● Power Blackout Looms, AI Data Centers Choke, Grid Bottleneck Beats GPUs
The ‘Power Crisis’ of 2026 Determines AI Investment Success: More Urgent than GPUs – Power Infrastructure (Cooling, Onsite Generation, SMR) Explained
This article includes the core points below.
1) Why the scenario of “having GPUs but unable to run the data center” is actually occurring
2) Why ‘electricity’ is the most formidable variable in the cost structure (construction cost vs. IT equipment vs. operating cost) of data centers
3) The transition from air cooling to direct-to-chip (D2C) cooling to immersion cooling, and why cooling is synonymous with Power Usage Effectiveness (PUE)
4) Investment points in the ‘power value chain’ including onsite generation (gas turbines, fuel cells, energy storage systems) and SMR (Small Modular Reactor)
5) The real point that other news doesn’t often talk about: “The bottleneck in the power grid (transmission, substation, distribution) absorbs CAPEX, changing the winners”
1) News Briefing: In 2026, the AI Bottleneck is ‘Power,’ Not ‘GPU’
Core Point
AI has two bottlenecks: the first is GPU supply and the second is power supply.
Currently, there are reports from the United States saying, “The data center is built, but it can’t be operated due to a power shortage.”
The rapid expansion of generative AI outpaces the power infrastructure expansion, making power bottlenecks a more structural issue.
Market Background (Where Money Flows)
Spending on data center construction has already reached billions of dollars, and big tech repeatedly raises its capital expenditure (capex) above expectations during earnings announcements.
Amidst the debate over the ‘AI bubble,’ it is a fact that there is concentration in AI infrastructure investment regardless of the bubble.
SEO Point Inserted Naturally
This trend is particularly important as ‘AI infrastructure’ remains a priority for investment despite inflation and interest rate levels centered in the United States, and the risk of a global economic recession.
This is likely to continue until the power infrastructure bottleneck is resolved, regardless of how U.S. Federal Reserve policy stances change.
2) Data Center Money Structure: “GPUs Consume CAPEX, and Electricity Consumes OPEX”
① Initial Investment (CAPEX) Structure
The largest portion of data center construction costs is IT equipment (servers/GPU/network).
Thus, GPU supply chains like Nvidia take a significant share of CAPEX profits.
Next is construction cost (building/electrical room/mechanical room/piping), with land having a relatively low proportion (usually in the outskirts/industrial complex).
② Operating Costs (OPEX) Structure
Operating costs are essentially mostly ‘electricity bills’.
Therefore, AI data centers are both “hardware competition” and “power procurement competition.”
Summary
The problem is not the expensive GPUs, but rather the structure where the more GPUs you buy, the more urgent electricity becomes.
3) Why Power is a Problem: AI Data Centers are ‘Heat Monsters’, Not Just ‘Electricity Hogs’
Power Consumption Increase → Heat Increase
AI data centers use much more electricity than traditional data centers.
When a lot of electricity is used, heat inevitably increases significantly.
Actual Regional Concentration: Virginia, Texas
In the U.S., data centers are concentrated in certain regions, leading to cases where local power grids can’t handle the load.
Virginia is often mentioned as a leading example of a data center “cluster,” and Texas is quickly following suit.
Conclusion
Data center competitiveness lies more in “power accessibility (substation, transmission line, power contract)” than in “site location.”
4) Power Value Chain in Three Stages: Generation → Transmission → Distribution, and Where Bottlenecks Occur
Basic Structure of the Power Grid
1) Generation: Producing electricity through nuclear, gas, wind, solar, coal, etc.
2) Transmission: Stepping up the voltage at substations for long-distance transmission through the grid
3) Distribution: Stepping down the voltage at substations near demand areas and supplying it to industrial/commercial/residential areas
How Data Centers Exacerbate the Problem
Data centers are “large-scale industrial demand” that suddenly spikes.
Thus, simply increasing the number of power plants doesn’t solve the problem; expanding substation, transmission, and distribution facilities (including permits) takes longer, causing bottlenecks.
5) Cooling System Trends: Air Cooling → D2C (Direct-to-Chip) → Immersion Cooling
Why Cooling is Fundamental
Next-generation GPUs evolve towards higher power consumption (TDP), meaning that if heat is not removed, performance cannot be utilized.
① Air Cooling
It was the standard for existing data centers, but quickly reaches its limits in high-density AI racks.
② D2C (Direct-to-Chip Liquid Cooling)
This method removes heat by directly circulating coolant to the chip.
As AI rack power density increases, D2C adoption becomes inevitable.
