The next phase of the AI boom may be less about technological adoption than about who ultimately owns the financial risk. After several years in which venture capital deployed enormous sums into private technology companies while exits lagged, some of the largest AI businesses ever created are approaching public markets.

The central question is not whether artificial intelligence is real or commercially valuable. It clearly is. The question is whether IPOs are becoming the mechanism through which extraordinary private valuations, persistent operating losses and increasingly leveraged infrastructure commitments are transferred from venture investors to public-market balance sheets.

The incentive for liquidity is substantial. PitchBook and the NVCA estimate that cash flows to venture LPs have been negative by roughly $197 billion since 2022. Despite improving exits, 2025 remained far below the 2021 liquidity peak, while hundreds of unicorns continued to sit in private portfolios. The NVCA counted 859 U.S. unicorns worth roughly $4.34 trillion entering 2026, while only a small fraction exited in the prior year. The result is a structural DPI problem: funds can continue marking portfolio companies higher, but LPs ultimately need cash distributions rather than paper appreciation.

AI has intensified that imbalance. In Q1 2026 alone, $242 billion—80% of all global venture funding—went into AI companies. OpenAI, Anthropic, xAI and Waymo absorbed $188 billion between them. Record capital formation has therefore become unusually concentrated in a handful of businesses whose eventual exits could determine returns for large portions of the venture ecosystem.

The valuation divergence is important. Much of the broader venture cohort has undergone a painful reset from 2021 marks, while leading AI companies have moved in the opposite direction. That creates an unusual exit environment: public investors may be asked to absorb businesses at valuations dramatically above previous private rounds even while profitability remains uncertain. It does not prove that VCs are deliberately “dumping” companies on the public market. But it creates a clear incentive to monetize positions when market enthusiasm makes exceptionally large exits possible.

Long-run IPO data reinforce the concern. University of Florida professor Jay Ritter finds that only 24% of U.S. technology IPOs in 2025 were profitable on trailing earnings, compared with 44% in 2004 and 36% in 2005. Meanwhile, the median tech IPO offer-price-to-sales multiple reached 11.8x in 2025, versus 6.4x in 2004 and 4.5x in 2005. Public investors are therefore accepting substantially higher sales valuations while profitability among issuers remains considerably weaker.

The AI pipeline pushes this tension to an extreme. OpenAI's audited 2025 accounts showed $13.07 billion of revenue against $34 billion of costs and expenses, producing a $20.92 billion operating loss. Its $38.5 billion headline net loss included large accounting charges, making the operating result a cleaner indicator of the underlying economics. At the same time, OpenAI completed a private financing at an $852 billion valuation in early 2026.

Anthropic represents a somewhat stronger operating trajectory, but it is still being valued primarily on expectations of extraordinary future growth rather than established earnings. Reuters reported that bankers have been considering valuations potentially approaching $2 trillion, partly justified by projections of $190–200 billion of revenue in 2028. The company is simultaneously arranging a $15 billion revolving credit facility as it prepares for its IPO.

Databricks provides the obvious counterexample. It was valued at $190 billion in August 2026, reached a $7 billion annualized revenue run rate and remained cash-flow positive. Cerebras also reached public markets after materially improving its financial profile. The thesis, therefore, should not be “all AI IPOs are bad businesses.” The more defensible argument is that the largest and most strategically important frontier-AI issuers may reach public markets before their economics have demonstrated that they can support their valuations without continuous access to external capital.

The valuation-versus-revenue comparison makes that distinction visible. The market is not simply paying for today's revenue.

It is capitalizing expectations about future market dominance, falling compute costs and enormous future AI demand. That may prove correct, but it also means relatively small changes in long-run growth or margin assumptions can produce very large changes in implied equity value.

High revenue multiples are particularly consequential when earnings are negative. A profitable company can eventually defend its valuation through cash generation. A loss-making company valued on revenue requires continued confidence in future margins. If that confidence breaks, there is far less fundamental valuation support.

CoreWeave is arguably the most useful preview of this model. The company is experiencing exceptional demand: Q2 2026 revenue reached $2.58 billion. Yet it still generated a $626 million net loss, while quarterly net interest expense reached $640 million. Its recourse and non-recourse debt totaled roughly $35 billion at June 30, and total liabilities had reached $72 billion.

