2. Electric Power Revolution: The "Lifeline" of AI Computing Power

All content is for sharing and discussion only; creation is not easy. Thank you all for your support with a single click.
This article aims to help everyone see future trends, understand underlying logic, and think with tomorrow in mind. I will try to use the simplest language to explain professional terminology. I recommend everyone read carefully. I believe those who finish reading are eager to learn, and after reading, you will surely find the corresponding key.
Why has "electricity" become the "lifeline" of the AI era?
You might not know: the daily power consumption of a large AI data center is equivalent to the total electricity used by 50,000 ordinary households in one day (over 500,000 kWh)! The International Energy Agency warns that by 2026, global AI-related electricity consumption will soar to 1000 TWh, almost equal to Japan's entire annual electricity usage. What's more challenging is that AI computing power's electricity usage pattern is like a "roller coaster"—power consumption spikes instantly during computation and drops sharply when idle. Traditional power grids cannot withstand such "highs and lows."
If we compare AI large models to a "superbrain," then the power system is the "circulatory system." In 2023, the market is focused on GPUs and optical modules; by 2026, the main theme will inevitably be the "electric power revolution"—who can solve the "power supply problem" for AI will hold the next decade's wealth code. Today, let's discuss: how energy storage, HVDC, and solid-state transformers (SST) can support the "energy lifeline" of AI computing power?
Core Link 1: Energy Storage—The "Power Buffer" for AI Factories
You can understand it this way: energy storage is like a "power bank + emergency power supply" for AI data centers, which needs to "store electricity" during low demand periods and "rescue" instantly during sudden power outages. It is the "ballast stone" ensuring stable operation of computing power.
Why has energy storage suddenly become a "necessity"?
In the past, data centers relied on diesel generators for emergencies, but diesel engines take seconds to start, and even millisecond interruptions in AI computing power can cause trained models to become "obsolete." Now, lithium batteries combined with supercapacitors can achieve "seamless switching" in milliseconds, instantly compensating for grid voltage fluctuations. More importantly, electricity costs are a major expense for AI factories—storing electricity at night when rates are low and discharging during peak daytime hours can reduce average electricity costs by 30%. Elon Musk vividly said: "Batteries are like 'energy banks'—saving money at night, withdrawing during the day."
How big is the market potential?
China's "Action Plan for the Large-Scale Construction of New Energy Storage" clearly states that by 2027, new energy storage installed capacity should reach 180 GW, with an average annual increase of 35 GW over the next three years—equivalent to building 35 Three Gorges-scale energy storage projects each year. Globally, AI data centers are now equipped with 2-hour backup energy storage as standard, and Europe and the US are even extending this to 4 hours. This is not just a "policy task" but a "survival necessity" for AI enterprises.
A beginner's reminder: the energy storage industry has shifted from "policy-driven" to "market-driven." When choosing companies, look at their "system integration capability"—companies that only sell batteries have thin profit margins, while those capable of providing "integrated solar-storage solutions" command higher premiums.
Core Link 2: HVDC Power Supply—The "Highway" of Power Transmission
You can understand it this way: traditional data centers use UPS for power supply, like a "rural country road," with about 10% transmission loss; HVDC (High Voltage Direct Current) is like a "high-speed highway," improving efficiency to 98% and reducing copper busbar usage by 45%.
How is the technology upgraded?
Previously, servers used 48V low-voltage power supplies, like "small water pipes," which caused "traffic jams" as power increased; now, AI server cabinets' power has surged from 120kW to 1MW (equivalent to electricity for 1000 households), requiring an upgrade to 800V high-voltage direct current "big water pipes." For example, NVIDIA's Rubin cabinets adopting HVDC have a power supply value ten times higher than before. This is not "tech show-off," but "a must change"—otherwise, cables would be thicker than a person, and the server room couldn't fit.
How big is the market potential?
In 2023, the domestic data center HVDC penetration rate was only 10%, expected to rise to 30% by 2026. Giants like China Mobile and Huawei have begun full-scale promotion, and in the next three years, the HVDC power supply board market will surpass 20 billion yuan.
A beginner's reminder: HVDC is not an isolated device; it must be paired with energy storage and servers. Therefore, prioritize companies that can be "bound" with AI computing power manufacturers.
Core Link 3: Solid-State Transformer (SST)—The "Ultimate Form" of Power Revolution
You can understand it this way: traditional transformers are like "green trains," requiring multiple step-down transfers; SST is like "high-speed rail," using new materials like silicon carbide (SiC) and gallium nitride (GaN) to directly convert 10kV high-voltage grid electricity into 800V DC, reducing losses by half and shrinking size by 60%.
Why is SST considered the "future"?
Deploying SST in a large AI data center can save tens of millions of kWh annually. Even more impressive, it acts like a "smart energy router," capable of connecting to photovoltaic, wind power, and energy storage simultaneously, using solar energy during the day and stored energy at night, achieving "campus-level microgrid" self-sufficiency. "By 2026, SST will be the most explosive mainline in power," and top institutions are quietly laying out plans.
Market potential?
Currently, SST is still in the early stage of commercialization, but estimates suggest that by 2027, the global market size will exceed $5 billion, with China accounting for over 60%. On the policy level, the "Action Plan for the Large-Scale Construction of New Energy Storage" explicitly supports "advanced power electronic technologies," with SST being a key focus.
A beginner's reminder: SST technology has high barriers and is still in the "concept stage." It is suitable for long-term layout; avoid chasing short-term gains.
Investment logic summary: How to seize this wave of "electricity dividends"?
1. Choose "system-level players": prioritize integrated companies with "energy storage + inverters + microgrids" over companies that only sell batteries or equipment, as they are more resilient.
2. Focus on "AI data center orders": for example, a company establishing an AIDC division in 2025 to serve AI computing clients, which tends to have higher gross margins.
3. Beware of "technological route risks": for instance, HVDC may upgrade from 800V to 1500V in the future; solid-state batteries, if mass-produced, could impact traditional energy storage batteries.
Final words for beginners:
The competition for AI computing power is essentially a competition for "electricity efficiency." By 2026, energy storage will solve "stability," HVDC will ensure "efficiency," and SST will address "the future"—these three links are like the "three horses" of the "electric power revolution." Without any one, the race cannot run. Remember: don't just focus on GPUs and optical modules; the power companies silently "feeding" AI may hold the next "ten-bagger" stock.
(Next article preview: Liquid cooling technology—The "necessity revolution" for AI cooling, discussing the logic of the cooling track)