Private Market FOMO Over AI Will Lead to Mistakes, Bravo Says: How PE Firms Can Navigate the Valuation Minefield

The FOMO Paradox: Why Institutional Capital Cannot Stay Still

The foundation of current private market dysfunction rests on a psychological phenomenon that has driven bubbles throughout financial history: the fear of missing out, or FOMO. But unlike retail investor FOMO, which is often relegated to meme stocks and speculative plays, institutional FOMO in private AI markets carries systemic weight because it involves capital that shapes entire industries[1][9]. Thoma Bravo’s Orlando Bravo identified this dynamic with precision, noting that the “enormous anxiety gripping investors” is leading to decisions made at the expense of thorough due diligence[1]. When every competitor in the market is raising AI-focused funds, when every portfolio company meeting begins with questions about AI capabilities, and when founders know that attaching “AI” to their pitch deck can increase valuation multiples by 40 or 50 percent, the pressure to participate becomes nearly irresistible for institutional investors bound by fiduciary duty to generate returns.

This dynamic creates what behavioral finance researchers recognize as a collective action problem. No individual investor wants to make rushed decisions or participate in what they suspect might be bubble valuations. Yet from the perspective of any single fund manager, sitting on the sidelines while competitors deploy capital into AI opportunities that later compound at 10x or 100x multiples represents an existential threat to fund performance, limited partner confidence, and institutional survival[18]. As one venture capitalist quoted in research on the topic observed, “Every dollar going into venture is going into AI”—and in such an environment, choosing not to participate feels like a career-limiting move rather than a prudent risk management decision[30]. The result is a market dynamic where rational individual actors create collectively irrational outcomes, with capital pooling at the top among mega-rounds to companies like Anthropic, xAI, and Mistral AI while the distribution of capital across the broader ecosystem becomes increasingly concentrated and risky[27].

The human psychology driving this phenomenon is not new, but the scale and speed at which it operates in 2025 is historically novel. In previous cycles—the dot-com boom, the housing bubble, the cryptocurrency surge—there were warning signals and a longer runway before the inevitable correction. This time, the infrastructure for rapid capital deployment has become so efficient, the information flows so instantaneous, and the competitive pressure among institutional investors so intense that the compression from hype to deployment to potential implosion could occur at an accelerated pace[21][5]. The MIT NANDA initiative’s finding that 95 percent of enterprise generative AI pilots are failing, with only 5 percent achieving rapid revenue acceleration, provides empirical grounding to what many sophisticated investors have quietly begun to suspect: the gap between AI’s transformative potential and its near-term monetization reality is far wider than valuations currently reflect[42].

Valuation Disconnect: When Multiple Expansion Becomes Detached From Reality

At the core of Bravo’s warning lies a mathematical problem that cannot be rationalized away through appeals to long-term potential or disruptive innovation. As he told CNBC, a company generating $50 million in annual recurring revenue cannot credibly be valued at $10 billion unless investors believe that company will generate $1 billion in free cash flow, and even then, achieving a full return on capital becomes a generational accomplishment[7]. This is not conservative thinking; it is basic arithmetic. Yet precisely these valuations have become commonplace in private AI markets. Anthropic, valued at $20 billion in its Series D round, is estimated to generate perhaps $300 million to $500 million in annual revenue. This implies a revenue multiple of 40 to 70 times, compared to the 5 to 15 times multiples typical for profitable SaaS companies operating at scale[15][39].

The historical precedent for such disconnects is instructive. During the dot-com bubble, which peaked in March 2000, companies trading at revenue multiples exceeding 50 times became normalized, with participants justifying these valuations through appeals to “new economy” metrics that supposedly rendered traditional financial analysis obsolete[46]. The error was not in recognizing the internet’s transformative potential—that recognition was entirely correct—but rather in assuming that all companies building on that infrastructure would capture proportional value[46]. The dot-com crash saw the NASDAQ decline nearly 80 percent by October 2002, with hundreds of companies losing 80 to 95 percent of their value. Priceline, despite ultimately becoming a successful company, fell 94 percent from its peak[46]. Today’s private AI valuations follow an eerily similar pattern, with the critical difference that this time the capital is not flowing to thousands of diverse startups but rather concentrating among a handful of mega-rounds to foundation model companies and AI infrastructure players[27][30].