③ Immersion Cooling
This minimizes power usage by submerging the entire server in coolant to maximize cooling efficiency.
It is advantageous for extreme PUE (Power Usage Effectiveness), making it more attractive in the “era of power shortages.”
6) Why ‘Onsite Generation’ is Rising: If the Power Grid is Slow, Data Centers Generate Their Own
Definition of Onsite Generation
Traditionally, power is pulled from the grid, but now strategies for having power generation facilities on/near data center sites are spreading.
The reason is simple.
Data centers need to operate stably 24/7, while power grid expansion takes time.
Three Representative Options
① Gas Turbines
They are relatively quick to install and allow for flexible output.
There are increasing reports of satellite images showing gas turbines next to data centers.
② Fuel Cells
They generate power through chemical reactions based on hydrogen/gas.
When combined with a low-carbon transition story, they can be advantageous in large customer contracts.
③ Energy Storage Systems (ESS) + UPS
Since “blackouts” are critical to data centers, UPS is essential, and ESS is becoming increasingly important.
Especially in response to grid volatility (peak-time pricing, demand surges), storage is virtually mandatory.
7) Why Nuclear Power and SMR are Re-evaluated as the ‘Baseload of the AI Era’
The Power Data Centers Want is ‘Stability’
In terms of operating rate, nuclear power can provide very stable power supply.
Wind/solar have variability due to natural conditions, making them unsuitable as standalone baseload sources for AI data centers.
Advantages of SMR (Small Modular Reactors)
Traditional large nuclear plants have long construction periods and high project risks.
SMR offers advantages with modular production, reducing installation time, and potential onsite forms near data centers.
But the Reality Investors Shouldn’t Miss
SMR holds great promise, but commercialization, licensing, and supply chain establishment remain variables.
Thus, within the “power infrastructure theme,” it’s essential to differentiate between areas with immediate revenue potential (cooling, power equipment, substation/distribution, auxiliary power plant facilities) and promising long-term options (SMR).
8) (Important) A Summary of the ‘Real Core Points’ Not Often Addressed by Other Channels/News
Core Point 1: The Power Bottleneck More Often Occurs in “Power Grid (Substation, Transmission, Distribution)” than in “Generation”
It takes longer to expand substations, build new transmission lines, obtain permits, and deal with local community issues than to build power plants.
As a result, data centers are delayed while waiting for the power grid, leading to detours through onsite generation.
Core Point 2: Cooling Is Not About Cost Saving, It’s About ‘Power Acquisition’
Better cooling efficiency means extracting more computations from the same power contract (PPA/agreed power).
This means cooling technology is less about “saving electricity bills” and more about “overcoming power bottlenecks.”
Core Point 3: In 2026, the Winners Come from the ‘Power Layer that Enables AI’ rather than ‘AI Itself’
Even if GPU supply catches up to a certain extent, sales are delayed when power/cooling/UPS/substation installations get stuck.
Ultimately, the market sets the upper limit of performance based on “power infrastructure build rate” rather than “AI growth rate.”
Core Point 4: In SMR, ‘Timeline’ Determines Investment Performance More than ‘Technology’
SMR is on the right path, but commercialization timing and licensing create extreme stock price volatility.
Conversely, cooling/power equipment/ESS already face real demand, and pricing/lead time is being realized faster.
9) Checklist for Blogs: A Guide to Analyzing the Power Infrastructure Sector
A. Inside the Data Center (Immediate Demand)
– Cooling: Air → D2C → Immersion
– Power Stabilization: UPS, PDU, Switchgear
– Energy Storage: ESS (Peak Response/Backup)
B. Site/Power Procurement (Determines Operation)
– Onsite Generation: Gas Turbine, Fuel Cell
– Power Contract: Long-term Power Contract, Local Power Grid Capacity
C. Power Grid (Where Time Consumes Money)
– Substation Expansion
– Transmission/Distribution Network Expansion
– Permits/Local Regulations
D. Long-term Option (High Expectations with Many Variables)
– SMR (Small Modular Reactor)
< Summary >
The real bottleneck in AI expansion in 2026 is power, not GPUs.
Data center CAPEX is driven by IT equipment, and OPEX by electricity costs.
With the spike in heat generation from AI densification, there is a shift from air cooling to D2C and immersion cooling.
As power grid expansion lags, onsite generation (gas turbine, fuel cell) and ESS/UPS are rising sharply.
SMR is the most suitable power source for data centers but its commercialization timeline is a critical variable.
