJM’s current tariff framework. [4]
Why it matters for stocks:
This is the regulatory front line for the “power bottleneck” narrative. Any rule changes that speed interconnections (or force stricter cost allocation) can shift the economics of data center development—and ripple into REIT leasing pipelines, developer margins, and equipment demand timing.
3) Record data center dealmaking in 2025—over $60B through November
Global data center dealmaking hit a historic high in 2025, topping 100 transactions totaling nearly $61 billion through November, according to S&P Global Market Intelligence (as reported by Reuters). The surge is tied to AI-driven compute demand and investor appetite—especially private equity—amid limited supply of “high-quality” assets. [5]
Why it matters for stocks:
M&A volume reinforces the “institutionalization” of data centers as an asset class. But record activity often coincides with late-cycle concerns: high entry multiples, competitive bidding, and debt structures that look great in spreadsheets—until rates, utilization, or power prices move the wrong way.
Forecast: Hyperscaler Capex in 2026 Looks Enormous—and Still Rising
The single most important macro variable for data center stocks is not “AI hype.” It’s capex from a handful of hyperscalers and AI platforms—because that spend cascades into the entire supply chain.
CreditSights: Top-5 hyperscalers projected at ~$602B capex in 2026
A CreditSights analysis projects capex for the top five hyperscalers rising from about $443B in 2025 to around $602B in 2026 (a 36% increase), after a major jump in 2025 versus 2024. The same analysis estimates ~75% of 2026 capex could be AI-related infrastructure—roughly $450B—and highlights a continued shift toward leasing data centers rather than owning everything outright, plus broadening financing structures (IG bonds, private credit, project finance, even GPU leasing). [6]
Goldman Sachs: 2026 capex consensus up to $527B—and estimates keep being revised higher
Goldman’s Dec. 18 note says the consensus estimate for 2026 capex among hyperscaler AI companies reached $527B, up from $465B at the start of the third-quarter earnings season—continuing a pattern of upward revisions. Goldman also notes that capex estimates have been too low for two years running, and that supply bottlenecks or investor appetite may constrain spending more than balance sheets. [7]
Market implication:
This is simultaneously bullish and destabilizing. Massive capex supports revenue visibility for chips, networking, cooling, and power gear. But it also raises the bar for proof of payback—especially when capex is funded through debt or pressures free cash flow.
Power Demand Outlook: The Constraint That Keeps Tightening
BloombergNEF: U.S. data center power demand could reach 106 GW by 2035
BloombergNEF projects U.S. data center power demand could reach 106 GW by 2035, a forecast Utility Dive reports as 36% higher than BNEF’s April estimate. The article also notes the U.S. had about 25 GW of operating data centers in 2024 (citing Bloom Energy) and that a meaningful share of newly announced projects are extremely large (over 500 MW). At the same time, analysts caution that speculative proposals and “AI bubble” dynamics can inflate load forecasts. [8]
PJM’s capacity market is flashing red—again
PJM’s latest capacity auction cleared at the $333.44/MW-day price cap, and the grid operator procured capacity below its reliability target. Utility Dive reports that a 5,250 MW increase in PJM’s demand forecast was a major driver—and that it was almost entirely driven by data centers. [9]
Why this matters for data center stocks:
Power isn’t just a cost line item. It’s becoming a gating factor that determines which metros can grow, how quickly projects can deliver, and whether developers must pay new tariffs or build dedicated energy solutions.
The New “Large Load Tariff” Era: Who Pays for the Megawatts?
Utilities and regulators are increasingly trying to separate “serious” data center demand from speculative interconnection requests.
A Utility Dive report summarizing Enverus analysis says dozens of U.S. utilities have adopted new large-load tariffs, with some tailored to data centers. The article highlights examples where new rate structures can materially raise early-year costs for large facilities—and potentially reduce speculative interconnection requests. [10]
Separately, Latitude Media reports that as 2025 ends there are at least 65 large-load tariffs pending or in place across 34 states, with a large number introduced this year (citing Smart Electric Power Alliance tracking). [11]
Stock-market read-through:
- Winners tend to be companies selling the “picks and shovels” of grid buildout (switchgear, transformers, conductors, substations, backup power, thermal management), because tariffs and queue reform usually don’t eliminate demand—they reshape timelines and cost allocation.
- Higher-risk names are those whose economics assume frictionless interconnection, cheap power, and uninterrupted permitting.
Stock Catalysts Across the Data Center Supply Chain
1) Chips: Nvidia stays the heartbeat, but geopolitics is back in focus
Nvidia remains the anchor of the AI data center stack. Reuters reported its data center segment sales rose to $51.2B in the quarter ended Oct. 26, exceeding analyst expectations, and the company forecast quarterly revenue of $65B, also above estimates. [12]
But late 2025 also shows how policy can swing narratives:
- Reuters reported the U.S. would allow exports of Nvidia’s H200 processors to China while collecting a 25% fee on such sales, and separately that Nvidia was evaluating additional capacity for H200 amid strong China demand. [13]
- Meanwhile, reporting about Chinese access to advanced Nvidia chips via cloud infrastructure highlights the ongoing cat-and-mouse dynamic around export controls and data center geography. [14]
What investors watch: sustained shipment volume, supply-chain execution, and any tightening/loosening of export rules that changes the global data center build map.
