The Age of AI Reasoning: Everything You Need to Know from Energy to Investment

on

 

The Age of AI Reasoning: Everything You Need to Know from Energy to Investment

Hello! Are you curious about how AI will evolve and which investment sectors you should watch in this changing AI landscape?

As AI technology evolves from simple computation to a 'thinking' stage, massive changes are happening—from energy consumption to data center infrastructure. Today, we'll explore the shift from computation to reasoning in AI, what's changing as a result, and which investment sectors are worth watching.

The Evolution of AI: From Computation to Reasoning

To understand AI, you need to know the difference between 'computation' and 'reasoning'. Computation is the process of processing data according to predetermined rules. Classic examples include a calculator computing 2+2=4 or image filters enhancing photos. It's a deterministic process where the same input always produces the same output.

In contrast, reasoning is the process of understanding meaning beyond data and making judgments. Like humans, it grasps context and draws conclusions in uncertain situations. When AI looks at a photo and determines "this is a cat," or understands user intent in conversation to provide answers—that's reasoning. Modern AI, especially Large Language Models (LLMs), is developing in this direction.

Let me explain with an analogy. Think about an apple and a photo of an apple. Computation only compares pixels and colors between the two objects and analyzes them as "similar." But reasoning distinguishes the essence: "This is a three-dimensional real apple, that's a flat image." Using Aristotle's concept of 'form and matter,' AI is now beginning to understand meaning (form) beyond data (matter).

How Reasoning AI Actually Works

So how does reasoning AI work specifically? Traditional AI immediately outputs learned patterns. When asked a question, it generates an answer instantly. But the latest reasoning models take 'thinking time.'

For example, when solving a complex math problem, reasoning AI goes through these steps:

  1. Breaking down the problem into sub-problems
  2. Exploring possible solutions for each step
  3. Verifying intermediate results and correcting errors
  4. Re-examining before deriving the final answer

This process is remarkably similar to how humans solve difficult problems. OpenAI's o1 model and Google's Gemini 2.0 adopt this approach. The goal isn't just fast answers, but accurate and reliable ones.

The AI market is experiencing explosive growth reflecting this trend. It's projected to grow from approximately $63.8 billion in 2025 to $368 billion in 2034—roughly 6x growth. AI services with reasoning capabilities command premium prices, becoming a new growth engine for the market.

Benefits for Humanity and New Challenges

AI reasoning will bring three major changes:

1. Productivity Revolution: AI handles complex analyses like medical diagnosis and personalized education, allowing us to focus on creative work. Doctors can spend time communicating with patients instead of reviewing vast medical literature, and teachers can present customized learning paths for individual students.

2. Problem-Solving Capabilities: AI suggests optimal solutions in climate change and economic forecasting. It considers thousands of variables simultaneously and discovers patterns humans might miss. For example, in urban traffic congestion, it can comprehensively analyze signal systems, public transport scheduling, and weather patterns to suggest optimal solutions.

3. Enhanced Daily Convenience: Natural conversation improves accessibility assistance and real-time translation. AI for the visually impaired can provide specific guidance like "There's a hinged door 2 meters away at 3 o'clock, handle on the right" rather than simply saying "There's a door ahead."

Energy Consumption: The Biggest Challenge

However, challenges exist. The biggest issue is energy consumption. AI reasoning accounts for 80-90% of energy consumption compared to training. Why so much?

AI model training happens once, but reasoning executes every time a user asks a question. When millions access services like ChatGPT simultaneously, reasoning operations repeat accordingly. Research shows a single ChatGPT conversation consumes 10 times more power than a regular search.

Looking at specific numbers, a Google search uses about 0.3Wh of electricity, while AI chatbots like ChatGPT consume 2.9Wh per query. If just 10% of worldwide Google searches were replaced by AI reasoning, the annual additional power consumption would equal Ireland's entire electricity consumption.

Two Balancing Factors

Fortunately, two balancing factors exist:

Reduced Query Effect: When reasoning AI provides in-depth answers, users ask fewer follow-up questions. Traditional search required sequential queries like "iPhone 15 price" → "iPhone 15 lowest price" → "iPhone 15 discount." But Google AI Overviews or ChatGPT provide price, discount information, and purchase comparison all at once. Research suggests total queries could decrease by 30-40%.

Efficiency Improvements: Hardware and algorithms are advancing simultaneously. Google's TPU improved energy efficiency 30x over 10 years. NVIDIA's latest H200 chip doubled performance per watt compared to the previous generation. AI models themselves are becoming lighter too. 'Small giant' models are emerging that deliver the same performance at 1/10th the size.

According to MIT research, AI efficiency improvement rates are catching up with demand growth, suggesting energy consumption could stabilize by 2030. Of course, this must happen alongside renewable energy transition.

Energy Revolution: The Rise of ESS and Renewable Energy

With surging power demand from AI data centers, Energy Storage Systems (ESS) and batteries are gaining attention. The global ESS market in 2025 stands at 92GW/247GWh, worth $28.9 billion. It's projected to exceed 500GW by 2030.

