Zhongtai Securities released a research report analyzing AI computing demand from an electricity consumption perspective. By conducting high-frequency monitoring of the PJM grid covering key U.S. data center clusters (Virginia and Ohio), the firm observed significant increases in both power load and electricity prices. The report identifies four key thresholds for AI applications and suggests that investment targets capable of overcoming these barriers will see substantially improved success rates. Vertical industry scenarios with proprietary data also present viable investment opportunities, provided the AI-driven added value is sufficiently high to avoid complete displacement by large model companies.
**Key Insight 1: High-frequency power data confirms accelerating AI computing demand.** Monitoring the PJM grid revealed: 1. **Virginia DOM region**: In 2025, the average monthly load increase was approximately 3GW (excluding baseline load), up 0.98GW from 2024. Year-over-year growth in September–November surged by 73%, 53.2%, and 56.4%, respectively. Three grid pricing nodes—ARCOLA (dominated by Google with AWS support), SHILOHDP (AWS-centric), and BOYDTNDP (Microsoft-centric)—showed notable trends: - **Nighttime price differentials rose sharply**, especially in October and November. The ARCOLA node (Google’s primary consumption hub) saw the steepest increase, with October’s differential reaching $7.94/MWh (+197% YoY) and November’s spiking to $13.11/MWh (+680% YoY, +65% MoM), coinciding with Gemini3’s release. - **Price volatility intensified**, with standard deviations at these nodes exceeding PJM’s regional average, indicating grid capacity nearing limits. - **Congestion fees surged**, particularly in November. ARCOLA’s congestion fee differentials far outpaced other nodes, rising 223% and 890% YoY in October and November, respectively.
2. **Ohio AEP region**: The 2025 monthly load increase averaged 1.34GW (excluding baseline), up nearly 0.8GW from 2024. September–November load growth soared by 158%, 223%, and 180% YoY.
**Key Insight 2: AI’s expansion from 1 to N favors Google’s ecosystem.** The report outlines four AI adoption barriers: 1) Weak economies of scale per user → 2) Subscription models limiting user growth → 3) Higher ROI and value-add requirements → 4) Data flywheel effects. While the first three relate to capital and sector dynamics, the fourth hinges on ecosystem integration.
**Google’s AI ecosystem spans**: - Hardware (TPU chips, Tensor G5) - Smart devices (Pixel, Samsung integration) - Cloud infrastructure (Vertex AI) - Software (Android, Chrome, Antigravity IDE) - Applications (YouTube, Search, NotebookLM, Maps, Workspace, Waymo, Project Astra).
Leveraging Gemini’s multimodal capabilities and in-house TPUs to cut computing costs, Google embeds AI across its ecosystem.
**Investment Recommendations**: Targets overcoming the four barriers offer higher success probabilities. Vertical sectors with unique data remain investable if AI delivers sufficient value-add to resist displacement by large models.
**Risks**: 1) Slower-than-expected AI model/application development 2) Technological/environmental shifts invalidating assumptions 3) Delays in data updates.
Comments