Introduction
Artificial intelligence is the most significant thematic driver of M&A valuations since the cloud computing wave of the 2010s. About one-third of the 100 largest corporate M&A transactions announced in 2025 cited AI as part of the strategic rationale, according to PwC's review of the year's megadeals, and in the technology sector nearly all of the largest deals referenced it. Infrastructure and frontier-model assets now carry the richest marks in the market, far above the mid-single-digit revenue multiples typical of standard SaaS companies.
But AI's impact on valuation is more nuanced than a uniform premium, and since early 2026 it has cut both ways: the agents behind near-trillion-dollar marks for model developers have also made investors question the revenue of established software companies. The market is developing a tiered framework that differentiates AI companies by the depth and defensibility of their technology and applies very different multiples to each tier, down to an outright AI discount at the bottom.
The AI Valuation Tier Framework
Tier 1: AI Infrastructure (Highest Multiples)
Companies building the foundational layers of AI: semiconductor designers (NVIDIA, AMD), cloud computing infrastructure (hyperscalers), data center operators (Equinix, Digital Realty), and large language model developers (OpenAI, Anthropic). These assets command the most extreme valuations because they are the "picks and shovels" of the AI revolution: every AI application depends on them, and supply is constrained (the AI investment cycle article traces the capex behind that demand).
The reference marks for the tier are the private rounds of the two leading frontier labs: OpenAI at $852 billion after its $122 billion round in March 2026, and Anthropic at $965 billion after its $65 billion Series H in May 2026, the first time it topped OpenAI, as CNBC reported. Against OpenAI's disclosed revenue of about $2 billion a month and Anthropic's $47 billion run rate, those marks equal roughly 35x and 21x annualized revenue, multiples that still flatter both because run-rate revenue annualizes the latest month (see how investors value OpenAI and Anthropic). NVIDIA, whose record highs in early October 2026 lifted its market value close to $6 trillion, trades at about 14x the annualized value of its third-quarter fiscal 2027 revenue guidance of $108 billion. The gap is the lesson: NVIDIA's multiple prices profitable revenue at gross margins near 75%, while the lab marks price an option on dominance for companies that still burn cash.
Tier 2: AI-Native Applications (High Multiples)
Companies whose core product is built around AI: AI-powered cybersecurity platforms (like Wiz, acquired by Alphabet for $32 billion in a deal that closed in March 2026), AI-driven drug discovery (Recursion Pharmaceuticals, Isomorphic Labs), and AI-native enterprise software. These companies have AI embedded in their core architecture, not bolted on as a feature. They command 30-50% premiums over comparable non-AI software.
Tier 3: AI-Enhanced Products (Moderate Premium)
Established software and technology companies that have integrated AI capabilities into existing products (CRM with AI-powered recommendations, ERP with AI-driven forecasting, marketing platforms with AI content generation). The AI premium for Tier 3 companies was always the smallest (10-20% above non-AI peers) and came under pressure as AI features became table stakes rather than differentiators.
The 2026 software sell-off showed that this premium can flip into a discount. As AI labs released agents for everyday office work, investors feared those agents would replace the human users that per-seat pricing bills for, the model used by Adobe, Salesforce and many others, as CNBC reported in January 2026. By early February the S&P 500 Software and Services Index was down about 20% for the year, and the iShares Expanded Tech-Software Sector ETF ended the first quarter down 24%, its worst quarter since 2008, even as AI companies took 86% of private deal value in the first half, per PitchBook data cited by CNBC.
The discount did not stay uniform, which is the real test of the framework. By early August the ETF had recovered to a 3% loss for the year as results sorted the group: Atlassian and Twilio each jumped more than 20% after reporting, while HubSpot had its worst day in a decade. A Tier 3 valuation now needs an explicit view on whether AI adds to or erodes revenue, with seat compression in the downside case; the SaaS valuation reset covers the sell-off's effect on multiples and on 2021-vintage PE take-privates.