Three Reasons Why ESS Matters

1. Peak Demand Response: AI reasoning uses power instantaneously in bursts, which ESS buffers. Data center power consumption isn't constant. It concentrates during business hours and spikes during large-scale model training. ESS stores energy when demand is low and releases it during peak times. This prevents grid overload and reduces electricity costs.

2. Renewable Energy Complement: It compensates for the intermittency of solar/wind power, enabling stable supply. Solar doesn't generate at night, wind doesn't generate without wind. But AI data centers run 24/7. Without ESS, stable operation with renewable energy is impossible. Large-scale ESS like Tesla's Megapack can support an entire data center for up to 4 hours.

3. Carbon Neutrality: Running AI on clean energy reduces environmental impact. Microsoft aims for carbon negative by 2030, and Google declared it will operate all data centers on 24/7 carbon-free energy by 2030. ESS is key technology for achieving these goals.

Combining Data Centers with Renewable Energy

Google operates the Orion Solar project in Ohio, and Amazon runs large solar farms in Texas, building data centers near renewable energy sources. This isn't just environmental protection—it's an economic choice. As renewable energy prices plummet, purchasing electricity through long-term contracts has become cheaper than fossil fuels.

Countries with vast territories like the US, Canada, and Australia have advantages in this strategy. They can build large-scale solar farms and data centers together on cheap land. Meanwhile, countries like Japan and South Korea are responding with tower-type high-density data centers due to space constraints.

Evolution of Data Center Infrastructure

The physical location and structure of data centers are also innovating. Beyond traditional large building approaches, various experiments are underway.

Underground Data Centers

Companies like Iron Mountain converted Pennsylvania's abandoned mines into data centers. Underground advantages are clear:

  • Natural Cooling: Underground temperatures remain constant year-round, reducing cooling costs by 70%.
  • Physical Security: Limited access enhances security.
  • Noise Isolation: Server noise from thousands of machines doesn't leak outside.

But disadvantages exist too: flood risks, limited scalability, and difficulty responding to emergencies. Currently, they're mainly used for archive data storage.

Subsea Data Centers

Microsoft's Project Natick was an experimental project installing a data center on the Scottish seabed. After 2 years of operation, failure rates were 8x lower than land-based data centers. Reasons include:

  • Stable Environment: Subsea conditions have almost no temperature changes, humidity changes, or vibrations.
  • Seawater Cooling: Infinite cooling water provides high energy efficiency.
  • Renewable Energy Connection: Can directly connect to offshore wind power.

But a fatal disadvantage exists: maintenance is nearly impossible. Hardware can't be replaced without retrieving the container. That's why Microsoft kept this project at the research stage without commercialization.

Future Hybrid Models

Ultimately, future data centers will likely be hybrid:

  • Urban Edge Centers: Small centers for ultra-low latency services (autonomous driving, VR)
  • Suburban Mega Centers: For large-scale operations like AI training
  • Renewable Energy-Connected Centers: Dedicated facilities near solar/wind farms
  • Special Purpose Centers: Utilizing underground (security), cold regions (natural cooling)

Each type will be distributed according to purpose and connected through networks—this will become the standard structure.

Investment Sectors to Watch

⚠️ Disclaimer: The following content is not investment advice and is for informational purposes only. Please consult with professionals before making investment decisions and invest based on your own judgment and responsibility. Past performance does not guarantee future returns.

Now let's dive into specific investment perspectives. Here are 5 sectors worth watching in the AI reasoning era.

1. AI Semiconductors & Software

NVIDIA (NVDA): The absolute leader in the AI chip market. Expected market value of $4 trillion in 2025, with data center sales accounting for 80% of total revenue. GPUs like H100 and H200 are used for both AI training and reasoning. Recently investing in reasoning-specific chip development, expected to maintain leadership in the reasoning era.

Palantir (PLTR): An AI reasoning software platform. Provides data analysis solutions to governments and enterprises, recently strengthening AI reasoning capabilities. Used across various sectors including defense, healthcare, and finance, with 2024 revenue up 30% year-over-year.

2. Energy Storage Systems (ESS)

Tesla (TSLA): Supplies ESS to data centers with Megapack. One Megapack has 3.9MWh capacity; large data centers install dozens. Tesla's energy division revenue exceeded $6 billion in 2024, with significant portion from Megapack. This segment's growth will accelerate as the AI boom continues.

CATL (Contemporary Amperex Technology): China's global battery manufacturing leader. Holds #1 market share in both EV batteries and ESS markets. Shows strength in cost-sensitive data center markets with excellent price competitiveness.

3. Renewable Energy

NextEra Energy (NEE): America's largest renewable energy company, with multiple long-term contracts supplying clean energy to AI data centers. Continuously expanding solar and wind capacity, with Q4 2024 results exceeding expectations thanks to AI data center demand.

First Solar (FSLR): America's largest solar panel manufacturer. Specializes in thin-film solar cells, highly efficient even in high temperatures, making it suitable for US Southwest data centers. Recent AI-related demand brought order backlog to all-time highs.

