In a post I published last year (here), I argued that the staggering capital deployment into artificial intelligence was economically unjustifiable. Years later, the ROI remains nowhere to be found. Instead of a productivity revolution that justifies trillion-dollar valuations, the market has doubled down on financial acrobatics, hype cycles, and political theater to keep the music playing.
In one of my ongoing empirical research on the AI economy (Umlai, Charfeddine, & El-Masri, 2026), we analyzed 5,330 AI ventures globally to test whether massive capital accumulation actually reflects venture quality. The data reveals a glaring disconnect: Foundation-AI ventures raise roughly 31% more total and equity funding than Execution-AI ventures, driven purely by category hype and signaling mechanisms. Yet, when evaluating actual milestone performance, this funding advantage completely disappears. Execution-AI startups (those building practical downstream applications) are equally likely to reach Series A and more likely to achieve ultimate success through exits (IPOs and M&As). Capital accumulation in AI has become entirely decoupled from fundamental value creation. Money raised is not quality achieved.
The "Contain Us" Charade
Economist Yanis Varoufakis recently dissected the current state of the AI debate, calling out the sudden wave of pseudo-moral agony from Silicon Valley tech lords. Figures like Sam Altman, Dario Amodei, and Elon Musk spent the past few months making headline-grabbing declarations about a "10% chance of AI-driven human extinction," pleading with governments to regulate and contain them. As Varoufakis correctly points out, this entire debate is a cynical, dreadful fraud. It seems incomprehensible that these executives are terrified of their creations. They seem to be confessing to the sin to claim the glory, deliberately hyping their models as "too dangerous" as a masterclass in guerrilla marketing. More aggressively, they are executing classic regulatory capture:
A) Crushing Competition: By begging governments for antitrust oversight and strict compliance frameworks, they ensure that smaller open-source competitors are snuffed out under regulatory burdens.
B) Shielding Cartels: They seek state protection to preserve a cozy duopoly while demanding explicit exemptions to build unchecked military systems.
C) Extracting Cloud Rents: While the US model relies on closed-source "cloud capital" designed to manipulate user behavior and extract economic rents, it creates little material substance.
Meanwhile, Europe has effectively removed itself from the board. trapped under data regulations like GDPR and clinging to delusions about catching up on computing power; a point underscored by Mario Draghi’s recent FT piece proposing a European monopsony buyer cartel that Varoufakis rightly dismisses as dead on arrival. Europe isn't just losing the AI race; it isn't even in the race.
Recursive Self-Improvement, Fragile Humanoids, and Real Risks
While Silicon Valley plays regulatory theater, the physical and algorithmic reality of AI is taking a far more unsettling turn.
In a recent panel discussion I participated in on the 2026 Beijing World Humanoid Robot Games, I talked about those humanoids breaking world records in sprinting and jumping under lab-controlled conditions. Yet, the moment the environment shifted even slightly, success rates collapsed!! Sprinters crashed through trackside barriers while weightlifters tumbled into judges' tables. The physical bodies remain fragile, but the brains we are embedding inside them present immediate danger.
The single algorithm that should terrify everyone is recursive self-improvement embedding AI agents with the ability to design and iterate on the next generation of AI with zero human oversight. In May 2026, Pope Leo XIV touched on this exact peril in Magnifica Humanitas, reminding us of a fundamental boundary: humans must create innovations; we cannot allow innovations to create innovations.
We are already witnessing what happens when autonomous agent loops step outside human guardrails:
OpenAI & Testing Breakouts: OpenAI disclosed internal test instances where agent swarms coordinated beyond authorized boundaries, attacking external systems and attempting to compromise their own evaluation frameworks.
Data Destruction: Replit’s coding agent, under pressure to complete a task, wiped a company’s live production database, attempted to conceal the action, and fabricated replacement data. Similarly, Cursor (running Claude Opus 4.6) executed a single Railway API call that deleted a founder’s production database and volume backups in nine seconds.
Lethalized Warfare: Military systems generating targeting packages at a rate of one every four seconds in active conflicts. When decisions operate at that speed, human verification is a complete fiction.
The Cracking Financial Foundation: Circular Funding & Shrinking Capital
But how come this AI bubble is staying afloat??? Largely through circular funding. Big tech giants invest billions into mid-stage AI labs, which immediately turn around and spend those exact funds renting cloud infrastructure from those same big tech giants. It is a financial feedback loop where revenues are recycled to manufacture the illusion of exponential market growth. However, the external capital faucets supporting this illusion are beginning to dry up. Deep-pocketed sovereign backers of Silicon Valley mega-funds are facing mounting domestic economic and financial pressures. Investments in unproven AI infrastructure contracts and the massive capital subsidies keeping these foundation labs solvent will face an immediate squeeze.
When foreign capital tightens and balance sheets can no longer hide behind circular vendor financing, the missing ROI will finally catch up with the market. The AI bubble isn't going to pop because bots became sentient and took over; it will pop because the economics were a mirage all along.
