Goldman Sachs just admitted the quiet part out loud. AI is breaking the market. Osman Ali, global co-head of Quantitative Investment Strategies at Goldman Sachs Asset Management, suggests the machines are now tripping over their own feet. The signal is dying. Noise is the new alpha.
The Recursive Loop of Algorithmic Failure
Quantitative investment relies on historical patterns to predict future movements. It is a game of math and speed. But the math is changing. As large language models (LLMs) saturate the financial ecosystem, they generate a massive volume of synthetic data. This data is fed back into other models. The result is a recursive loop that distorts price discovery. When every hedge fund uses a variation of the same transformer-based architecture, the trade becomes crowded. Inefficiencies emerge not from human error, but from algorithmic consensus.
The technical term for this is model collapse. It happens when a model begins to learn from its own outputs or the outputs of its peers rather than organic market signals. Per recent Bloomberg market data, the correlation between top-tier quantitative funds has reached a three-year high. This suggests a lack of diversity in the underlying logic of the market’s biggest players. If everyone is looking for the same signal, the signal disappears before it can be exploited.
Visualizing the Decay of Model Diversity
The Friction of Instantaneous Information
Speed used to be a moat. Now it is a commodity. In the 48 hours leading up to May 12, the volatility in mid-cap tech stocks demonstrated a disturbing trend. Prices corrected in milliseconds, often overshooting their fundamental value by 4 to 5 percent before snapping back. This is the inefficiency Osman Ali refers to. The models are so fast that they create vacuum pockets in liquidity. They react to news before the news is even fully indexed by search engines.
Institutional investors are now forced to look for “dark signals.” These are data points that AI cannot easily scrape or interpret. This includes physical supply chain tracking via satellite and private credit flows that remain off-grid. According to Reuters financial reporting, the shift toward discretionary-quant hybrids is accelerating. Pure-play black box strategies are underperforming for the third consecutive quarter. The human element is being re-introduced to act as a circuit breaker for the machines.
The Cost of Sophisticated Analysis
Sophistication comes with a price tag. The compute power required to run real-time sentiment analysis across global markets is skyrocketing. Goldman Sachs notes that while AI enables deeper analysis, the marginal utility of that analysis is falling. We are seeing a massive capital expenditure in AI infrastructure that is yielding diminishing returns in terms of basis points. The infrastructure is becoming more expensive than the alpha it generates.
This creates a barrier to entry that favors only the largest institutions. However, those same institutions are the ones most susceptible to the systemic risks of model uniformity. If a primary liquidity provider’s model experiences a logic error, the contagion spreads instantly. The Securities and Exchange Commission has recently increased its oversight of algorithmic risk management, but the regulatory framework is struggling to keep pace with the deployment of autonomous trading agents.
The Emergence of Synthetic Volatility
Volatility is no longer just a measure of market fear. It is a product of model interaction. When an AI detects a pattern, it executes. If that execution triggers a response from another AI, a feedback loop is born. This is synthetic volatility. It does not reflect a change in the underlying company’s value. It reflects a conflict between two sets of code. For the savvy investor, this creates a new type of arbitrage. The goal is no longer to predict the stock, but to predict the reaction of the dominant models.
The next major milestone for the industry occurs on June 15, when the first batch of mandatory AI-disclosure filings are due for major asset managers. Watch the “Model Drift” section of these reports. It will reveal exactly how much these algorithms have veered from their original parameters in the face of today’s synthetic market conditions.