- AI Valuation Premium
The incremental valuation multiple a company commands because of its AI capabilities, measured against comparable non-AI businesses. It reflects the market's view that AI creates durable advantages through proprietary data, unique model architectures, network effects (more users generate more data, which improves the model) and high switching costs. The premium varies by tier: infrastructure-grade AI assets can command 3-5x the multiple of comparable non-AI businesses, while feature-level AI may earn only a 10-20% premium that erodes as competitors catch up, or turns negative when AI threatens the core revenue model.
Why Traditional Valuation Methods Struggle with AI
Revenue Quality Is Harder to Assess
Not all AI revenue is created equal. Consumption-based pricing (common for AI API services, where the customer pays per query or per token) creates variable revenue streams that are harder to project than traditional subscription models. A company reporting $100 million in AI-related revenue may have very different visibility depending on whether that revenue comes from annual contracts (predictable) or usage-based pricing (volatile).
The Moat Question Is Unresolved
The most consequential valuation debate in AI is whether foundation models will become commoditized (in which case the premium dissipates) or whether network effects and data advantages create durable moats (in which case the premium is justified). No lead has proved safe so far: OpenAI defined the category, yet Anthropic's May 2026 round priced it above OpenAI, and open-source alternatives (Meta's Llama, Mistral) keep narrowing the quality gap. If the model layer commoditizes, value accrues to the application layer (companies using AI to solve specific vertical problems) rather than the infrastructure layer.
Landmark AI Deals and Their Valuation Implications
| Deal | Value | Implied Multiple | AI Rationale |
|---|---|---|---|
| Alphabet / Wiz (closed March 2026) | $32B | ~30x+ estimated ARR | Cloud and AI security platform at scale |
| NVIDIA / Groq (December 2025) | ~$20B | ~3x Groq's last private round | Inference chip IP and leadership team, via license |
| Electronic Arts LBO (closed August 2026) | $55B | Record all-cash LBO | AI in game development and player experience |
| NVIDIA (trading) | ~$6T market cap | ~14x annualized revenue guidance | Foundational AI compute infrastructure |
These deals show the range of AI-driven valuations. At one end is pure infrastructure: data center sales were about 92% of NVIDIA's second-quarter fiscal 2027 revenue. Next come acqui-license structures: NVIDIA paid about $20 billion for a non-exclusive license to Groq's inference technology plus its senior team, roughly three times Groq's $6.9 billion September 2025 valuation, according to CNBC, without buying the company. Alphabet/Wiz is a strategic deal in which AI security was one dimension of a cloud platform strategy. EA sits at the edge of the category: its closing announcement quotes the buyers citing AI in game development, but the price rests on established franchise cash flows, a reminder that an AI rationale is not the same thing as an AI valuation.
Valuing AI Companies: A Practical Framework
For investment bankers advising on AI-related transactions, the valuation approach combines standard tools with AI-specific adjustments:
Use EV/Revenue as the primary multiple for high-growth AI companies, because most are pre-profit or have EBITDA margins compressed by growth investment. Segment the peer group into the appropriate tier (infrastructure vs. application vs. feature-level) to avoid comparing NVIDIA to a small AI SaaS startup.
Supplement with a DCF that models the path to profitability. AI companies investing heavily in model training, compute infrastructure, and talent will show near-term losses that give way to operating leverage as revenue scales. The projection period should be 7-10 years to capture this transition, and the terminal value should reflect mature, at-scale margins (which for software platforms can reach 30-40% EBITDA).
Assess the moat. The valuation should explicitly address whether the company's AI advantage is defensible. Proprietary training data that improves with use (a data flywheel), patented model architectures, and exclusive partnerships create lasting value. Generic AI features that any competitor can replicate with off-the-shelf models do not.
- Data Flywheel (AI Moat)
A competitive advantage in which a company's AI models improve as more users generate more data, which trains better models, which attract more users. Strong flywheels (Google Search, Amazon product recommendations) widen the moat with use. In AI valuation, a proven flywheel (measurable model improvement tied to data volume) justifies a structurally higher multiple, because the advantage persists even as the underlying AI technology commoditizes.
Price the downside as well as the upside. Since the 2026 sell-off, a credible AI valuation also models how agents could erode the target's own revenue (fewer seats, lower pricing, shorter contracts) instead of treating AI only as a source of growth.
