Thesis
The Moat Is Memory, Not Intelligence
Every fund is buying the same intelligence. The durable advantage isn't a smarter model — it's institutional memory that compounds deal by deal.
Every fund is buying the same intelligence. Anthropic, OpenAI, Google, and the frontier labs are racing each other toward zero marginal cost per token. Within eighteen months, the model your fund uses will not distinguish you from the fund next door. It never will again.
The advantage GPs have left is not access to a smarter model. It is what accumulates behind it: the judgment a firm builds by running deals, working with partners, and living through cycles. That accumulation is memory. Unlike intelligence, memory cannot be bought off a shelf. It has to be built, deal by deal, inside a firm.
MIT's 2025 NANDA research found that 95% of enterprise generative AI pilots delivered no measurable P&L impact, despite $30 to $40 billion in enterprise spending. The report's central finding was not a model problem. It was a memory problem: the tools evaluated did not learn from, retain, or adapt to the organizations using them. They reset with every session.
Private equity is repeating this pattern with more sophistication and higher stakes. FTI Consulting's 2026 Private Equity AI Radar found that 95% of funds report their AI initiatives meeting or exceeding expectations, a number that should make an LP more skeptical, not more comfortable. Those business cases were often conservatively scoped from the start, and a self-graded pilot with modest goals will succeed almost by definition. Meanwhile, Allvue's 2026 GP Outlook found most GPs still run core workflows on fragmented, manual, Excel-based systems, the exact condition MIT identified as the reason AI tools fail to compound into anything durable.
The pattern holds everywhere it has been measured: tools that retain nothing at session end are optimizing the wrong layer. A due diligence copilot that summarizes one CIM well is useful once. It is not more useful the fiftieth time, because it does not remember the forty-nine that came before it: the pricing patterns that emerged, the diligence questions that actually mattered, which operating partners were right and which weren't.
That is the distinction Pegasus is built around. Not a faster way to read a document. A system that gets sharper with every deal a fund runs through it, because it retains the judgment that produced the outcome, not just the outcome itself.
The firms that separate from the pack over the next fund cycle will not be the ones with the best model subscription. They will be the ones whose institutional memory compounds faster than their competitors', because that compounding, unlike model access, cannot be replicated by writing a check.
The first brief is useful. The fiftieth is a different product entirely.