2026 SaaS M&A Retrospective: What Software Can Command a Premium in the AI Era

At the beginning of this year, the SaaS (software-as-a-service) industry went through a round of sharp emotional turmoil. As AI Agents began to enter more specific work scenarios such as legal, finance, and sales, the market began to ask: if AI can complete a large amount of the work that software originally handled, and do it faster, how much are those traditional software products still worth? Thus, a rather dramatic new term, "SaaSpocalypse," began to circulate.

Nine months later, the market gave another answer. Among SaaS companies also looking for buyers, some were bid up and snapped up, while others could only accept any offer. What exactly happened in between? Let us review some of the landmark moments of this year.

A few deals prop up the whole year

When Anthropic added plugins for roles such as legal, finance, and sales in the first quarter, investors began to ask a question: which software can AI agents directly replace? What the market worried about was not any single company losing orders, but the subscription model itself. The charging model of traditional SaaS is deeply tied to headcount. If AI agents finish the work for enterprises, why would enterprises still buy an account for every employee? Once this logic holds, what software companies have to answer becomes whether their revenue model still holds up.

By June, the impact of the stock selloff in related industries earlier in the year gradually faded. The largest new deals of the year were announced one after another between June and September. SpaceX acquired Cursor for $60 billion in an all-stock deal, Nvidia agreed to acquire Hugging Face for a reported $12.9 billion, Stripe agreed to acquire OpenRouter for a reported $7.5 billion, and AMD agreed at the end of September to acquire World Labs for $8.2 billion. Together with two large security deals signed earlier in the year—Google's $32 billion acquisition of Wiz and Palo Alto Networks' $25 billion acquisition of CyberArk—these five deals totaled more than $137 billion.

It is worth noting that these deals showed an extreme concentration at the top. The Cursor deal alone, at $60 billion, far exceeded the other deals. Of these five deals, two bought "security," and three bought "entry points to models and developers." In other words, what buyers paid for was precisely developer distribution capability and security capability, and these two are exactly what most AI deployments lack.

Two markets, two sets of valuation logic

These deals send a signal that is not relaxed but very clear: the market has not given up on SaaS; it has simply begun to judge more selectively what kind of software is worth buying and what kind of software can only be repriced.

At the top, buyers are raising their bids. OpenRouter is a model aggregation startup platform founded in 2023, valued at $1.3 billion in its May 2026 financing; in August, Stripe agreed to acquire it for a reported $7.5 billion, only three months later. A few days after its own IPO, SpaceX paid for the Cursor acquisition in full with stock. Premiums are the norm in these areas.

In other areas, sellers are accepting whatever price the market gives. Airtable's peak valuation at the end of 2021 exceeded $11 billion, but it ultimately agreed to sell to Bending Spoons for about $1.3 billion.

The difference between the two ends lies in the nature of the assets. What OpenRouter holds is a distribution channel for model calls; by buying it, the buyer directly plugs into that channel. What Airtable holds is a set of collaborative spreadsheets, a type of function that is easiest to rewrite, and after rewriting, customers may not follow. The former sells a strategic position; the latter is more about selling operating results.

The total volume looks prosperous, but the buyer structure is tightening. There are more deals, but fewer people making them.

So far in 2026, the number of AI companies acquiring each other in the United States is 14% higher than in all of 2025. OpenAI leads with 10 deals, but the number of independent buyers grew by only 2%. A small number of winners take the vast majority of acquisitions, and the current market has not yet cultivated a broader buyer base. For sellers, the number of available counterparties is smaller than imagined. The buyer list is highly repetitive, and each one clearly knows its scarce position.

In January this year, the question was whether AI would make software worthless. Nine months later, companies have given their answer. After code became cheaper, the center of value for software companies moved toward two ends: one end is data and workflows that others cannot obtain, and the other end is distribution channels that others cannot enter. Companies stuck in the middle with only a feature list were repriced by the market this year.

In the past, "already has customers, already has revenue, already has a mature product" was in itself enough to constitute considerable value. Now, this is still important. But if a product lacks unique data, hard-to-replace workflow positions, strong distribution capability, or security and compliance barriers, buyers are more likely to see it as a financial asset rather than a strategic asset.

The transaction prices this year actually offer three very simple tests.

The first is the open-source test. Suppose this company posted all its code online tomorrow; would customers still renew? If the answer is yes, it means what customers truly cannot do without is the data and processes inside, and the code is only the shell; if the answer is no, then being replaced by AI is only a matter of time.

The second is "whether this money buys growth or an existing stock." Also software, someone is willing to pay 20 times annual revenue because that part of revenue will be larger next year; Airtable sold for only $1.3 billion because the buyer is calculating how many customers will leave and how long the remaining ones can still be charged. One buys the future, the other buys residual value; under the same word are actually two kinds of business.

The third is "the reason customers have no choice but to buy." Every enterprise has its own reason for purchasing software services, but this year their questions are also changing: if we do not acquire, how long would it take us to build it ourselves? If we build it, can we simultaneously gain its customer trust, data accumulation, developer ecosystem, and workflow position? Can we pass security reviews, compliance certifications, and the long procurement process inside the organization? For SaaS companies, the fewer these questions are, the more passive they are in bargaining.

"Software that can be sold" and "software that customers have no choice but to buy"—whether a company will be regarded as "an asset worth a strategic acquisition" does not have to wait until the day it is sold to be revealed. The answer often begins to take shape today from product choices, customer relationships, and capability accumulation. AI is making the supply of software services richer, and the M&A market is redefining scarcity in a stricter way.

This article comes from the WeChat public account "Hongshan Hui" (ID: Sequoiacap), author: Hong Shan, published by 36Kr with authorization.