AI Infrastructure Valuations: Q2 2026
AI infrastructure valuations in Q2 2026 are defined by a sharp split: capital-light data, inference and retrieval software command 15-25x revenue on consumption-led growth, while capital-intensive neoclouds re-rate to 6-10x on forward revenue. A $725B hyperscaler capex supercycle, reopened IPOs such as Cerebras and CoreWeave, and IBM's $11B acquisition of Confluent underscore record demand and strategic consolidation across the AI compute and data substrate.
- Sector
- AI
- Focus
- Valuations
- Published
- April 15, 2026
- Length
- 7 slides
- Reading time
- 11 minutes
Slide deck
7-slide deck. Desktop readers can page through the embedded viewer below. Mobile readers can open the direct PDF link.
Open slide deck PDF Key findings
- Hyperscaler capital spending is climbing to roughly $725B for 2026, up about 77% year on year, underwriting multi-year demand across the AI infrastructure stack.
- AI-native data platforms such as Databricks trade near 25x annualised revenue while neoclouds such as CoreWeave sit near 6.5x forward revenue, producing a valuation spread of roughly 10x compared to 4x in 2023.
- IBM completed its $11B acquisition of Confluent in March 2026, marking a landmark consolidation in the AI data-pipeline segment.
- Cerebras priced at $185 and surged on its May 2026 debut, signalling a meaningful reopening of the AI infrastructure IPO market.
- Snowflake reported 125% net revenue retention and Databricks reported net dollar retention above 140%, turning AI adoption into compounding revenue growth.
- Private equity invested approximately $45.7B in US data centers in 2025, a five-year high, with the $40B Aligned Data Centers sale setting a record for the asset class.
- Data and infrastructure software clearing the Rule of 40 threshold trades at roughly 2 to 3 times the revenue multiple of peers that fall below it.
- CoreWeave's roughly $100B contracted backlog anchors its equity story despite more than $20B of debt on the largest neocloud balance sheets compressing forward multiples.
- Inference platforms including Fireworks and Baseten reached roughly $800M and $600M in annualised revenue respectively, though GPU costs hold gross margins near 50%.
- Windsor Drake's working benchmark for public data-infrastructure software has settled near 11x EV/Revenue, with private AI-native leaders running at 20–25x.
Methodology
This report is based on Windsor Drake's synthesis and calibration of data drawn from Goldman Sachs AI capital-expenditure and infrastructure research, Morgan Stanley financing analysis, McKinsey & Company's Global Private Markets Report 2026, Bain & Company's Global Private Equity Report 2026, S&P Global Market Intelligence data-center M&A and private-capital analyses, PitchBook AI and data-infrastructure comparable-company data, CB Insights State of AI, PwC US Technology Deals 2026 Outlook, EY-Parthenon M&A Outlook 2026, Federal Reserve FOMC statements and Summary of Economic Projections, and Bloomberg AI infrastructure funding and valuation data. Windsor Drake applied its proprietary valuation framework and Rule of 40 benchmarking methodology to calibrate subsegment multiples and identify the primary drivers of premium and discount across the AI infrastructure stack.
Frequently asked questions
What multiples are AI infrastructure companies trading at in Q2 2026?
Multiples vary sharply by subsegment. AI-native data platforms trade at 18x–25x revenue, inference and model-serving platforms at 15x–19x, and cloud data platforms at 8.5x–12x. Capital-intensive GPU and accelerated cloud neoclouds compress to 6x–10x on forward revenue, reflecting backlog quality and debt loads rather than topline growth. Windsor Drake's broad public data-infrastructure benchmark has settled near 11x EV/Revenue.
Who is buying AI infrastructure companies right now?
Strategic incumbents are the most active acquirers, exemplified by IBM's $11B acquisition of Confluent completed in March 2026. Private equity is also pressing in, having invested roughly $45.7B in US data centers in 2025, a five-year high. Sovereign vehicles such as MGX from the Gulf are deploying record sums into compute and data centers, creating both buyers and cross-border partners.
How does the Rule of 40 affect AI infrastructure valuations in 2026?
The Rule of 40 remains the primary filter for a premium multiple. Software clearing the threshold trades at roughly 2 to 3 times the revenue multiple of peers below it. Top-quartile performers scoring above 50 average 18x EV/Revenue or higher, while those below 30 are discounted to 4x–8x. Balance between growth and margin matters as much as the headline score.
How long does an AI infrastructure M&A or IPO process take in 2026?
A full transaction process runs 12 to 18 months end to end. Windsor Drake advises founders who intend to meet the market while the $725B capex supercycle, reopened IPO listings, and roughly $3.7T of dry powder still align to begin preparation within the current cycle, given the lead time required.
What valuation metric should I use for a neocloud or GPU cloud business?
EV/Revenue is appropriate only as a starting point for neoclouds. An EV/contracted-backlog lens paired with the cost of capital values the equity more honestly because depreciation and debt sit between revenue and free cash flow. CoreWeave, for example, is anchored by roughly $100B of backlog rather than trailing revenue. EV/EBITDA is preferred for mature, cash-generative compute and physical data-center assets.
What is the difference between public and private AI infrastructure valuations in 2026?
The gap is widening rather than converging. Public data-infrastructure software has re-rated to about 11x EV/Revenue, while private AI-native leaders have run to 20x–25x, widening the spread from roughly 4x in 2023 to about 10x today. Scarcity of scaled independent AI-native assets, led by Databricks at a $134B valuation, keeps private marks running ahead of public comparables.
How does net revenue retention drive AI infrastructure valuations?
Net revenue retention is the single strongest driver of premium multiples in AI infrastructure. Snowflake at 125% and Databricks above 140% set the current benchmark, meaning existing customers expand spend automatically as AI workloads grow. Software gross margins of roughly 75% or better compound that retention into durable profit, whereas resold compute margins near 50% attract a significantly lower multiple.
Companies covered
Public and private companies referenced in this report.