The press-release math on AI deals in 2025 and 2026 is doing real damage to founder expectations. Every quarter brings another headline — OpenAI buying Windsurf, Meta writing a $14.3 billion check for 49% of Scale AI, ServiceNow paying $3 billion for Moveworks, Alphabet at $32 billion for Wiz — and every quarter, founders walk into conversations expecting their business to be valued against those prints.
Paul Inouye believes it doesn't work that way. There are three fundamentally different types of deal happening inside the "AI M&A" label, and the multiples don't transfer between them. If you don't know which one you're building for, you're going to pitch the wrong story to the wrong buyer, and the gap between what you expect and what clears will be painful.
Type one is the capability grab. Pure talent and model plays. OpenAI acquiring Windsurf to own a coding agent. Meta paying $14.3 billion for 49% of Scale AI — effectively an investment with pre-negotiated hiring rights. These deals are priced on the cost to replicate the team and the data internally. Revenue barely matters. The multiples you read about are meaningless to anyone whose business isn't one of three or four labs or hyperscalers. If you're a $20 million ARR founder reading about Windsurf, that is not your comp.
The capability-grab buyer universe is small — maybe a dozen companies globally that can justify the price. Frontier labs, hyperscalers, and one or two strategic acquirers on a specific thesis. Everyone else doesn't have the strategic need, capital base, or integration capacity. If your business is in this bucket, you don't run a conventional process. You have a small number of direct conversations, often through relationships, and the price gets set in negotiation rather than an auction.
Type two is the hybrid — real product, AI-capability driven. ServiceNow paid roughly $3 billion for Moveworks because Moveworks had a genuine product, Fortune 500 customers, and an AI capability ServiceNow couldn't quickly build. MongoDB buying Voyage AI fits the same pattern — real asset, but the premium is priced on the AI IP, not the recurring revenue. Thomson Reuters buying Casetext for $650 million set the template a couple of years ago. Multiples here run hot because buyers are paying up for capabilities that would cost years and hundreds of millions to replicate internally. These deals are replicable for founders with genuine technical moat and commercial validation — but the bar on the moat is high, and it has to be real, not marketed.
The type-two buyer universe is broader — enterprise software incumbents, data infrastructure players, security platforms, fintech infrastructure — anyone needing to accelerate an AI capability and would rather acquire than build. The premium typically runs two to three times what a traditional strategic would pay for the same revenue, because the buyer isn't pricing the ARR; they're pricing the capability and the time saved. A Rule-of-40 company with NRR above 120% and a proprietary AI integration is trading at 7x to 9x revenue in this bucket, per one recent SaaS M&A framework. The top 5%, where competitive process tension holds, prints at 10x to 12x.
Type three is traditional M&A at scale. Alphabet's $32 billion for Wiz. Palo Alto Networks' $24.5 billion for CyberArk. Real businesses, real revenue, real integration thesis. These deals clear at multiples consistent with where the sector trades publicly, adjusted for strategic premium. This is the category where most founder-led businesses in the $25 million to $250 million range actually land. Less glamorous, maybe. But it is real, and it is where process mechanics matter most — buyer coverage, deal pacing, competitive tension.
The type-three buyer universe is widest and most process-dependent. Strategic acquirers evaluate on fit — customer base, distribution, product complementarity, geographic coverage. Financial sponsors evaluate on Rule of 40, NRR, and platform potential. The premium over public comps comes from a competitive process that forces the top two or three buyers to compete, and from the advisor's job: coverage, positioning, pacing. A well-run type-three process in 2026 still generates 20% to 35% premiums over a quiet negotiated sale. Those points matter when the underlying business is trading at 3x to 5x revenue.
The trap founders fall into is mixing the categories. They pitch type-two AI capability stories when they are really type-three businesses, which confuses buyers and kills credibility on the first call. Or they benchmark to type-one acqui-hire multiples when they are building a type-three company, which blows up on the first round of comp analysis. The tell, when it happens, is a deck that leads with AI on slide two and doesn't get to revenue quality until slide fifteen. Sophisticated buyers will read three slides before asking how much of the story is real.
Here is a simple self-diagnostic. Under $15 million in ARR with most of the value in proprietary model IP and elite technical talent — type one; plan accordingly. Real revenue, meaningful customer traction, and a defensible technical moat a strategic buyer would pay a premium to absorb quickly — type two; prep emphasizes the moat, and your buyer list is incumbents who need what you have. Durable recurring revenue in the $25 million to $250 million range with an AI roadmap incremental to a fundamentally strong software business — type three; run a disciplined process, focus on strategic fit and competitive tension, and don't try to price yourself as type two without the goods to back it up.
Before you plan the exit, decide which type you are building. The prep is different. The team shape is different. The buyer universe is different. The multiple math is different. Founders who know which trade they are running walk into their process with a story the buyer recognizes. Everyone else shows up with the wrong deck for the wrong room.