AI Collapse Warning, Who Survives

● AI Revenue Collapse Warning

AI 95% Wave of Failures Warning: Is This a Real Crisis? A News-Style Breakdown of Companies That Survive and Companies That Collapse

The MIT report stating that 95% of AI projects are failing to generate revenue is shaking the market.

As mentioned by a Google vice president, warnings are also growing that companies centered on simple API wrappers and AI intermediary platforms are structurally vulnerable.

Meanwhile, analysis is gaining traction that companies with semiconductor supply chains, inference AI infrastructure, and proprietary data are more likely to survive.

In this article, we will systematically break down what truly matters in the AI industry right now, which companies are at risk, which companies are strong, and the core points that other news reports often overlook.

1. What Is Happening in the AI Market Right Now?

The AI market still looks hot on the surface, but internally, it is rapidly moving into a phase of profitability validation.

Generative AI services are flooding the market on the surface, but far fewer companies than expected are actually making money.

The figure mentioned in the MIT report, “a 5% success rate in monetizing generative AI,” should be read not merely as a number, but as a signal that the survival probability of the AI industry is lower than many assume.

In particular, companies with large investment funding but weak customer retention and recurring revenue are likely to be the first to shake during an economic slowdown.

In other words, the AI bubble debate is shifting from “Is the technology good?” to “Who is actually making money?”

2. The Most Vulnerable Group: Simple API Wrappers

The companies most likely to be shaken first are services that call existing AI models and only change the UI and UX.

Examples include the following types.

PDF summarization services
Character-based conversational AI
Marketing copy generation AI
Legal consultation assistant AI
Psychological counseling-style AI
Writing assistant services

These services may look useful on the surface, but in many cases, their core technology depends on external API connections rather than proprietary models.

The problem lies here.

When model providers strengthen their own features, the differentiation of wrapper services quickly decreases.

From the customer’s perspective, the question becomes, “Do I really need to use an intermediate service?”

Ultimately, this creates a structure with low barriers to entry and high substitutability.

From an economic news perspective, this is a very important point.

Even though there appear to be many AI companies, many of them do not actually have a moat that can defend excess profits.

3. The Second Most Vulnerable Group: AI Aggregators

Another risk group pointed out by Google is aggregators.

Simply put, these are intermediary platforms that bundle multiple AI models in one place and connect them to users.

Examples include structures like these.

Connecting multiple LLM APIs at once
Automatically routing models based on use cases
Providing image, video, and text models on one screen
Bundling multiple AI tools for enterprise customers

This model looks efficient on the surface.

However, in the long run, it is closer to a brokerage fee business than a true platform.

If model companies, the suppliers, strengthen their own consumer-facing products, the reason for middle platforms to exist weakens.

This is similar to distribution economics.

When there are hundreds of thousands of chicken restaurants, delivery platforms are powerful, but when suppliers are compressed into a few giant brands, consumers begin using the brand apps directly.

In other words, if the AI market also becomes a structure where only a few suppliers remain, aggregators’ bargaining power will drop sharply.

The core point here is pricing control.

If model providers control pricing, intermediary platforms will struggle to defend their margins.

4. Companies More Likely to Survive: Semiconductors and Infrastructure

The bottleneck in the AI industry is increasingly shifting from software to physical computing power.

This means that the real winners of the AI era will not only be companies that build good models, but also companies that own computing resources and supply chains.

The sectors viewed most favorably are as follows.

4-1. The Top of the Semiconductor Chain: Nvidia, TSMC, Samsung Electronics, and SK Hynix

Nvidia is at the center of the AI accelerator market.

Demand for GPUs remains strong in both AI training and inference, and because Nvidia has built an ecosystem, it is unlikely to end as a short-term trend.

TSMC is a core production base for advanced process technology and is effectively difficult to replace.

Samsung Electronics and SK Hynix are important in memory, especially in HBM supply.

AI servers cannot operate properly without high-performance memory, so companies controlling the memory supply chain are likely to directly benefit from AI growth.

This is also important from the perspective of the global economic outlook.

As AI investment continues, capital will ultimately flow into chips, equipment, processes, and packaging.

