# The AI "Super Bubble" Warning Is a Filter, Not a Funeral > Source: > Published: 2026-06-28 01:43:07+00:00 [Back to Blog](/blog) # The AI “Super Bubble” Warning Is a Filter, Not a Funeral Published by Pentesty · Market Analysis · AI Security Two of China's most respected hedge fund managers just told their investors that the global AI rally has become a “super bubble” and that the collapse point may not be far away. One of them, Yang Dong of Wealspring, has credibility on this kind of call. He flagged the top of the market right before the 2007 crash. The other, Shanghai Banxia, went further and said the trigger for the bubble to burst has already appeared. When people who called the last crash start using words like “super bubble,” it is worth listening. But not for the reason most headlines want you to read it. This is not a story about AI being fake. It is a story about which AI was real all along. ## What the Funds Are Actually Warning About Read past the scary headline and the argument gets specific. Wealspring's complaint is that a huge slice of AI companies lack a durable competitive moat, run fairly ordinary business models, and need constant capital spending just to keep growing. Banxia pointed at slowing revenue growth at frontier model labs as the early crack. That is not a critique of artificial intelligence. That is a critique of valuation without fundamentals. The funds are saying the same thing investors say at the top of every cycle. Too much money chased too many companies that had a buzzword and a burn rate but no defensible business underneath. Notice what they did not say. They did not say AI does not work. They did not say the technology is a trick. They said the market priced a lot of companies as if continuous capital injection equals a real business, and that assumption is about to get tested. Every bubble works this way. The dot-com crash did not kill the internet. It killed the companies that had a domain name and no revenue. Amazon and Google walked out the other side stronger because they were solving a real problem people paid for. The crash was a filter. It separated the businesses with actual cash flow from the ones running on narrative. The same logic applies to AI capability warnings. When [frontier models demonstrate real exploitation capabilities](/blog/frontier-ai-hacking-security-posture-gpt5-2026) in Chromium and Firefox, the question is not whether the capability is genuine. It is whether the business built around it has genuine demand on the other side. ## The Question Every AI Company Should Be Asking Right Now If a correction is coming, the only thing that matters is which side of the filter you land on. There are really two kinds of AI companies in this market. The first kind uses AI as the pitch. The product is “we have AI” and the value proposition lives in the demo and the funding round. When capital tightens, these companies have nothing to fall back on because the AI was the whole story. The second kind uses AI as the engine for a job people already needed done and already paid for. The customer does not care that it is AI. They care that the problem got solved faster, cheaper, or better than the alternative. For these companies, an AI winter is not an extinction event. It is a clearing of the field. The test is simple. If you stripped the words “AI powered” out of your pitch, would anyone still buy? If the answer is no, the bubble warning is about you. If the answer is yes, the warning is your opportunity. ## Why Security Is on the Right Side of the Filter Here is where cybersecurity sits in this picture, and it is not a coincidence that it sits where it does. Nobody buys security because it has AI in it. They buy security because the cost of not having it is catastrophic and concrete. A breach has a number attached to it. Regulatory fines, customer churn, incident response, ransom payments, legal exposure, the deal that died in due diligence because the buyer's security review failed. The ROI of finding a vulnerability before an attacker does is not a projection on a pitch deck. It is the breach that did not happen. That is the difference between spending that survives a downturn and spending that gets cut first. When budgets tighten, companies cut the speculative bets and the nice-to-haves. They do not cut the thing standing between them and a seven-figure incident. Security is one of the few line items where AI just makes an already-essential job faster and cheaper, instead of being the reason for the purchase in the first place. [AI in security](/blog/ai-powered-cyber-attacks-2026) is not a story about a future that might pay off. It is a tool that compresses work that used to take a senior researcher a week into an analysis that runs in minutes. The value is realized the moment the report lands, not three funding rounds from now. That is what real demand looks like. It does not need a