Key takeaways

  • As private AI giants look to launch IPOs in 2026, PitchBook research suggests the market has rewarded story over substance, as companies carrying the highest valuations are built on the weakest business fundamentals.
  • OpenAI commands the biggest brand and the highest valuation, but its most important business metric—how much revenue it retains from enterprise customers—has never been disclosed.
  • Databricks has faster growth than its closest public comparable, positive free cash flow, and a valuation that is cheaper by every adjusted measure.
  • Anthropic’s path to a premium valuation runs through a single near-term hurdle: whether its gross margins reach 40% in 2026.

Three of the biggest names in the artificial intelligence revolution appear on track to launch IPOs in 2026 amid a wave of hype, massive valuations, and for most investors, very little visibility into the businesses behind them.

OpenAI, Anthropic, and Databricks are expected to go public this year, and they carry combined last-round private valuations approaching $1.4 trillion. But that math is based on funding announcements without context, leaving investors to compare incomparable metrics across companies with radically different revenue models. Without a consistent framework, the category is nearly impossible to price.

Once companies file their S-1 forms with the Securities and Exchange Commission, more details will emerge. Until then, PitchBook’s framework evaluates these AI giants on business quality fundamentals, anchored by net revenue retention, which measures what a company earns from existing customers. This framework reveals one clear takeaway: The private market is systematically mispricing these three companies.

Navigating the hype: A framework for a market without one

Rather than ranking on model benchmarks or annual recurring revenue—a company’s annualised subscription and contract revenue—PitchBook’s AI Business Quality framework scores each company across five business quality dimensions weighted for long-run enterprise value relevance:

  • Revenue quality (25% of the score), measuring whether the revenue base compounds through enterprise concentration and net revenue retention
  • Capital efficiency (20%), capturing annual recurring revenue per dollar of cumulative capital raised
  • Governance optionality (20%), evaluating whether corporate structure accelerates or constrains the path to public markets
  • Moat durability (20%), assessing defensibility against model commoditisation
  • Computing independence (15%), measuring control over inference infrastructure—the dimension most directly tied to margin expansion as API pricing compresses

Its central finding cuts through the noise. The private market has produced a systematic quality inversion: The companies commanding the highest valuations score lowest on the five business quality measures public markets actually price. OpenAI, at a valuation of $840 billion, scores weakest of the three. Databricks, at $134 billion, scores strongest.

Capital efficiency makes this concrete. Anthropic generates $0.23 in ARR per dollar raised, Databricks $0.16 on positive free cash flow, and OpenAI $0.11. That reading on OpenAI is down from $0.31 18 months ago—the steepest single-event deterioration of any major AI company under PitchBook’s coverage.

Why NRR will define these IPOs

When these companies file S-1s, most investors will anchor their valuations on one metric above all others: net revenue retention, which measures recurring revenue retained from existing customers, including expansion from additional usage minus churn. Above 120% is best in class for enterprise software; above 140% is exceptional by any standard.

NRR matters particularly in AI because the dominant revenue model is consumption-based—customers pay per token consumed rather than per seat licensed. Token consumption compounds automatically as enterprise customers build on a model. A single successful product launch can multiply a customer’s spend rapidly without any sales motion. Databricks and Anthropic both sit at approximately 140% NRR, while OpenAI’s enterprise NRR has never been disclosed.

Databricks: Cheaper than it looks

Databricks’ framework predictions and public data align most cleanly. The firm crossed $5.4 billion in ARR in the fourth quarter of 2025 with 65% year-over-year growth (nearly twice Snowflake’s SNOW 29% rate at comparable revenue scale) while delivering positive free cash flow. With 140% net dollar retention, more than 700 customers above $1 million in annual contract value, and AI products now representing $1.4 billion in ARR, the company arrives at the IPO gate without a profitability narrative to construct because it already has one.

