Live Review 2: Why AI Stocks Sold Off: The Real Risk May Be Data Center Financing, Not AI Demand

Live Review:AI Selloff, Fed Risks & the Next Investing Opportunity

From data center permits to Nvidia, SoftBank and Micron, the next phase of the AI investment cycle requires a closer look at capital allocation.

When AI stocks experience a sharp correction, the immediate reaction is often to ask whether the AI boom is ending.

But I believe that question is too simplistic.

In my recent Tiger Brokers livestream, I explained why investors need to distinguish between three very different issues: the long-term demand for AI, the practical constraints on building AI infrastructure, and the financial returns generated by those investments.

My central argument is that AI can continue growing while certain AI investments deliver disappointing returns.

That distinction is particularly important when evaluating data centers, semiconductor companies, and highly concentrated AI investors.

1. Both political parties support AI—but disagree over who pays for its expansion

A common interpretation of the U.S. midterm election is that one political party supports AI infrastructure while the other opposes it.

I believe that interpretation misses the underlying policy debate.

Both parties recognize the strategic importance of artificial intelligence. Their priorities, however, differ.

Democrats tend to emphasize electricity affordability, environmental protections, water consumption, workers, and local community interests.

Republicans tend to emphasize construction speed, power availability, and American competitiveness in AI.

The most important disagreement is therefore not whether AI should expand.

It is who should bear the cost of expansion and how quickly infrastructure should be built.

This matters because data centers are not simply collections of expensive GPUs.

They require electricity generation, transmission capacity, cooling systems, water infrastructure, financing, and regulatory approvals.

In my presentation, I used Texas as a case study of how state-level permitting discussions could affect the development of AI infrastructure.

The examples illustrate a broader investment risk: a project can have strong underlying demand but still face delays because its supporting infrastructure is not ready.

Investors should also distinguish between state-level restrictions and national policy. A local permitting delay does not necessarily imply that AI investment is being abandoned nationwide.

2. Data center delays can ripple through the entire semiconductor supply chain

One of the most important connections I highlighted is the relationship between infrastructure approvals and semiconductor demand.

The chain works approximately as follows:

A delay at the beginning of this chain can affect the timing of orders further downstream.

For companies supplying GPUs, high-bandwidth memory, conventional DRAM, and networking equipment, the timing of infrastructure deployment is therefore an important determinant of revenue growth.

However, I want to emphasize a critical distinction:

A delayed project is not necessarily a canceled project.

Investors should examine whether customers are postponing orders, renegotiating contracts, or abandoning planned capacity altogether.

Those outcomes have very different implications for earnings.

3. A fully occupied data center can still be a poor investment

This is perhaps the most important and differentiated point from my presentation.

Many investors focus on the enormous demand for AI computing capacity.

But demand alone does not determine whether an investment creates shareholder value.

A data center generates revenue from capacity contracts, usage fees, and potentially price escalators.

At the same time, its owners must pay for servers, electricity, cooling, maintenance, financing, and infrastructure upgrades.

The fundamental investment test is:

The key test over the asset's economic life

A data center can be fully utilized and still produce inadequate returns if construction costs and borrowing expenses are too high.

This is why higher electricity costs, grid connection expenses, and financing rates matter so much.

Consider two hypothetical projects.

Project A generates a 15% return on invested capital against an 8% cost of capital.

Project B generates an 8% return against an 11% cost of capital.

Both may experience strong customer demand, but only the first clearly earns more than its cost of financing.

The implication for AI investors is straightforward: Revenue growth is not the same as value creation.

4. Nvidia, SoftBank and Micron reveal three different types of AI risk

I used three companies in the livestream to demonstrate why investors should avoid treating the entire AI sector as a single trade.

$NVIDIA(NVDA)$ — Governance and policy risk

The presentation examined reported governance concerns alongside export-control uncertainty. My approach is to separate confirmed findings from allegations and then assess whether developments could materially affect leadership, customers, or access to international markets.

$Softbank Group Corp(SFTBY)$ — Concentration and collateral risk

SoftBank illustrates how large investments in AI companies can create balance-sheet sensitivity. I discussed its reported OpenAI commitments, loan-to-value ratio, liquidity position, and exposure to valuation changes.

$Micron Technology(MU)$ — Earnings and demand-timing risk

Micron illustrates why investors must distinguish between a falling share price and deteriorating fundamentals. The important questions involve memory contract prices, customer orders, HBM demand, and earnings revisions.

For Nvidia, the key is whether company-specific events alter the business's fundamental prospects. Allegations must not be confused with verified findings.

For SoftBank, I emphasized the importance of monitoring financial leverage. The livestream cited a loan-to-value ratio of approximately 15.9% as of September 30. This does not, by itself, establish insolvency risk.

The real question is how a decline in investment values might affect borrowing capacity, collateral requirements, and liquidity.

For Micron, the presentation contrasted a substantial reported share-price decline with expectations for rising DRAM prices.

The apparent contradiction is important. A stock can fall even when industry pricing remains favorable because investors are discounting potential future risks.

I also distinguished between high-bandwidth memory used in AI accelerators and conventional DRAM and NAND markets. Their demand drivers and pricing cycles are not identical.

5. I winners will be determined by financial discipline

I remain interested in the long-term economic opportunities created by AI.

However, I believe investors should increasingly focus on the quality of capital allocation rather than simply the size of announced investment commitments.

Three questions deserve particular attention.

First, are infrastructure projects progressing on schedule?

Second, are customers maintaining their contractual commitments?

Third, can the companies involved generate returns that justify the capital they are spending?

If the answers remain positive, a correction may reflect a temporary increase in risk premiums.

If project delays spread, orders weaken, and financing costs rise, the market may be signaling a more fundamental problem.

My Final Takeaway

The AI boom should not be judged solely by stock prices or headline capital expenditure.

The next phase of AI investing will be about translating enormous technological demand into sustainable financial returns.

That is why I believe investors should monitor permitting, power infrastructure, financing conditions, customer orders, and free cash flow together.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Report

Comment(1)

  • Top
  • Latest
  • Marialina
    ·15:23
    Long-dated, fixed-rate debt could help data centers ride out construction delays. How much financing matures before these projects start generating cash?
    Reply
    Report