No Big Tech Debt Concerns Until 2028

NICE Credit Rating released a report analyzing the financial statements of the five major big tech companies. The conclusion is that their debt is surging, but they can hold out until 2028.

Kwon Soon-woo introduces the report and shifts the direction of concern. The services we use are structured to model on this, and concerns about what happens if big tech borrows too much money and cannot earn enough to fail are unlikely to materialize until 2028. He says if you are going to worry, you should look at whether companies like OpenAI and Anthropic actually make money, and that worrying about big tech is as unnecessary as worrying about celebrities.

AI end-of-days rhetoric is a recently emerging topic, but the expression appears that it keeps holding us back like a ghost haunting us. The theory of AI investment reduction repeats whenever something comes out, saying it will keep being reduced.

However, Kwon Soon-woo's diagnosis is that whenever earnings and guidance come out, the pattern of increasing again, increasing again, keeps repeating.

It is said that nearly 30% or more of total U.S. GDP is AI investment. There is a point that Korea's double-digit nominal growth rate is thanks to those five companies.

Selling memory, selling power equipment, and selling gas turbines are all ultimately connected to AI investment. These numbers are presented as grounds that there is no need to think complicatedly about an AI bubble.

It is a different concept from accounting profit and appears in the cash flow statement. Kwon Soon-woo explains that by that measure, it is true that more is already going out than coming in.

Since 2024, the annual investment growth rate expanded from 38% to 209%. Investment relative to cash flow also rose from 42% to 95%.

By individual company, Oracle already turned negative on cash flow as of 2025. Amazon and Alphabet each recorded net outflows in the first and second quarters. Meta was positive in the first half but showed a trend of decreasing by 25% compared to the previous period.

The reason to look at this indicator is the fear that liquidity could be cut off. It is not a question of whether there are profits, but a question of what happens if there is no cash to repay.

If you do not look closely at that part, the view emerges that the company will fail because debt is so large.

After analyzing maturity conditions and the like, the credit rating agency found that debt due in the short term is less than about 10% of the total and is divided over 30 years, so there is no major problem for immediate survival.

Amazon also draws talk whenever it issues bonds or does a paid-in capital increase. If it increases capital, it becomes a problem, and if it issues bonds, people make a fuss saying debt is increasing. Credit rating agencies actually rate it highly as preemptively securing liquidity at an appropriate time.

However, it is also pointed out that when liquidity is created, money that must be repaid in the short term is also created.

The fact that Meta's cash is decreasing significantly is a concerning part. Nevertheless, there is an assessment that it is unusual to achieve such performance without commercializing its own data centers, along with an assessment that there is room to secure more liquidity.

The point to note is not default, but that cash flow is not good. Because it invests heavily in AI data centers net, that part is not large, and as a result shareholder returns can decrease. Credit rating agencies keep in mind that shareholder returns at Microsoft and Meta could decrease when making credit ratings. In an urgent situation, repaying money becomes more important than shareholder returns.

Order backlog means customers who have reserved to buy data centers in the future, and it is read as a signal that customers are clearly increasing over the past year. Order backlog means people who have told me they will buy data centers in the future.

It is a signal that many customers are clearly being created over the past year, and looking at cloud alone, it is about five times the level.

Oracle's customers are mostly OpenAI. In terms of contracted promises, Oracle appears to have the highest growth potential, but the reason it also appears most vulnerable is that it is too heavily concentrated on OpenAI.

The burden of supplying services to the order backlog is also pointed out.

The key is how much end users pay. Regarding the parts where end users pay, there are still statements that spending continues to increase.

Places like Anthropic are already showing positive figures. Concerns about big tech investment have been resolved to some extent, and instead end users' ability to pay remains a concern.

How would consumers know which LLM a service company used?

If it is cheap and produces similar results, users can just use it. Korea uses ChatGPT and Gemini, and this is because models and services are linked, so the service used becomes the model.

If another service company offers a well-made service and the model behind it is Chinese, consumers could be replaced without realizing it. However, such a trend has not yet appeared.

These figures are for data centers only, and data centers are also leading the increase. There is also burden on the power grid from renewable energy use.

Insulation treatment, one of the bottlenecks, uses cellulose material even for bolts and nuts and contains no iron, so it is 100% manual.

Reporter Lee Ju-ho met a skilled worker who has done insulation treatment inside transformers for 40 years. It is said that after about five years of training a skilled worker, they are treated roughly like the youngest apprentice, and there was also talk that Chinese and American people cannot do it well because their fingers are thick. Demand suddenly increased, but workforce training has not kept up.

Ultra-high-voltage bushings, a component, have lead times delayed by about three years. Bushings are the wrinkled parts sticking up like antennas on top of transformers, and if you order one, you must wait three years.

High-spec electrical steel sheet is also lacking. There is only one place in the U.S. that makes electrical steel sheet. Even that is not suitable for ultra-high voltage, so about half is supplied from Japan and about half from Korea.

In Korea, POSCO representatively makes this, so Korea is evaluated to be in a very advantageous position in these areas.

The scale is 2,060GW, more than double the 2020 level.

Generation demand waiting for transmission grid connection is 2,060GW. Previously, it was said that about 50GW could be connected per year.

As the waiting list has grown, the outlook emerges that demand for power equipment will last 50 years. The talk of three years means that if it exceeds three years, they will not order.

By U.S. standard, Mexico is first and Korea is next. Quite a few power equipment companies that were in the U.S. moved factories to Mexico due to profitability issues. While Mexico is the traditional No. 1, Korea has rapidly caught up to rise to No. 2.

The cast takes a more aggressive view than the credit rating agencies. They believe the speed of bottleneck shifts could become faster. Now they talk about switchboards next to power plants, but by then, switchboards next to power plants plus transmission will be built, and bottlenecks could emerge on both sides of the switchboard.

After discussing stories prepared around news and reporting, they moved to the NICE Credit Rating report and covered big tech financial analysis and the future status of power equipment.

If you only look at headlines, you wonder whether investment can continue with so much debt, but those who look directly at the finances see that only 10% must be repaid quickly and the rest is spread over 30 years. The summary is that it was time to calmly look at whether this is truly a crisis that could collapse right now.