4. Data Center Infrastructure

Equinix (EQIX): The world's largest data center REIT. Operates 240+ data centers in 70 cities worldwide, showing strong 2025 performance. Vacancy rates at historic lows due to AI demand, with rents also rising. REIT structure offers attractive dividend yields.

Digital Realty (DLR): Along with Equinix, this company dominates the global data center market. Particularly strong in hyperscale data centers for large-scale AI training. Continuing stable growth by signing long-term lease contracts with big tech companies like Microsoft and Meta.

5. Utilities

Constellation Energy (CEG): A clean energy supplier centered on nuclear power. AI data centers need stable 24/7 power, making nuclear optimal. Leader in AI energy supply contracts, including a 20-year power supply agreement with Microsoft.

Vistra (VST): America's largest independent power producer, a portfolio company with natural gas, nuclear, and renewable energy. Played a leading role in the utility sector's 21% rise in 2025, with power contracts with AI data centers driving performance.

Investment Strategy: The 'Picks and Shovels' Approach

What do these companies have in common? Rather than creating AI models themselves, they provide essential infrastructure for AI to operate. In the 19th-century California Gold Rush, the people who made the most money weren't the miners who panned for gold, but the merchants who sold picks and shovels to miners.

Similarly, in the AI era, companies providing power, cooling, and computing infrastructure are more likely to generate stable profits than flashy AI startups. The appeal also lies in being 'proven businesses' already generating revenue while benefiting from AI growth.

Regional AI Infrastructure Competition

Interestingly, the AI reasoning era is also changing regional competitive dynamics.

United States: Holds overwhelming dominance. Vast territory, cheap power, and abundant renewable energy resources form a perfect combination. States like Texas, Arizona, and Nevada are rapidly emerging as data center hubs.

China: Making massive government-led investments in AI infrastructure. Building mega data centers in western regions like Inner Mongolia and Guizhou. While dependent on coal power creating environmental issues, rapidly transitioning to renewable energy.

Europe: Strict environmental regulations and high power costs are weaknesses. However, Northern Europe (Iceland, Norway, Sweden) has competitiveness with cheap hydropower and natural cooling. That's why Microsoft and Google are making large investments in Northern Europe.

Emerging Asia: Singapore is a data center hub despite small territory. Political stability, excellent internet infrastructure, and central Asian location are strengths. But high power costs and environmental regulations limit growth, so recently expanding to Malaysia and Indonesia.

Future Outlook: AI Infrastructure in 2030

What will AI infrastructure look like in 2030? Here are some predictions:

Power Consumption: Global AI data center power consumption is projected to increase from about 100TWh in 2025 to 300-500TWh in 2030. This equals Germany's entire electricity consumption. Due to power efficiency improvements and distribution issues, growth rates will slow.

Renewable Energy Share: By 2030, over 80% of AI data centers will operate on renewable energy. As carbon taxes and ESG regulations strengthen, fossil fuel dependence will become economically disadvantageous.

Decentralization: Dualization of mega data centers and edge centers will progress. Large-scale AI training in suburbs with abundant renewable energy, real-time reasoning in urban areas close to users.

New Technologies: Quantum computing may commercialize for specific AI tasks. Also, neuromorphic chips (chips mimicking human brains) could dramatically improve energy efficiency.

Conclusion: 3 Things to Remember in the AI Era

1. AI is evolving from tool to partner

AI that reasons beyond simple calculation will greatly enhance our productivity and creativity. What matters is the ability to understand and utilize AI. Rather than fearing or rejecting it, we must actively learn and adapt. What AI replaces isn't jobs but 'repetitive tasks.' We can focus on more human work.

2. Energy is the new bottleneck

The key to AI advancement is no longer algorithms but energy. No matter how excellent an AI model is, it's useless without power to run it. ESS and renewable energy are the keys to opening a sustainable AI era. Climate change response and AI development aren't conflicting—they're paths we must take together.

3. Focus investments on infrastructure

Rather than flashy AI models or startups, semiconductors, data centers, and energy infrastructure supporting them can be more stable long-term investments. They're already generating revenue while being directly related to AI. Of course, investment always carries risks, so diversification and long-term perspective are important.

The AI reasoning era is an exciting time where technology, environment, and economics intersect. Beyond simple technological advancement, it's reshaping our entire social and economic structure. Understanding and preparing for this change will become future competitiveness.

How are you preparing for the AI reasoning era? What excites you and what concerns you? Please share your thoughts in the comments. Let's discuss together and prepare for a better future!

References

[1] AI Market Size Statistics (2025-2032) - Exploding Topics

[2] Artificial Intelligence Market - Precedence Research

[3] Global Energy Storage Growth - BloombergNEF

[4] Energy Storage Systems Market - Yahoo Finance

[5] 2025 Global Data Center Outlook - JLL

[6] AI Data Center Power - CNBC

[7] NVDA Stock Price Prediction - Yahoo Finance

[8] Tesla Megapack

[9] NextEra Energy AI Data Center Demand - Investopedia

[10] Microsoft Project Natick - Microsoft News

Comments

Post a Comment