This is where the AI story expands beyond equity valuation. Infrastructure providers must finance GPUs, power and data centers before much of the contracted revenue arrives. The Bank of England has warned that AI's use of external financing is accelerating quickly: AI-related issuers accounted for 41% of U.S. high-yield issuance for non-refinancing purposes in 2026, despite accounting for only around 1% of the high-yield index at the end of 2025. Private credit's share of AI investment financing increased from 9% in 2024 to 34% in 2025.

That creates a second risk channel beyond IPO valuations: leverage turns an equity repricing into a credit event.

An additional vulnerability is that many of these businesses are not financially independent from one another. Nvidia invests in AI companies that purchase Nvidia chips. Cloud and infrastructure providers receive commitments from frontier labs while simultaneously investing in, partnering with or financing those same labs. OpenAI receives capital from major technology companies and then spends enormous amounts purchasing infrastructure and services from parts of the same ecosystem.

Bloomberg's mapping of these relationships illustrates an increasingly circular system of investments, cloud contracts, chip purchases and venture financing. Circularity does not make the transactions fictitious: GPUs are delivered, computing services are consumed and revenue is recognized. The concern is economic dependency. If Company A's revenue depends on Company B's spending, while Company B's ability to spend depends on funding from Company C, which itself benefits from Company A's demand, apparently diversified revenue streams can ultimately depend on the same underlying assumption: continued exponential growth in AI capital expenditure.

A slowdown could therefore propagate through the network. Frontier labs reduce compute commitments; infrastructure providers lose expected utilization; highly leveraged data-center operators face weaker debt-service coverage; chip orders slow; valuations decline; and financing conditions tighten precisely when loss-making firms need additional capital.

The final risk is the destination of the equity. An IPO converts concentrated private ownership into securities that can eventually enter institutional portfolios and indexes. The transmission into retirement savings is not instantaneous: S&P currently requires at least 12 months of trading history, positive GAAP earnings and sufficient free float before a new company can enter the S&P 500. That provides a meaningful profitability filter.

Other benchmarks can absorb major IPOs more quickly, however, and public-market exposure also arrives through actively managed funds, hedge funds and broad technology products well before S&P 500 inclusion. At the same time, the AI financing boom is spreading through investment-grade bonds, high yield, private credit and structured financing. The systemic issue is therefore broader than whether one unprofitable company eventually enters a 401(k) index fund.

The emerging question is whether private markets are creating companies whose capital requirements and valuations have become too large to remain private, forcing public markets to become the next source of liquidity before the economics are fully proven.

That does not mean an AI crash is inevitable. Revenue growth is real, demand is real and some companies already demonstrate sustainable economics. But the combination of VC liquidity pressure, unusually high valuations, weak profitability, expanding leverage and circular financial relationships creates a structure worth examining closely. If AI growth continues at extraordinary rates, the system may successfully grow into today's prices. If growth or monetization disappoints, the IPO wave could mark the point at which losses previously concentrated among sophisticated private investors became substantially more distributed across the financial system.

Publishing note: the Value Add VC charts should be treated as valuation estimates/snapshots rather than authoritative current valuations. Several figures have already moved materially—for example, Databricks was revalued to $190B in August and Anthropic's potential IPO valuation is now being discussed as high as $2T. Also, the first valuation-table image has an unrelated title (“AI Impact on Job Openings…”); I would retitle it before publishing.

BirJob. (2026). AI IPO Wave 2026: OpenAI $852B, Anthropic $380B, Databricks $134B. https://www.birjob.com/blog/openai-anthropic-databricks-largest-ai-ipo-wave-2026 

Flowingdata. (2025). Circular deals among AI companies. https://flowingdata.com/2025/10/13/circular-deals-among-ai-companies/ 

KloverAI. (2026). Complete 2026 AI IPO Landscape: Every Company, Every Bet, Every Risk. https://www.klover.ai/complete_2026_ai_ipo_landscape_every_company_every_bet_every_risk_indepth_analysis_2026/ 

Reuters. (2026). Why SpaceX faces a longer wait to join S&P 500. https://www.reuters.com/legal/transactional/why-spacex-faces-longer-wait-join-sp-500-2026-06-05/ 

Value Add VC. (2026). IPO Pipeline 2026: Every Company Expected to Go Public This Year. https://valueaddvc.com/blog/ipo-pipeline-2026-every-company-expected-to-go-public-this-year-and-current-status 

Value Add VC. (2026). $4T+ in AI Companies Heading to Public Markets. https://valueaddvc.com/ai-ipo-pipeline