AI Drawdowns and De-Leveraging: The Cost of Stability
By Nikhil Raghavan · Reporting from San Francisco ·
The conversation highlighted how strong cumulative gains can mask immediate danger posed by extreme AI sector volatility.
The latest episode of "TBPN" offered a masterclass in financial fragility disguised as strategic resilience. While much attention was rightly paid to the breathless reports of AI model efficiency—OpenAI slashing costs for models like Luna, making them highly efficient per task—the most arresting details concerned the mechanics of massive capital flows and risk management at the highest levels of finance.
On the podcast, Leopold detailed a significant drawdown in July, exacerbated by extreme moves in core positions, particularly AI names. To stabilize the fund and protect LP capital, he executed a block transaction, selling a portion of the public equity book to Citadel, thereby closing all shorts and removing leverage risk. He stated clearly: "I take full responsibility for these events." While this level of accountability is commendable, it simultaneously underscored how deeply exposed specialized investment vehicles remain to volatility.
The Illusion of Stability in High-Velocity Capital
The discussion around fund performance was a sharp reminder that market resilience is often an accounting exercise rather than an economic one. Despite the negative month-to-date performance for July (estimated at -67%), Leopold noted that the net year-to-date performance remains up 80%. This juxtaposition—a massive monthly loss offset by strong cumulative gains—is standard finance rhetoric, but it obscures the immediate danger posed by extreme volatility in large caps and AI names.
The conversation touched on the core challenge of fund management: balancing getting the trend right with nailing portfolio construction. The lack of leverage suggested by Leopold’s recent 13F filings was therefore interpreted not as caution, but perhaps as a strategic "learning lesson." This focus on minimizing risk exposure—the shift to managing the public book on a fully paid-for basis—is critical for understanding platform stability in any sector. When capital structures are overly leveraged, the first adverse event risks triggering margin calls, which are fundamentally different from, yet often mistaken for, government bailouts or being "too big to fail."
Capex vs. Topline: The New Bear Market Signal
Shifting focus to technology, the market commentary provided a clear lesson in how investor attention has shifted away from traditional metrics. Major tech companies reported earnings that generally exceeded expectations, but stock movements were often driven less by topline/bottom line beats and more by discussions of quarterly capex forecasts and AI messaging. Meta serves as the perfect example: despite posting stronger-than-expected revenue ($60.8B), its stock fell sharply because its EPS missed analyst expectations, leading investors to focus intensely on its high quarterly capex ($31.1B).
This dynamic—where sheer spending capability dictates valuation regardless of current profitability—is the defining feature of today’s platform economy. The investment thesis is no longer about maximizing profit; it is about demonstrating capacity for continuous infrastructure build-out and technological differentiation. This makes the efficiency gains reported by OpenAI, such as slashing costs on models like Luna, incredibly valuable from a policy perspective. Model efficiency must be measured by cost per task, not just tokens—a crucial distinction that governs whether cutting-edge AI becomes broadly accessible or remains confined to deep pockets.
The Architecture of Progress and Risk
The underlying tension running through the episode is the relationship between technological progress and financial risk. We are witnessing an era where infrastructure investment (capex) has become the primary indicator of value, overshadowing traditional metrics like net income. Whether discussing a hedge fund reducing leverage to protect capital or Meta spending billions on AI data centers, the common thread is managing structural uncertainty. The ARC-AGI V3 benchmark was useful precisely because it offered a tangible way to measure this progress—a visual metric that transcends technical jargon and speaks to general capability.
The market’s current state is not one of fundamental collapse, but rather an extreme re-pricing of risk across platforms. Capital remains aggressive, but its deployment is now highly sensitive to demonstrated efficiency gains and the management of massive capital expenditure cycles. The story of Leopold's drawdown and subsequent defensive maneuvers provides a stark warning: even sophisticated investment structures must prioritize foundational stability over sheer speculative growth.