[Related Articles…]
- Reasons Power Infrastructure Bottlenecks Hinder AI Data Center Expansion
- SMR Commercialization Timeline and Global Nuclear Investment Points
*Source: [ 월텍남 – 월스트리트 테크남 ]
– 26년 10배 오를 유망주, 전력인프라 총정리
● China EV Platform Coup, Huawei and CATL Rewrite the Auto Profit Map
China’s Electric Cars and Autonomous Driving ‘K-Beauty Platform Revolution’: The Moment Huawei and CATL Become the Automotive Cosmax
Today’s article focuses on just four aspects.
1) Explaining why Chinese electric cars can turn a profit in just two years, even for latecomers, through structural analysis.
2) How Huawei aims to become the ‘TSMC of the automotive industry’ with its horizontal platform (HIMA).
3) The real reason CATL is shifting from selling batteries to leasing them (credit risk/standard dominance).
4) How these changes create both pressure and opportunities for Korean companies (car manufacturers, parts, batteries, software), including actionable points.
1) One-Line News Summary: “The Chinese Electric Car Market Has Moved from a Growth Phase to a ‘Mature Platform War’”
The core point is that the Chinese electric car industry is rapidly shifting from ‘vertical integration where one company does everything’ to ‘horizontal division where specialists sell in modules’.
Just as K-Beauty lowered brand entry barriers with ODM (Cosmax, Kolmar) + distribution (Olive Young) + SNS marketing, Chinese electric cars are lowering the entry barriers with the combination of Huawei (autonomous driving/in-cabin electronics) + CATL (batteries/standards).
2) Why Latecomers in Chinese Electric Cars Are Booming (Like Xiaomi Turning a Profit in Just Two Years)
① The market is already mature, so there are virtually no “market education costs.”
Tesla grew by creating the market, but latecomers like Xiaomi can ride on existing demand and infrastructure.
This difference completely separates the speed of turning a profit.
② Battery prices and supply chain efficiency have opened the ‘bottom cost’ of product pricing.
Batteries typically account for about 30% of the cost of an electric car, the largest portion.
In China, prices centered on LFP have dropped to “business viable levels,” allowing latecomers to design prices and margins.
③ With dark factories (automation) and module procurement, the development cycle has been halved.
If traditional automakers take three years for new car development, China can talk about 18 months.
This is not just due to technical prowess but also thanks to an industry structure where modules are bought and assembled.
④ Conclusion: Electric cars are moving closer to being ‘product planning’ companies rather than ‘manufacturing’ companies.
This is the frightening point.
The focus is quickly shifting to “what UX/features/prices to combine” instead of building a new chassis.
3) Huawei: Targeting the “TSMC of the Automotive Industry” with a Horizontal Platform Strategy (HIMA)
If Huawei remains just a company that directly makes and sells cars, there are limits to its expansion.
Instead, Huawei is collaborating with traditional automakers, implementing autonomous driving, electrical architecture, and software packages like ‘Intel Inside’ to set the stage.
Core Functions of HIMA (Harmony Intelligent Mobility Alliance)
① Providing autonomous driving/ADAS stacks: Packaging from sensors (LiDAR, camera) to control software.
② Integrating electronic architecture: Standardizing the in-vehicle computing/network structure.
③ Supporting from design to sales: Not just selling technology but helping commercialization in a “turnkey” manner.
Why this is similar to ‘K-Beauty ODM’
If the platform handles the high-barrier areas (autonomous driving software, electronic integration), finished car companies can launch products quickly based only on the chassis/brand/concept.
Consequently, the competitive density of the entire industry skyrockets.
Economically: ‘Economies of scale + standardization’ happen simultaneously
The bigger the platform gets, the lower the costs, which attracts more companies, reinforcing the standard.
This is a classic network effect, and it is the essence of the game China is playing.
4) CATL: Transforming from a Company that ‘Sells’ Batteries to One that ‘Leases’ Them
This is the sharpest point in the original text.
CATL’s reason for battery swapping is not only for “convenience” but with larger goals like standard dominance + risk structure redesign.
① For battery swapping to work, ‘standardization’ is essential
Like a gas station, it only works when there are a few defined types; if batteries vary by vehicle model, inventory explodes.
Therefore, a dominant company like CATL must be able to say “match this standard.”
② Removing batteries from car prices drastically lowers up-front vehicle costs.
When batteries are ‘included in the car sale,’ consumers face high initial costs, but separating batteries (chassis + battery lease) reduces the entry price.
This is a direct weapon for broadening demand.
③ CATL’s real motive: Reducing the default risk of car companies
The “sell batteries to finished car companies on credit” structure means CATL is affected if car companies falter.
By holding batteries as assets and leasing them to end customers, credit risk is dispersed, and the collection model changes.