2) ASICs, networking silicon, and the “AI margin debate”: Broadcom in the spotlight
Broadcom forecast quarterly revenue of about $19.1B on strong AI-related demand, but warned that margins could dip due to AI mix—sending shares lower on the day. [15]
Reuters also reported that the selloff reflected investor sensitivity to profitability optics even when demand is strong, noting Broadcom’s AI margin commentary helped reignite “AI bubble” fears in the broader market. [16]
Why it matters for data center stocks:
The market is moving from “growth at any price” to “growth with quality.” That shift can amplify volatility around earnings calls across the entire AI infrastructure ecosystem.
3) Memory: Micron’s HBM-driven outlook signals continued tightness
Micron’s latest outlook points to a market where high-bandwidth memory (HBM) demand remains a structural tailwind. Reuters reported Micron forecast quarterly revenue around $18.7B and said it would boost fiscal 2026 capex to about $20B as AI demand pressures supply, with capacity constraints cited into the future. [17]
Why it matters:
Memory is a “second-order” AI bottleneck. Tight supply can support pricing power for memory makers, but it can also create delivery constraints for server OEMs and cloud deployments.
4) Data center landlords and operators: strong demand, but execution is everything
For public-market exposure, investors often gravitate to data center REITs and listed operators because they offer (in theory) steadier cash flows than single-project developers.
At the same time, research cited this week suggests the demand-supply balance may tighten unevenly:
- A Goldman Sachs framework summarized by ITPro projects AI workloads could increase their share of the data center market, with scenarios ranging from sustained tightness to oversupply if AI demand falls or enterprise cloud spending slows. [18]
Where the market is most focused:
- lease signings and renewal spreads,
- preleasing rates for new capacity,
- and, above all, power availability in top metros (which can be the real limiter on “pipeline to revenue”).
5) Developers and “AI cloud challengers”: financing risk is no longer theoretical
If 2024–mid-2025 was about announcing mega-projects, late 2025 is about proving they can be financed and delivered.
Reuters reported Oracle’s planned $10B Michigan data center project faced uncertainty after Blue Owl funding talks stalled, underscoring how deal terms and capital markets can reshape AI buildout timelines. [19]
At the same time, Reuters reported major private capital is still committing to hyperscale projects—such as Pure Data Centres (backed by Oaktree) unveiling a roughly €1B Amsterdam hyperscale campus, with phased completion expected from 2028. [20]
Takeaway for data center stocks:
Capital is available—but it’s getting more selective. The highest-confidence projects increasingly come with clearer tenants, clearer power, and clearer financing.
The Emerging Political Risk: Data Centers and Household Power Bills
As data center load becomes more visible in consumer electricity prices, political scrutiny is rising.
The Guardian reported that three U.S. senators (Elizabeth Warren, Chris Van Hollen, Richard Blumenthal) are investigating whether data centers are contributing to higher power costs for consumers, sending letters to major tech companies and data center operators and requesting details on energy use, incentives, and related lobbying. [21]
What investors should infer:
Even if demand stays strong, the “social license to operate” is becoming a factor. Delays, stricter tariffs, and changing incentive structures can all affect the timing of revenue recognition across data center stocks.
So… Are Data Center Stocks Still an “AI Trade” in 2026?
Yes—but it’s evolving into a more segmented market:
Likely beneficiaries if capex stays elevated
- Compute & memory leaders with clear order visibility and execution.
- Networking and connectivity suppliers benefiting from higher-bandwidth architectures.
- Power and thermal management infrastructure tied to high-density racks and faster retrofit cycles.
- Select landlords/operators with power-secured capacity in supply-constrained metros.
Where risk is rising
- Projects dependent on cheap, fast interconnection in congested regions.
- Highly leveraged developers that need tight credit spreads and perfect occupancy to meet return targets.
- Names priced for uninterrupted hypergrowth, where even a “slowdown from extreme to merely high” can compress multiples.
Goldman’s own framing captures the mood: the timing of a capex growth slowdown is a valuation risk for infrastructure companies, and investors are increasingly looking beyond infrastructure to later “phases” of the AI trade. [22]
What to Watch Next Week and Into Early 2026
If you’re tracking data center stocks into the new year, these are the catalysts most likely to move the group:
- Grid rulemaking and tariffs
Watch PJM filings, FERC actions, and how large-load tariffs spread—because they directly change project economics and timelines. [23] - Hyperscaler capex guidance revisions
Consensus has been climbing; any “upward surprise” can re-rate the supply chain, while any hint of discipline can hit the most crowded infrastructure positions. [24] - Financing conditions for mega-projects
The gap between “announced” and “funded” is widening in some places; markets will reward projects that show tenant certainty, power certainty, and low-cost capital. [25] - Power-price and capacity-market signals
PJM’s auction dynamics and state-level utility proceedings (like Georgia’s) are increasingly leading indicators for where data center growth can happen fastest—and at what cost. [26]
References
1. apnews.com, 2. www.goldmansachs.com, 3. apnews.com, 4. www.reuters.com, 5. www.reuters.com, 6. know.creditsights.com, 7. www.goldmansachs.com, 8. www.utilitydive.com, 9. www.utilitydive.com, 10. www.utilitydive.com, 11. www.latitudemedia.com, 12. www.reuters.com, 13. www.reuters.com, 14. www.barrons.com, 15. www.reuters.com, 16. www.reuters.com, 17. www.reuters.com, 18. www.itpro.com, 19. www.reuters.com, 20. www.reuters.com, 21. www.theguardian.com, 22. www.goldmansachs.com, 23. www.reuters.com, 24. www.goldmansachs.com, 25. www.reuters.com, 26. www.utilitydive.com
