“The AI Runaway Creates a Golden Age for Cybersecurity”—The 10-Trillion Parameter Leak and OpenAI’s Cold-Blooded Decision to Abandon Sora

While Anthropic is recording phenomenal revenue growth, OpenAI has effectively abandoned its highly anticipated video generation AI, "Sora." Furthermore, "vanity revenue (ARR)" is rampant in the startup world, and the shadow of the Chinese government is looming over cross-border M&A.

In this article, based on discussions among top U.S. investors, we unravel the "true state of the AI market and investment risks" that cannot be seen through superficial news alone, focusing on four key themes.

1. The 10-Trillion Parameter Leak and the Paradox of "Cyber Defense"

What shook the industry was the data leak of Anthropic's unannounced model, "Claude Mythos."

1-1. The Irony of Security AI Leaking Due to "Human Error"

What was leaked was data related to a 10-trillion parameter next-generation model, which is considered extremely adept at detecting cybersecurity threats. Anthropic explains this leak as being "due to human error," creating a powerful irony where a model specialized in security is leaked due to sloppy management.

Following this news, major cybersecurity stocks such as CrowdStrike and Palo Alto Networks fell sharply for a time. The market panicked, fearing that "the emergence of powerful AI would make existing security companies unnecessary."

1-2. The AI Runaway Creates a "Golden Age of Security"

However, the reality is quite the opposite. In an era where AI agents write code and mass-produce apps 24/7, security loopholes and bugs will increase by orders of magnitude.

“Agents have 1,000 times the productivity of humans, but that also means they create 100 times the mistakes and vulnerabilities.”

In other words, the more widespread "Vibe Coding" (writing code based on intuition) becomes, the more new threats will explode, and the correct market reading is that "an unprecedented golden age is coming for cybersecurity companies."

2. OpenAI’s "Sora Withdrawal" Reveals Cold-Blooded Economic Rationality

While Anthropic shows solid execution, OpenAI is being forced into a major strategic shift (or retreat).

2-1. Massive Computing Costs and Inadequate Revenue

OpenAI has effectively halted development of its much-hyped video generation AI, "Sora." The reason is extremely simple: "It consumes too many computing resources (Compute) while generating too little revenue."

“Finally, the economists and accountants have entered the room and started saying, ‘Let’s allocate scarce computing resources to those who can pay the highest price.’”

Video generation is excellent for entertainment, but compared to coding and B2B enterprise models, the ROI (Return on Investment) is hopelessly mismatched. While the withdrawal of Sora is an admission of OpenAI's strategic failure, it is also a positive signal of the company's maturity in that it has "begun to prioritize profitability."

2-2. The Shift to "Advertising," the Lifeline of Consumer Business

Having abandoned Sora, OpenAI is now focusing on its "advertising business." It has already been reported that it has surpassed $100 million in Annual Recurring Revenue (ARR). With the percentage of consumers paying $20 a month being limited, there is no path left to justify a massive market capitalization like Facebook or Google other than ultimately succeeding with a large-scale advertising model.

3. The "Vanity ARR" Rampant in the AI Market

What investors should be most wary of is the "revenue metric bubble" rampant in the AI startup market.

3-1. "Triple counting of revenue" through token reselling

Many AI application companies do not possess their own proprietary AI models; they are merely calling the APIs (tokens) of Anthropic or OpenAI in the background.

"Anthropic records revenue, Cursor uses that to record revenue, and the startup above them records revenue as well. Every time the same token is resold, the ARR of the entire industry is fictitiously inflated."

Despite gross margins being effectively near zero, many companies claiming rapid "$100 million ARR achievement" are creating the appearance of explosive growth through this "token pass-through." When the time comes to account for profit margins, this house of cards will collapse in an instant.

4. Geopolitical Risk: The End of "Singapore Washing"

Finally, I would like to touch upon the fact that geopolitical risk has reached a level that cannot be ignored in cross-border M&A.

4-1. "Talent hoarding" by the Chinese government

The case of a startup called "Manus," which was acquired by Meta, is causing ripples. The company was originally based in China, but it moved its headquarters to Singapore (so-called Singapore washing) and was acquired by Meta as a US company.

However, according to reports, after the acquisition was finalized, **"the founder is in a state of being unable to leave China (effectively under house arrest) due to the intentions of the Chinese government."** The Chinese government is highly vigilant against the "brain drain" of its own talented AI personnel to US companies.

This incident demonstrates that the method of disguising nationality to accept Western capital has reached its limit, and it proves that "founder nationality and location" will become an extremely critical due diligence item in future AI startup investments.

Summary

The current AI market is a mixture of technological evolution, bubble-like frenzy, and the cold logic of capital.

With Masayoshi Son (SoftBank) taking out a highly leveraged $40 billion bridge loan to acquire OpenAI shares, the concentration of funds toward winners is accelerating. However, it can be said that investors have entered a phase where they must identify traps that are difficult to surface, such as "triple-counted ARR" and "geopolitical risk," and select companies with truly sustainable business models (e.g., essential cyber defense or grounded B2B software).

Recommended Articles

Next Big Wave—Growth Stocks, Seeds of Ideas, and Deep Dives into Trends