Goldman Sachs’ analysis provides crucial nuance to this valuation story. While public AI stocks trade at 30 times forward earnings on average—well below the 50 times multiples of the dot-com peak—private AI companies exhibit what Kostin describes as “unsustainable” valuations driven by “reflexivity” rather than fundamentals[2][13]. The distinction matters because it reveals that the bubble mechanism operating in private markets is qualitatively different from what occurred in public equities during previous cycles. In public markets, there is price discovery through trading, short-selling, and the continuous adjustment of valuations based on updated information. Private markets have none of these mechanisms. Instead, they operate on a series of discrete financing rounds, each priced by the next round of sophisticated investors, with information asymmetries that allow valuations to detach from reality for years before correction forces a reckoning[2][25].

The specific mechanics of this valuation disconnect deserve close examination. When Foundation Capital partner Maria Palma observed that the sector is experiencing “swift technological advancements and the pressure on investors to maintain pace,” she identified the mechanism by which forward-looking models become increasingly speculative[15]. Each new capability announcement from OpenAI, Anthropic, or Mistral AI triggers a new wave of investment thesis updates, revised market size projections, and assumed revenue acceleration curves. These projections, built on necessarily uncertain assumptions about AI adoption timelines and monetization success, then become embedded in valuation models. When those models are presented to investment committees composed of executives paid bonuses based on percentage returns and capital deployment velocity, the incentive structure inevitably skews toward aggressive assumptions and optimistic scenarios[27][28].

Circular Financing: The Structural Vulnerability Beneath the Boom

Perhaps the most sophisticated—and most concerning—aspect of current private market AI investment dynamics involves what financial analysts have termed “circular financing” or “vendor financing,” a mechanism that echoes the structures that preceded both the dot-com bust and the 2008 financial crisis[2][37][38][40]. In this arrangement, large technology companies like NVIDIA, Microsoft, and Oracle make strategic investments in AI startups and foundation model companies. These same startups then deploy the invested capital to purchase the vendors’ products and services—NVIDIA GPUs, Azure cloud capacity, Oracle infrastructure—at premium rates and long-term contracts. The result is that vendor revenue and growth rates appear to be driven by organic market demand, when in reality they are being driven by capital that originated with the vendors themselves, recycled through intermediary investments[37][40].

The scale of these arrangements is staggering. OpenAI has announced up to a $100 billion commitment to NVIDIA for deploying 10 gigawatts of AI computing capacity, equivalent to approximately $500 billion in total capex investment[40]. Simultaneously, OpenAI has secured a $300 billion cloud infrastructure agreement with Oracle and a $10 billion custom-chip partnership with Broadcom[40]. These are not isolated transactions but rather interconnected pieces of an ecosystem where capital flows in complex patterns, creating the appearance of robust market demand when closer inspection reveals that much of the apparent demand is vendor-financed[37][40]. UBS analysis notes that while the OpenAI-NVIDIA arrangement represents significant capital commitment, it accounts for only approximately 13 percent of NVIDIA’s projected 2026 revenue, suggesting that while circular financing exists, it has not yet reached the level where it comprises the entire revenue base[40]. However, this reassurance must be tempered by the recognition that concentration at this level is nevertheless substantial, and circular financing dynamics extend far beyond just NVIDIA into infrastructure players, data center operators, and GPU leasing firms[37][40].

