Gemini 3 × The 'Power Hell' of AI Infrastructure — What Are the Conditions for the Next AI Boom to Truly Last? _ Last Week's Market
In November 2025, Google (hereinafter Google) announced its latest generative AI model, 'Gemini 3,' sending shockwaves through the AI industry. It is widely considered to surpass OpenAI's ChatGPT and its successors in performance, response speed, and multimodal capabilities, with many corporate executives and technical experts stating that 'there is no turning back now.'
Meanwhile, the power demand for the data center clusters essential to AI evolution is expanding rapidly, raising the possibility that future growth could be hindered by 'power infrastructure.'
In this article, we will consider what the arrival of Gemini 3 signifies and the infrastructure issues supporting the AI boom—specifically the constraint of 'electricity'—along with recent data.
1. Gemini 3 — Will the 'Second AI Railroad' Change the World?
1-1. The Superiority of Gemini 3 and Market Reaction
On November 18, 2025, Google officially released Gemini 3. It features enhanced capabilities across the board, including reasoning, multimodal processing (image/video/audio), and even design support. It recorded numerous scores surpassing other models in industry benchmarks.
There are reports that over one million users tried it within 24 hours of its release, leading to the assessment that 'the power map of the AI race has completely changed.'
A prominent tech executive described Gemini 3 as a 'leap forward' and said, 'I can't go back to ChatGPT,' causing industry sentiment to shift rapidly.
Thus, many stakeholders are positioning Gemini 3 as the 'second railroad' (a new infrastructure supporting the AI era), and the lead in AI is beginning to shift.
1-2. Why the 'Second Railroad'?
This metaphor likens the situation to the railroad boom of the Industrial Revolution. In other words, while AI has the potential to fundamentally rewrite society and industrial structures, it also carries the risk of 'overinvestment → bubble → collapse,' much like the railroads of that era.
And now, Google, an 'existing giant player,' has seriously changed course to catch up with or overtake the frontrunner, OpenAI. This has ignited a full-scale battle for dominance in an AI landscape previously defined by 'first-mover advantage.'
2. Behind the AI Boom — Electricity as the 'True Bottleneck'
2-1. Data Center Power Consumption is Expanding Rapidly
With the spread of AI, energy demand for data centers is surging. According to multiple studies:
Since the mid-2020s, global power consumption from data centers has been growing at an annual rate of approximately 12–15%.
Forecasts suggest that by 2030, the amount of electricity consumed by data centers will reach nearly double current levels, accounting for a significant portion of global power consumption.
In the United States, power demand for data centers is expected to grow by 22% year-on-year in 2025, with some estimates suggesting that about three times the current amount of electricity could be needed by 2030.
In short, the development of AI—and the expansion of data centers—is not just a matter of 'increasing processing power,' but a situation that also requires the rapid expansion of power infrastructure.
2-2. Can Renewable Energy Alone Keep Up?
Currently, many data center operators rely on existing power grids for their electricity supply, and power derived from fossil fuels remains the primary source. One report suggests that more than 56% of this power is dependent on fossil fuels.
If the number of AI-focused data centers continues to grow at this rate, we may face a dilemma where the securing of new power sources and the large-scale adoption of renewable energy cannot keep up, leading to power grid overloads or increased CO₂ emissions.
3. The AI Dream and Infrastructure Reality — Conditions for Compatibility
3-1. Can 'High-Speed Processing Power' and 'Sustainability' Coexist?
Powerful models like Gemini 3 will significantly expand the scope of AI applications. Examples include bulk analysis of vast documents, multimedia analysis involving images, video, and audio, and large-scale planning and design support—truly embodying the potential for 'AI to exceed human limitations.'
On the other hand, behind this lies social and environmental costs such as massive power demand, greenhouse gas emissions, and water consumption. For instance, GPU clusters used for AI model training and inference often consume tens of megawatts of power.
How to resolve this contradiction—that is the biggest theme in future AI infrastructure development.
3-2. Direction of Necessary Measures
Several possibilities are being discussed:
Expanding the adoption of renewable energy and power derived from renewables
— Connecting wind and solar power directly to data centers—though this faces issues regarding stable supply and cost.Utilizing small modular reactors and on-site power generation
— Integrating power generation and consumption near data centers without relying on existing power grids. A challenge aimed at stabilizing and improving the efficiency of power supply.Efficient power operation and supply-demand management
— Dynamic power management based on AI load, peak shaving, and optimization of cooling efficiency. Recently, operational research using short-term power demand forecasting has also been advancing.Guidance through policy and legal systems
— Institutional design for the entire power sector, such as construction permits for data centers, mandates for renewable energy, and the introduction of carbon pricing, is essential.
4. Will the Golden Age of AI End as a 'Mirage,' or Will It Become a New Foundation for Civilization?
What powerful AI models like Gemini 3 bring may not be just technological innovation, but the reconstruction of information infrastructure. The future where all intellectual tasks—searching, designing, creating, and analyzing—can be handled by the 'universal tool' of AI is undoubtedly approaching.
However, behind this, massive costs in 'power,' 'environment,' and 'infrastructure development' are piling up. If these cannot be managed successfully, the 'AI boom' could hit a wall in an unexpected way.
Economic leaps and environmental/social sustainability—achieving both is the key to making the coming AI era truly valuable.
We may now be standing at the entrance to a second Industrial Revolution—but it is not technology, but rather infrastructure and social choices that will determine our destination.
