In 2026, tech buyers pay for durability. The companies commanding strong prices pair efficient growth with revenue that expands inside existing customers, data and workflows an AI agent cannot easily replicate, a team that will still be there in two years and a codebase that survives inspection. Growth on its own no longer carries a deal, and a thin layer of product sitting on someone else's model is being priced accordingly.

The money is certainly there. PwC counts global technology, media and telecoms deal value up 48% to US$472 billion in the first five months of 2026, even as deal volumes fell 9%.[1] More capital chasing fewer deals means buyers are concentrating on assets they believe will hold their value and walking away from the rest. What follows is our read on where that line now sits.

A Big Market With Narrow Tastes

The headline numbers look like a boom. Crunchbase recorded 24 venture-backed companies acquired for US$1 billion or more in the second quarter, worth US$113 billion combined and the highest quarter on record, with SpaceX's US$60 billion deal for Anysphere, the maker of Cursor, doing much of the lifting.[2]

Underneath, software has been repriced. PitchBook's median multiple for public enterprise SaaS companies fell to 3.3x EV/revenue at 31 March 2026, from 4.9x at the end of 2025 and 6.2x a year before that, a slide it partly attributes to Anthropic's January launch of Claude Cowork alongside soft earnings and tariff worries.[3] BCG describes the resulting split plainly: infrastructure assets such as data centres, compute and power draw extraordinary valuations, while application-layer companies are correcting.[4] If you run a mature application business, that repricing is the backdrop to every buyer conversation you will have.

Efficient Growth Is the Entry Ticket

The Rule of 40, which holds that revenue growth plus profit margin should total at least 40%, has gone from board-deck slogan to the clearest dividing line in software valuation. In PitchBook's second-quarter comp sheet, public SaaS companies meeting it traded at a median 6.6x trailing revenue; those below it traded at 2.3x.[5] The same data shows an estimated median of 13.2% revenue growth and a projected 23.3% EBITDA margin for 2026, which tells you what buyers now consider normal.[5]

For private companies the implication is uncomfortable but simple. Growth bought with heavy burn gets discounted, and profitable growth at a slower clip holds its value better than founders who raised in 2021 might expect. Buyers will recompute your number from your own ledgers, so the version in the deck had better survive the exercise.

Net Retention Is the Tell

If the Rule of 40 tells a buyer how efficiently you grow, net revenue retention tells them whether that growth compounds. SaaS Capital's survey of private SaaS companies found those with net retention at or above 110% grew faster than the median, while those below 100% grew slower.[6]

PwC frames the 2026 version of the question more sharply. Buyers now want to underwrite current growth and retention and also "the durability of the software model in the AI era", particularly where pricing depends on seat growth.[1] A customer that replaces ten seats with one agent is a retention problem that has not yet shown up in your cohort data. Expect to be asked about it anyway.

Proprietary Data Beats a Clever Wrapper

PwC now sorts software businesses into AI-native, AI-resilient and AI-exposed, and says buyers are testing for defensible data, embedded workflows, high switching costs and a clear route to charging for AI.[1] Those tests explain most of the spread in 2026 pricing. Advisers at FE International report that pure wrappers, meaning a chat interface layered over commodity foundation models, struggle to attract serious buyer interest.[7]

The underlying issue is dependency, and it is not theoretical. In June 2025 Windsurf, then reportedly being acquired by OpenAI for about US$3 billion, said Anthropic had cut its direct access to Claude 3.5 Sonnet and 3.7 Sonnet with less than five days' notice.[8] Acquirers remember weeks like that. They will ask what happens to your product if your model provider changes its pricing, its terms or its mind, and the answers that hold up are specific: an architecture that can swap models, evaluation results on your own data, and inference costs already counted in gross margin.

Buying the Team Without Losing It

In AI, the team is often the asset, which makes retention risk a valuation input rather than an HR footnote. Research by Daniel Kim using US Census Bureau data found that 33% of acquired startup workers left within their first year, against 12% of regular hires.[9]

That risk has produced its own deal structure. In a reverse acquihire, the buyer hires the founders and key engineers, pays a licence fee for the technology and repays investors, leaving the original company standing.[10] Nvidia's reported US$20 billion non-exclusive licensing deal with Groq followed a version of that template, and in September 2026 it was reported that the US Justice Department is examining whether Nvidia tried to skirt antitrust scrutiny with it.[11] Earlier in the year, three US senators had urged regulators to treat Meta–Scale AI, Google–Windsurf and Nvidia–Groq as de facto mergers.[12]

For founders, the structure changes who gets paid. The central negotiation becomes how consideration splits between purchase price, which rewards shareholders, and retention packages, which reward whoever the buyer chooses to employ.[10] Staff who are not hired can walk away with very little unless someone argues for them. That conversation is worth having before a term sheet exists.

Diligence Now Reaches the Code

Technical diligence used to mean a senior engineer skimming an architecture diagram. Black Duck's 2026 open source report, built from audits it runs for M&A transactions, compliance and internal risk reviews, found licence conflicts in 68% of 947 codebases, the highest rate in the report's history, alongside a mean of 581 known vulnerabilities per codebase and high-risk vulnerabilities in 78% of them.[13] The same report flags AI coding assistants as a governance gap, so expect questions about who reviewed machine-written code and under what policy.

