Power Crunch Persists Despite AI Slowdown, Three Scenarios Examined

Attention is focusing on the fallout if growth in construction investment for artificial intelligence (AI) slows. Stocks related to data center construction have swung between declines and rebounds. Some Big Tech companies are expected to maintain growth in capital expenditure. Even so, caution toward AI-related stocks has intensified in equity markets. Behind this is the view that slowing training demand will lead to curbs on capital investment.

TMTPost’s coverage of Barron’s breaks down the market impact in a slowdown. It lays out three scenarios investors could face. The coverage can be summarized in the following statement.

If AI training slows, Big Tech companies could cut capital expenditure, reducing new data center construction

This outlook would ripple through the entire supply chain. It would affect demand for semiconductors, power equipment, and cooling systems. Valuations of energy stocks would also face reassessment. This article organizes the background and implications based on the published summary.

Data Center Stocks Remain Volatile

Data center-related stocks have seen volatile trading. Debate over threats to humanity amplified market anxiety. Concerns over AI safety weighed as a selling factor. Buybacks later kicked in, with prices recovering at times. The choppy price action reflects a tug-of-war between expectations and fears.

Capital expenditure by Big Tech remains at high levels. Expanding cloud demand is cited as the rationale for investment. Training and inference for generative AI require large-scale computing infrastructure. Power consumption is also larger than for conventional IT infrastructure. Expectations for continued investment therefore remain strong.

Meanwhile, Wall Street is taking an increasingly strict view. A plateau in training demand is starting to be flagged. There is speculation that expansion of computing resources for training has run its course. Concerns have emerged over delays and downsizing of new projects. Some voices also say valuations of related stocks are excessive.

Reading the Cloud Giants’ Investment Plans

Cloud operators have pushed ahead with expanding computing infrastructure. Based on demand forecasts, they have secured land and power capacity. They have also procured power through long-term contracts. Construction requires a long lead time. Changing plans entails significant costs.

Training demand has been supported by increasingly large models. Competition to develop large language models has driven up computing demand. Image and video generation has also increased computational loads. Expanding enterprise adoption has also boosted demand. These trends have justified expanded capital investment.

The slowdown is on the radar because efficiency gains have advanced. Improved training methods have reduced computational requirements. Higher utilization of existing infrastructure has also contributed. Views have emerged that the need for additional investment is fading. Market attention has shifted to whether investment plans can be maintained.

What Are the Three Slowdown Scenarios

Three scenarios were presented. The first is a gradual slowdown. The second is a shakeout of marginal projects. The third is slowing technological progress combined with expanding inference demand. In each case, the impact on energy stocks is the key issue.

In a gradual slowdown, major revisions to plans are unlikely. Cloud operators act with long-term demand in mind. There is little benefit in changing infrastructure plans over short-term fluctuations. Power contracts and land acquisitions will continue. The blow to the supply chain would be limited.

In a shakeout of marginal projects, low-profitability plans stall. Projects in areas where securing power is difficult are reconsidered. Businesses facing tight financing conditions are also affected. Even so, core large-scale projects remain. This is a stage where selection advances across the market.

If technological progress slows, growth in training demand is curbed. The assumption is that model performance improvements slow. Even so, inference demand is expected to keep expanding. Everyday business use underpins power demand. The structural shift between training and inference comes into question.

Power Crunch Seen Persisting

The notable point is the analysis that power shortages will remain. Even if marginal projects are weeded out, supply and demand will not ease. Existing large-scale projects alone create large power demand. Expanding the transmission grid takes a long time. New generating facilities also cannot be built in a short period.

Regional imbalances are also hard to resolve. Computing infrastructure is concentrated in areas where power and cooling water can be secured. Transmission capacity in certain regions is already tight. Expanding substation facilities requires permits and approvals. Coordination with local communities also takes time.

The impact on energy stocks is not uniform. Power generation businesses are supported by stable demand. Demand for transmission and distribution equipment upgrades also continues. Meanwhile, equipment orders dependent on new construction will fluctuate. Investors need to check the business mix of each stock.

Inference Demand Underpins Power Demand

Inference has different load characteristics from training. It contrasts with training, which performs massive computations at once. Inference consumes power through accumulated usage. Constant operation is needed to ensure response speed. Even in distributed setups, the total volume tends to be large.

Expanding business use drives inference higher. Use for document summarization and customer service has increased. Coding assistance and design support have also spread. Audio and video processing also consumes computing power. The broadening user base is the core of demand.

Even if technological progress slows, the impact is seen as small. Use continues even if model updates slow. Many applications can be served by operating existing models. If efficiency gains lower unit costs, usage will increase further. Total power demand is seen remaining resilient.

Editorial Opinion

Turning to short-term impacts, we expect screening of data center-related orders to progress over the next three to six months. We expect more delays for projects with difficult profitability or power procurement. We expect stocks tied to power generation and transmission to remain relatively stable.

From a long-term perspective, we assess that the shift in focus from training to inference will advance. We see inference smoothing demand fluctuations as it consumes power through accumulated usage. If expansion of power infrastructure lags, we see siting constraints determining business success or failure.

We pose questions from the editors. Which uses should be prioritized under power constraints. Who should capture the gains from efficiency improvements. Is the investment slowdown an adjustment of excess, or the start of a demand shift.

References

• “AI发展放缓对能源股意味着什么”, by Barrons巴伦 — 钛媒体, 2026-09-21T15:00:27.000Z (ARR)
• Source URL: https://www.tmtpost.com/8148083.html