4-2. Leaders in Design and Optimization: Broadcom, Marvell, and Qualcomm

In semiconductors, having a design alone is not enough.

In reality, system optimization, testing, power efficiency, packaging, and use-case-specific customization are all important.

Broadcom and Marvell have strengths in customized chip design and integration.

Qualcomm may secure a strong position in on-device AI and edge inference.

In particular, as AI moves from the cloud into devices, demand for inference chips in smartphones, PCs, automobiles, and IoT devices may increase.

4-3. Companies with Data Moats: Palantir and Recursion

As AI becomes more powerful, public data available for training is gradually becoming scarcer.

In other words, going forward, proprietary data will become a source of competitiveness.

Palantir has strengths in structured data processing across defense, manufacturing, and public sectors.

Recursion is gaining attention for its data-driven competitiveness in biotechnology.

Of course, both companies also carry risks.

However, the market’s focus is clear.

Companies that possess data that AI cannot easily replicate are more likely to survive over the long term.

5. Why the AI Market Is Shifting Toward Inference

Many people still think of AI mainly in terms of “training.”

However, the market has already moved toward inference.

Training is an area that requires massive data centers, electricity, and capital.

Inference, on the other hand, is the stage where AI responds to actual user requests, so it is much more directly connected to commercialization and revenue.

Going forward, a significant portion of AI computing demand is likely to shift toward inference.

The reason this trend matters is clear.

Inference means real usage directly leads to revenue
It is connected to on-device AI
Demand for edge devices increases
Low-power, high-efficiency semiconductors become important

In other words, future AI beneficiaries may increasingly be companies that can “run inference cheaply and quickly” rather than companies that simply “build massive models.”

6. The Most Important Point Other News Often Does Not Cover

The truly important point here is not that “AI is collapsing,” but that value within the AI industry is being redistributed from the top layer to the lower layers.

This point is surprisingly undercovered.

People usually frame the issue as whether the entire AI industry will collapse or not.

In reality, the location of profits is changing.

Previously, it seemed as though apps and services would take control, but now foundational elements such as hardware, process technology, memory, electricity, and data have become far more important.

Simply put, money in the AI era is flowing more toward invisible infrastructure than flashy services on the surface.

This is also a core point for understanding the broader Fourth Industrial Revolution.

Digital transformation may look like software innovation on the surface, but in reality, it grows by consuming power grids, servers, semiconductors, process technology, and data pipelines.

7. Checkpoints Investors and Companies Should Watch Now

When evaluating AI companies, the following questions are important.

Does the company have its own model?
Does it have proprietary data?
Can it differentiate its API?
Does it have customer lock-in?
Does it have pricing control?
Can it reduce inference costs?
Can it survive even if access to external model suppliers is cut off?

If the answer to most of these questions is “no,” the company’s long-term survival probability declines.

On the other hand, companies with strong foundations, such as semiconductors, equipment, memory, cloud infrastructure, and data platforms, have the strength to endure even during market corrections.

Now is not the time to simply buy the AI theme, but a time to examine where profits are actually being generated within the AI ecosystem.

8. Conclusion: Is the 95% Collapse Warning Exaggerated or Realistic?

Frankly speaking, the phrase “95% of AI companies will fail” is somewhat sensational.

However, the direction itself is sufficiently convincing.

That is because the AI industry is already being divided into replaceable services and core infrastructure.

In other words, the companies likely to survive can be summarized as follows.

Companies that control semiconductor supply chains
Companies that own inference infrastructure
Companies that possess proprietary data
Companies with strong on-device AI capabilities
Companies with technological moats and pricing control

Conversely, the risky side is as follows.

API wrapper-centered services
AI-packaged services without differentiation
Aggregators that rely only on intermediary commissions
Startups without proprietary data or models

Ultimately, victory in the AI era is likely to depend not on “how flashy a company looks,” but on “how irreplaceable it is.”

The AI market is now entering a phase of monetization validation.

Simple API wrappers and AI aggregators are structurally weak, while companies with semiconductors, inference infrastructure, and proprietary data are more likely to survive.

The key takeaway is not that AI is collapsing, but that value is shifting from services toward infrastructure and moats.

[Related Articles…]

*Source: Softdragon SOD