The historical parallel to this structure is the telecom bubble of 1998-2002, when equipment vendors like Cisco, Nortel, and Lucent financed the telecoms companies that were their primary customers, who then deployed that capital back toward the vendors’ products[38][46]. Cisco not only sold routers and networking equipment to long-haul carriers and internet service providers; it financed their purchases through vendor loans, leasing arrangements, and in some cases equity stakes. When the carrying costs of this debt began to exceed the revenue being generated by end-user demand for bandwidth, the system collapsed[38]. WorldCom, Global Crossing, 360networks, and dozens of smaller telecoms filed for bankruptcy, taking with them not only their equity investors but also leaving equipment vendors with massive bad debts and stranded inventory[46]. The difference between that cycle and the current AI cycle is material but not necessarily reassuring. As UBS notes, today’s large technology companies are funding much of their AI capex through robust operating cash flows rather than debt, suggesting greater financial resilience[40]. Yet this advantage could prove temporary if AI infrastructure spending fails to generate proportional revenue growth at the application layer, as has historically occurred during technology cycles where infrastructure investment races ahead of adoption[38][46].

The risk inherent in circular financing becomes acute when examining the dependency structures it creates across the ecosystem. If NVIDIA experiences production constraints or demand disappointments, the GPUs that AI infrastructure startups depend on become scarcer or more expensive, disrupting the carefully calibrated economics of data center operators, who in turn cannot serve AI application companies, who cannot generate the revenues that justify their private valuations[37]. Conversely, if AI adoption at the enterprise level fails to materialize at the pace currently being modeled—a genuine possibility given that 95 percent of pilot programs are failing to drive rapid revenue growth—then the demand for compute infrastructure becomes markedly less robust than current capex projections suggest[42]. In that scenario, infrastructure players find themselves with massive stranded capacity, unable to refinance debt or continue operations at breakeven, forcing asset sales or restructurings that cascade through the ecosystem[37][38].

The Due Diligence Crisis: Separating Real Innovation From AI Washing

Beyond the macro-level concerns about valuations and circular financing, private equity and venture investors face a profound due diligence challenge that strikes at the foundation of dealmaking itself: distinguishing genuine artificial intelligence innovations from what industry observers have termed “AI washing”—the strategic deployment of AI language and capabilities in marketing materials and fundraising narratives without corresponding substantive technical or business model advantages[14][19][32]. This challenge is not incidental to current market conditions; it is central to why experienced dealmakers like Bravo are sounding warnings about FOMO-driven mistakes. When investors are under pressure to deploy capital quickly, when every investment committee meeting includes questions about AI exposure, and when founders know that AI-centered narratives command premium valuations, the incentives for misrepresentation and exaggeration become overwhelming[14][19].

The mechanics of AI washing operate at multiple levels. At the most basic level, companies that generate minimal revenue and have no clear path to profitability attach “AI-powered” language to their pitch decks and suddenly become fundable, raising capital at valuations that would be laughable for non-AI companies[14][19][32]. Software Improvement Group’s research on fake AI has identified specific red flags: vague data strategies where companies promise to “collect user data” without explaining what data, how it will be owned, or how privacy regulations will be managed[32]. This reflects a fundamental misunderstanding or misrepresentation of AI capabilities, since all successful AI applications require specific, high-quality, proprietary data sets that provide defensibility[32]. When founders cannot articulate their data strategy in concrete terms, it typically signals either fundamental confusion about their own business model or intentional obfuscation designed to pass due diligence[32].

More sophisticated AI washing involves companies that wrap third-party foundation models (like OpenAI’s GPT, Anthropic’s Claude, or Mistral’s models) with user interfaces or modest customization, then present themselves as AI companies with defensible moats[32][57]. While there are cases where excellent interface design and go-to-market execution can create real value even atop commodity foundation models—Cursor’s code completion tool is a notable example where workflow embedding has created genuine switching costs despite using underlying OpenAI models—these successes are the exception rather than the rule[57]. The majority of companies attempting this strategy face intense pressure from both incumbent software providers adding AI features and from other startups with similar approaches, creating a race-to-the-bottom dynamic where differentiation erodes quickly[32][57].