AI-specific diligence goes further. Skadden's guidance for acquirers covers data rights and provenance, whether compute costs remain viable at commercial volumes, concentration of expertise in a handful of people and undisclosed third-party dependencies.[14] It also shows where findings land in the contract: earnouts tied to AI milestones, escrow holdbacks, AI warranties treated as fundamental with longer survival periods and higher caps, and warranty insurers that increasingly exclude AI-specific risk.[14] Every gap a buyer finds turns into a price adjustment or a clause.

Regulatory Exposure Now Has Dates

The EU's Digital Omnibus moved the AI Act's high-risk obligations for Annex III systems from August 2026 to 2 December 2027, while systems generating synthetic content have until 2 December 2026 to meet transparency obligations.[15] In Australia, from 10 December 2026 privacy policies must disclose when personal information is used in automated decisions that could significantly affect individuals.[16] Geopolitics counts as well. In April 2026 China's National Development and Reform Commission blocked Meta's US$2 billion acquisition of Manus, an agent startup that had moved its headquarters to Singapore the year before.[17]

The Australian Angle

Australian deals now carry an extra step. Since 1 January 2026, acquisitions that meet the thresholds must be notified to the ACCC and cannot proceed until approved.[18] The thresholds reach beyond headline deals: a buyer group with at least A$500 million in Australian revenue must notify when acquiring a target with A$10 million or more in Australian revenue, and a merged group with A$200 million in combined Australian revenue must notify when the target has at least A$50 million in Australian revenue or the global deal value reaches A$250 million.[19] Build the review into the timetable from day one.

Local strategics are also buying rather than building. PwC's Australian outlook notes that companies are buying capabilities because speed matters.[20] Canva is the most visible example, with five acquisitions in 2026 by April, including AI collaboration platform Simtheory and marketing automation company Ortto.[21]

Preparing to Sell: What We Tell Founders

Our advice starts well before any process. None of it is complicated, and most of it is unglamorous.

  • Rebuild your metrics from the ledger. Rule of 40, net and gross retention by cohort, and gross margin with inference costs included. If the numbers move when a buyer recalculates them, you lose price and credibility together.
  • Document where the AI value lives. Data rights and consents, evaluation results on your own data, and a written answer to what happens if your model provider cuts you off.
  • Audit your own code first. Run an open source licence and vulnerability scan, fix what you can and disclose what you cannot. Surprises in diligence cost more than disclosures up front.
  • Put governance on paper. An inventory of where AI is used, an acceptable-use policy and a map of the rules that apply, including the EU and Australian dates above. Our AI policy work starts with exactly that kind of audit.
  • Reduce key-person risk, then plan retention. Decide early how you want consideration split between price and retention, and who you want protected if the buyer only wants part of the team.
  • Create competition. The best outcome rarely comes from a single inbound offer. Putting several credible strategic and financial buyers into a structured process is how tension gets built, and it is the core of our M&A practice.

The Price of Certainty

None of this means AI has broken tech M&A. Buyers are paying for what they always paid for: confidence that the business will still be worth the price after the integration team moves in. What has changed is how hard that confidence is to earn when agents threaten seat-based revenue, model providers can change terms at a few days' notice and regulators have started examining deal structures built to sidestep them.

Founders who can answer those questions with evidence rather than adjectives are being paid for it. If you are weighing a sale or an acquisition, start the conversation.

Frequently asked questions

How are AI companies valued in acquisitions in 2026?

Buyers increasingly separate AI-native, AI-resilient and AI-exposed businesses, and pay most for defensible proprietary data, embedded workflows and high switching costs. Efficient growth and strong net revenue retention still anchor the multiple. Thin wrappers over third-party foundation models, and seat-based software that AI agents could replace, are being marked down.

What is the Rule of 40 and why do acquirers care about it?

The Rule of 40 says a software company's revenue growth rate plus its profit margin should total at least 40%. It has become one of the clearest dividing lines in software valuation: PitchBook's Q2 2026 data showed public SaaS companies meeting it traded at a median 6.6x trailing revenue, against 2.3x for those below it.

What is a reverse acquihire?

A reverse acquihire is a deal in which a larger company hires a startup's founders and key engineers and pays a licence fee for its technology instead of acquiring the company outright. Investors are typically repaid and the original company keeps operating. The structure is under growing regulatory scrutiny, including a reported US Justice Department examination of Nvidia's licensing deal with Groq.

What do buyers check in AI due diligence?

Beyond the financials, buyers examine training data rights and provenance, dependence on third-party models, compute and inference costs at scale, concentration of key technical talent, open source licence compliance and exposure to AI regulation such as the EU AI Act. Gaps usually surface as price adjustments, escrows, earnouts or expanded warranties.

Do tech acquisitions in Australia need ACCC approval in 2026?

Since 1 January 2026, acquisitions that meet the ACCC's notification thresholds must be notified and cannot proceed until approved. The thresholds are based on Australian revenue and transaction value. For example, an acquirer group with at least A$500 million in Australian revenue must notify when buying a target with A$10 million or more in Australian revenue.

References