bull market to justify itself. This is also why [the quality of security evidence matters more than ever](/blog/why-your-pentest-report-is-lying-to-you) in a tighter market. When boards start asking harder questions, a real vulnerability report with real severity scores is a very different answer from a demo recording and a slide deck. ## What a Correction Would Actually Do to This Space If the funds are right and a correction comes, it would be healthy for serious security companies, not harmful. A downturn clears out the noise. The vendors selling “AI security” as a label rather than a result get exposed fast, because the buyer's first question in a tight budget cycle is “what did this actually catch.” It pushes the whole market toward proof over promises. The report you can hold in your hand beats the demo you watched in a webinar. It also changes how buyers think. In a boom, companies buy security to check a box. In a correction, they buy it because they cannot afford the downside. The motivation gets more serious, the buyer gets more sophisticated, and the products that deliver measurable results win the accounts that the hype machine used to scatter across dozens of vendors. This dynamic plays out across every attack surface. The [cloud misconfiguration problem](/blog/cloud-security-misconfigurations-2026) does not get cheaper to ignore when the market turns. The cost of an exposed storage bucket or a misconfigured IAM policy is the same regardless of where the Nasdaq is trading. And [government-level concern about offensive AI capabilities](/blog/fable5-mythos5-us-government-ban-cybersecurity) signals that the threat side is not slowing down just because valuations might. The companies that survive a bubble are never the ones with the best story. They are the ones with the clearest answer to “what do I get for my money.” In security, that answer is a vulnerability found, a patch shipped, a breach prevented. That answer does not get cheaper to ignore when the market turns. It gets more expensive. ## The Takeaway The Chinese funds may be early or they may be exactly on time. Calling the precise top is a fool's errand and even they admit avoiding AI has been costly so far this year. But the underlying warning is sound, and it cuts cleaner than the headline suggests. The market overpaid for AI companies that have a buzzword instead of a business. That is not a reason to fear AI. It is a reason to separate the AI that is a marketing layer from the AI that is doing real work for customers who would pay either way. At [pentesty.co](https://pentesty.co) we built our pipeline around a problem that existed long before this AI cycle and will exist long after it. Companies ship software with vulnerabilities, attackers find them, and the cost of getting caught off guard keeps climbing. AI is what lets us turn a target into a real, prioritized report in minutes instead of weeks. [But the AI is the engine, not the pitch.](/#servicos) The pitch is the breach that did not happen. If the super bubble bursts, the companies left standing will be the ones solving problems people would pay to solve in any market. Security has always been one of those problems. A correction just makes that more obvious, not less. Related on Pentesty Frontier AI Just Got Better at Hacking → GPT-5.6 Sol demonstrated real exploitation capabilities. Real demand does not care which direction the market is heading. AI-Powered Cyber Attacks in 2026 → How adversaries integrate AI across the kill chain. The threat side is not waiting for valuations to correct. Why Your Pentest Report Is Lying to You → In a tighter market, boards ask harder questions. A real vulnerability report is a very different answer from a demo. Cloud Security in 2026 → Misconfigurations do not get cheaper to ignore when the market turns. The exposure cost is the same. The Day the US Government Shut Down the World's Most Powerful AI → Government-level concern about offensive AI capabilities is not correlated with market sentiment. The threat moves regardless. TL;DR References [1] Wealspring's Yang Dong warns global AI rally has become a super bubble — *South China Morning Post* [2] Shanghai Banxia: the trigger for the AI bubble to burst has already appeared — *Reuters* [3] Yang Dong's 2007 China market top call and track record — *Bloomberg* [4] AI sector valuation analysis: price-to-revenue multiples vs. fundamental moat — *Bloomberg Intelligence* [5] Gartner forecast: enterprise cybersecurity spending remains non-discretionary through downturns — *Gartner* [6] IBM Cost of a Data Breach Report 2026 — *IBM Security* [7] AI investment flows and correction risk: lessons from the dot-com cycle — *Financial Times* [8] Slowing revenue growth at frontier AI labs: the early cracks — *The Information* Want to see the AI that is doing real work, not running on narrative? [Request early access to Pentesty](/#pricing).