The most insidious form of AI washing involves companies claiming AI capabilities that simply do not work or that have fundamental limitations not disclosed in fundraising materials. Historical examples cited in research include the COMPAS algorithm used in criminal justice risk assessment, which was found to exhibit significant racial bias in predicting recidivism, with Black defendants being disproportionately classified as higher risk[14]. When investors conduct due diligence and encounter such biased systems, they face not only investment losses but also profound legal, regulatory, and reputational risks. The EU AI Act and emerging US regulatory frameworks increasingly require explainability, fairness testing, and documented performance metrics across demographic groups[14][26]. Companies that have not built these considerations into their systems from the start face costly remediation, potential regulatory fines, and customer churn once bias or accuracy issues become public[14][26].

Bain & Company’s research on AI due diligence identifies five critical questions that serious dealmakers must answer before committing capital: First, will AI enable revolutionary transformation (entirely new business models), transformation of existing business models, or merely augmentation of current processes? Second, will market volumes be affected, and if so, will pricing power be maintained or eroded? Third, will the basis of competition change in ways that threaten existing moats? Fourth, will returns on capital invested in AI be sufficient to justify the expenditure, especially if competitors implement similar capabilities? And fifth, is the management team actually capable of executing on an AI strategy, or is there a fundamental skills gap?[17]. These questions force investors beyond surface-level pitch deck narratives into the underlying economic and operational reality of AI investments. Too many private equity firms, under FOMO pressure, are skipping this deeper diligence or conducting it perfunctorily, accepting management assertions about AI capabilities without rigorous validation[17][22].

Blackstone’s global head of compliance has identified accuracy and data security as the biggest risks in AI deployment, concerns that cascade directly into due diligence implications[6]. If an AI system exhibits accuracy issues or security vulnerabilities that were not identified during the investment diligence process, the resulting losses can be substantial. These are not theoretical concerns; they represent documented operational risks at leading firms attempting to deploy AI at scale[6]. Yet many PE firms conducting diligence on AI-focused targets are not incorporating the specialized expertise required to assess these risks. Data scientists capable of evaluating model performance, bias detection, and security architecture are expensive and scarce, yet PE firms frequently proceed with AI acquisitions without engaging these specialists, instead relying on salespeople and marketing materials to assess technical capabilities[14][22].

Public vs. Private: Why Markets Are Pricing AI Differently

A crucial insight emerging from current market conditions involves the stark difference in how public and private markets are valuing artificial intelligence. Goldman Sachs’ David Kostin has articulated this distinction with clarity: while public markets have appropriately disciplined the valuations of publicly traded AI stocks through continuous price discovery and earnings-to-price correlation, private markets have allowed valuations to detach from fundamentals in ways that echo classic bubble dynamics[2][13][25]. Nvidia, the semiconductor company most directly exposed to AI infrastructure buildout, has seen its share price increase 12-fold over three years—but its earnings have simultaneously increased 12-fold, maintaining appropriate price-to-earnings discipline even as valuations have expanded[2][13]. By contrast, private AI companies commanding billion-dollar and multibillion-dollar valuations are often generating minimal revenue, have no clear paths to profitability, and represent entirely speculative bets on future market development[2][13][25].

The mechanisms creating this divergence are structural and merit detailed examination. Public market stocks are subject to continuous repricing based on the flow of new information—earnings reports, product announcements, competitive developments, macroeconomic changes—and short-selling constraints force bearish investors to stake capital against inflated valuations, creating pressure for price correction[13][25]. Private market investments, by contrast, involve discrete funding rounds, often with information asymmetries between informed insiders and later-stage investors. The last round priced reflects what the most recent investors were willing to pay, which then becomes the baseline for subsequent pricing rounds, creating a ratchet effect where valuations only move upward, never downward, until a catastrophic liquidity event forces a reset[2][13]. This is precisely why private market FOMO becomes so dangerous: once a valuation has been established in a Series D or growth equity round at an inflated level, subsequent investors anchoring to that round face the prospect that walking away from future rounds at higher prices will mean admitting the original investment thesis was wrong—a politically and professionally difficult acknowledgment[2][25].