④ Protecting the ‘Customer Ecosystem’ in Competition with BYD
BYD internalizes battery production, giving advantages in cost/supply.
From CATL’s perspective, creating weapons for the cars that use their batteries to compete with BYD grows the market.
Standard + swapping + leasing is that weapon package.
5) Why This Structure Aligns Well with ‘Industrial Policy’ from a Chinese Government Perspective
① Industrial restructuring (mergers/consolidations) becomes easier
With platformization, fixed costs (R&D, electronic integration, battery costs) for automakers reduce, as does the burden of working capital.
Even if a failure occurs, the probability of the entire system collapsing is low.
② Leverage from mature domestic market to exports
Once the domestic market matures, price competitiveness, product completeness, and a solid supply chain enhance, boosting international expansion potential.
Entering places like Southeast Asia can be interpreted as market selection matching “technical characteristics” such as the low-temperature weakness of LFP.
6) Global Shockwave: Brand Perception of “Chinese Cars in General” Changes, Not Just “One Company like BYD”
From the perspective of overseas consumers, once they realize BYD is “better than expected,” this evaluation extends beyond BYD to ‘Chinese cars.’
This opens the door for followers like Geely, Xiaomi, and the HIMA alliance models.
This is similar to how the perception shifted from a single K-brand to “K-Beauty in general.”
7) The ‘Most Important Content’ Often Overlooked in Other YouTube/News (Key Blog Insights)
1) Huawei and CATL are not just ‘parts suppliers,’ but players rewriting the ‘risk/revenue distribution structure’ of the automotive industry.
The control over autonomous driving stacks and battery standards seems like a “technology” issue but it actually determines where the industry’s money flows (shift in profit pools).
2) The core of electric car competition is shifting from “finished car brand” to “standard platform.”
If a platform becomes standard, finished cars could enter a stage similar to Android manufacturers in the smartphone era where differentiation is tough.
3) CATL’s battery lease/swapping is closer to a ‘financial business’
When batteries are turned into assets, the focus is not just on manufacturing but on capabilities in recovery, residual value, data, insurance, and maintenance.
If successful, battery companies might be re-rated, moving from simple manufacturing valuations to service/finance models.
4) The essence of ‘18-month development speed’ is not grinding the workforce but a ‘modular supply chain.’
If this is misunderstood, it may end as “China is cheap and fast,” but it’s actually tied to global supply chain restructuring.
8) Implications for Korean Companies: Where Are the Risks and Opportunities?
Finished Cars (Hyundai/Kia, etc.)
Risk: If standards for electronics/autonomous driving solidify into external platforms, differentiation becomes more challenging.
Opportunity: Conversely, if “how well can platforms be combined to design a product” is a strength, then extending release speed/lineup is feasible.
Parts/Electronics Partners
Risk: Growth in Huawei-style turnkey solutions might pressure the intermediate supply chain.
Opportunity: Opportunities to supply key components (sensors, power semiconductors, thermal management) within platforms globally expand.
Battery Value Chain
Risk: If standardization solidifies under a certain camp, competition shifts from specs to “standard competition.”
Opportunity: The domain of battery service (diagnostics, residual value assessments, reuse/recycling) is a new market.
Startups/Software
Risk: If autonomous driving stacks are enveloped by platforms, independent software has less standing.
Opportunity: ‘Mobility services’ like insurance/maintenance/optimization based on vehicle data actually grow.
Macroscopically,
All these changes move together with variables like global supply chain restructuring, competition in the electric car market, the possibility of interest rate cuts, inflationary pressures (cost/pricing), and exchange rate volatility.
Ultimately, “whether Chinese platforms dominate standards” affects export price competitiveness and industry margins.
< Summary >
Chinese electric cars are being reshaped into a ‘horizontal platform’ structure that lowers entry barriers by Huawei (autonomous driving/electronics) and CATL (batteries/standards).
Thanks to this structure, latecomers like Xiaomi can rapidly turn a profit/expand, and the new car development cycle is shortened.
CATL’s battery swapping/leasing is fundamentally about “standard dominance + credit risk reduction + protecting the customer ecosystem,” beyond just convenience.
The core fight is moving beyond finished car brand competition to “a structure where standard platforms capture margins and power.”
[Related Articles…]
- Electric Vehicle Market Competition: Why Chinese Platform Strategy Shakes Global Prices
- Battery Standard War: The Mechanism by Which CATL’s Swapping Changes Industrial Order
*Source: [ 티타임즈TV ]
– “화웨이, CATL이 한국 코스맥스처럼 차 제조플랫폼” (김창현 CEIBS 교수)