Adjusting EV/ARR for growth rate—dividing the revenue multiple by the ARR growth rate, so that faster growth justifies a higher multiple—places Databricks at 0.38 times versus Snowflake at 0.69 times. Investors are paying less per unit of growth for the faster-growing business, an anomaly that typically compresses at IPO. At 25 times ARR on 65% growth with positive FCF, Databricks is the cleanest institutional entry point in this cohort by every relevant measure.

Anthropic: The margin curve is the investment thesis

Anthropic’s enterprise metrics justify the framework’s heaviest weighting on revenue quality. PitchBook estimates NRR at approximately 140%, with eight of the Fortune 10 as customers and 80% of revenue from enterprise contracts. Claude Code generates an estimated $2.5 billion in ARR with 54% market share in AI-assisted coding. However, the investment thesis rests not on current profitability but on the speed at which the margin structure is improving.

The margin tells the investment thesis directly, with the firm going from negative 94% gross margin in 2025 to an estimated 40% in 2026 and 77% by 2028—a trajectory that mirrors CrowdStrike CRWD and Cloudflare’s NET first four years as public companies and justifies a premium to current SaaS semiannual at the IPO gate.

The 2026 estimated milestone of 40% is the critical near-term confirmation signal where a miss reopens the valuation question entirely. Two risks warrant monitoring: Approximately $200 million in federal revenue is exposed to a Pentagon use restriction that could signal broader regulatory headwinds, and computing cost assumptions remain sensitive to model architecture decisions that could compress the 2027-28 margin targets.

OpenAI: The biggest brand, the biggest question mark

OpenAI is both the company most investors think they understand and the one whose framework’s warning is most direct. Its $20 billion in ARR on 800 million-900 million weekly active users is a consumer footprint without precedent. The structural problem is that roughly 85% of those users pay nothing. Consumer subscription economics do not compound the way enterprise contracts do, and the enterprise NRR that would justify or collapse the 42 times multiple has never been disclosed. Its disclosure will define the IPO.

The capital structure compounds the challenge. Approximately $390 billion in computing obligations to Microsoft’s MSFT Azure and Amazon’s AMZN AWS produced $20 billion in ARR at the cost of computing dependence, precisely when computing ownership is the primary moat differentiator and API pricing has already compressed more than 90% since 2023. Profitability is not expected before 2029. The nonprofit-to-for-profit governance conversion remains unresolved.

At 42 times ARR with an undisclosed NRR, investors are being asked to price a business whose most important metric has been withheld. That is not a reason to dismiss the position—it is a reason to wait for the S-1 before sizing it.

Alphabet: The AI Allocation You Can Make Today

The most defensible AI position available does not require waiting for any filing. Alphabet GOOG/GOOGL is the only entity in frontier AI—public or private—that controls the full vertical stack the AIBQ framework measures. The company has a benchmark-leading model in Gemini 3.1 Pro, seventh-generation TPU inference hardware, hyperscale cloud infrastructure, and Search and Workspace distribution at a scale no private company can replicate.

Google Cloud revenue grew 48% year over year to $17.7 billion in the fourth quarter of 2025. Google holds approximately 10% equity in Anthropic—simultaneously an infrastructure vendor, equity stakeholder, and model competitor—meaning Anthropic’s growth accrues to Google Cloud regardless of whether Anthropic itself reaches profitability.

Microsoft’s relationship with OpenAI carries asymmetric downside by comparison. OpenAI accounts for an estimated 10 points of Azure’s AI-attributed growth—meaning if OpenAI’s enterprise penetration stalls ahead of its IPO, Microsoft’s cloud growth narrative stalls with it. That is single-vector exposure. Alphabet, by contrast, earns across the model layer, the infrastructure layer, and the equity layer simultaneously—three compounding positions where Microsoft has one.

Private markets have produced genuine business quality in Databricks and Anthropic, real valuation risk in OpenAI, and a structural public market advantage in Alphabet that remains underappreciated.

Subscribe to get Morningstar insights in your inbox