The capital availability metrics provide additional evidence of private market disconnect from rational valuation. Kostin notes that while the US has seen approximately 55 IPOs larger than $25 million in 2025, this represents a fraction of the 280 IPOs in 2021 and nearly 400 in 1999, suggesting that while capital remains available in public markets, it is not characterized by the “ebullience” that marked previous bubbles[2][13]. Yet private markets have seen record levels of capital deployment, particularly into AI, with venture capital funding hitting $97 billion in Q3 2025 alone, a 38 percent year-over-year increase, with nearly half flowing to AI companies[27]. This divergence—public markets showing discipline, private markets showing exuberance—is precisely the diagnostic signal that FOMO and valuation disconnects are operating in the private sector[2][13][27].

Historical Lessons: How the Dot-Com Parallel Differs and Echoes

Understanding current private market AI dynamics requires careful historical analysis of the dot-com bubble of 1998-2000, particularly examining what aspects of that cycle are genuinely repeating and where important differences might provide either reassurance or mask new vulnerabilities[46][56][59]. The parallels are undeniable: in both eras, transformative technology captures imagination and capital flows, valuations become unmoored from fundamentals, companies with minimal revenue command billion-dollar valuations, and FOMO drives investors to participate despite acknowledged risks[46][56]. The NASDAQ index increased fivefold in the late 1990s before collapsing nearly 80 percent between March 2000 and October 2002, with hundreds of companies losing 80 to 95 percent of their value[46]. Current S&P 500 valuations, on Robert Shiller’s cyclically adjusted price-to-earnings ratio, are trading above 1929 and 2021 peaks and about 13 percent lower than 2000 peaks on that metric, suggesting comparable levels of valuation exuberance[5].

Yet important differences between the two cycles warrant examination. During the dot-com boom, venture capital and retail investors drove speculation, with the capital base supporting inflated valuations resting on relatively thin financial foundations. When the capital dried up—as it inevitably did when IPO markets closed and fundraising velocity slowed—the system collapsed quickly because there were no deep-pocketed institutional actors willing to maintain valuations or continue deploying capital[46][56]. Today’s AI boom, by contrast, is substantially funded by major technology companies with robust balance sheets and strategic imperatives to maintain AI investment. Microsoft, Google, Amazon, Apple, Meta, and NVIDIA collectively generate more than $200 billion in free cash flow annually, providing them with the capacity to sustain capital deployment even if near-term returns disappoint[40]. This difference is not trivial; it means that unlike the dot-com era, where funding could evaporate overnight, today’s AI ecosystem has structural support from actors with long time horizons and strategic reasons to maintain investment regardless of near-term financial returns[7][40].

Additionally, AI technology has already demonstrated genuine commercial utility in ways that challenged the internet during the 1990s. By December 2025, major enterprises are deploying AI systems at scale, revenue models are becoming clearer even if they are not yet generating profits at those high valuations, and there is no serious question about whether AI will be transformative. The question is not whether AI matters—it clearly does—but rather who will capture value and at what valuations[42][44]. By contrast, during the peak of the dot-com bubble, fundamental questions about internet viability remained unresolved in many investors’ minds, even as valuations soared[46]. From this perspective, today’s AI cycle might be described as having stronger technological fundamentals than the dot-com cycle while simultaneously showing comparable or greater valuation excesses[5][21][46][56].

However, this observation provides limited comfort when examined through the historical record of what occurred to incumbents and infrastructure investors during the dot-com crash. Goldman Sachs’ 25-year retrospective on the technology bubble notes that while the internet itself proved transformative, the infrastructure companies that built out the underlying systems—telecoms, equipment vendors, data center operators—often failed to capture commensurate value[46]. These companies built out capacity assuming sustained exponential growth in demand for bandwidth and internet access. When that demand growth failed to materialize at projected levels, the infrastructure became excess capacity, and pricing power evaporated[46]. The lesson for today’s AI infrastructure investors is stark: the capital being deployed to build data centers, GPU sourcing agreements, and cloud capacity may exceed the near-term utility for those resources, creating stranded assets and compressed returns precisely as occurred in the telecom bust

Get M&A headlines on X!