Value Creation · Deep Dive
By Atul Tiwary, El Dorado Capital · Last updated: 7 October 2026
Four fund-level models have emerged to run AI value creation across portfolios. They are not variations on one idea — different equations, different evidence bases, different failure modes — and the security read-across leads, because that is the market where the full trade has not yet been run in public. Company-level mechanics are on the value creation deep dive.
The security read-across. Managed-security consolidation is already running at classic services multiples — LevelBlue assembled Trustwave, Stroz Friedberg and Cybereason into the largest pure-play MSSP, and half of the sector's recent platform deals are sponsor-backed. Trustwave's own path into that platform is instructive on its own: a decade under a strategic telecom parent that wrote the asset down, sold off its healthiest sub-unit in isolation, and exited via a seller-financed note — read as a failure case on the evidence page — which is a reminder that the roll-up's inputs were not uniformly healthy assets bought cheap. Agentic SOC tooling is commercializing on a separate track. The full trade — buy sub-scale MSSP and MDR revenue at services prices, replace tier-1 and tier-2 analyst cost with an agentic platform, and hold or exit at the re-rated multiple — has not been run end to end in public. It came closest on 16 September 2026, when EQT's Mid Market Europe fund agreed to sell Ontinue, a Microsoft-first MXDR provider with more than 250 customers on a 24/7 delivery model, to Eterna-backed Quorum Cyber. Ontinue is the operating half of the thesis already executed (and not the first agentic-SOC asset to trade: Cribl bought Radiant Security’s AI-SOC technology on 19 August and Rapid7 bought Kenzo Security in March, neither with a disclosed price): carved out of Open Systems in 2023, three technology platforms consolidated into one AI-native stack, three add-on acquisitions in data science and AI, and scale that EQT describes as roughly doubled. Terms were not disclosed, so the first agentic-SOC platform at services scale to change hands did so without a public exit multiple, and the re-rating that carries most of the return in this trade stays unpriced. Security's 24/7 labor intensity and regulation-driven demand make it arguably the best-suited vertical for this trade on the board — with a failure mode none of the HOA or parking rollups carry, which is that an autonomous response error creates liability, not just churn. That gap between consolidation-as-usual and the unbuilt archetype is where the next fund in this market gets formed. The screen is now being run name by name against the public tape, starting with Rapid7 at 1.1× ARR.
Everything above describes what AI does inside a single company. In the past eighteen months the mechanisms have acquired a fund layer — vehicles built specifically to run these plays across a portfolio — and the design choices those funds have made are the clearest available statement of which levers investors believe are real. Four models have emerged. They are not variations on one idea. They carry different value-creation equations, different evidence bases, and different failure modes, and conflating them is the fastest way to misprice a deal.
Bending Spoons is the reference case for the first model: buy mature digital revenue cheaply, replace the acquired cost base with one shared technology and monetization platform, reprice the captive user base, hold permanently. The record runs from Evernote (2023) through Brightcove ($233M), Vimeo ($1.38B all-cash), AOL (~$1.4–1.5B, financed with a $2.8B debt package) and Eventbrite (~$500M) and Tractive ($759M enterprise value, May 2026), to Airtable at a $1.285B enterprise value against roughly $480M of ARR — about 2.7× EV/ARR for a business that last raised private capital at $11B — and on to Miro, agreed 10 September 2026 at a $1.355B enterprise value on roughly $600M of ARR, about 2.3×. Miro's implied equity value of ~$1.79B sits roughly 90% below the $17.5B at which it was marked in 2022, equity basis against equity basis. Two former decacorns in five weeks inside a 2.3–2.7× ARR band is the useful part: the entry multiple is not a negotiated one-off, it is the model's standing bid for scaled software carrying a neglected cost base. The operating pattern is consistent: deep headcount cuts (Evernote went from 341 people to 60, roughly 82%, and about three-quarters of WeTransfer left; headcount at the post-IPO acquisitions is not disclosed), migration onto a central subscription and data platform, and hard repricing — Evernote's personal plan went from $80 to $130, a 63% increase. The July 2026 Nasdaq listing is what changes the analysis: the model now has a listed currency and auditable accounts, and both cut both ways. The currency is already load-bearing: in the Miro agreement certain selling shareholders committed $295M of their proceeds — roughly a sixth of the equity consideration — to newly issued Bending Spoons stock, a financing and alignment lever the private structure did not have. Revenue of $1.31B in 2025, up 95%, and roughly $2.6M of revenue per employee need to be read against how the growth was bought. Second-quarter 2026 revenue of $704M was up 126%, while organic growth was 3%. Debt principal stood at ~$4.9B at 30 June against $793M of cash: 2.4× leverage on the company’s own measure, 4.7× on Moody’s adjusted basis, which on 23 September upgraded the credit to Ba3 with S&P moving to BB-. A $1.25B and €395M term loan B add-on, upsized from $750M and €250M, priced on 25 September to fund Miro. The earnings quality is contested as well: the F-1 add-backs of $335M exceeded 2025 operating income of $278M, and the company now guides on adjusted operating income, where the Q2 add-backs were $141M against $240M of reported operating income, about 59%. Full-year guidance of $1.46–1.51B of adjusted operating income excludes Airtable and Miro. An unresolved dispute remains over whether retention on repriced bases is as stable as management claims. The equation is entry-multiple arbitrage plus opex collapse plus pricing power. The first two are proven. The third is where the argument lives.
The second model buys labor-priced revenue — accounting practices, property management, legal services, parking — at services multiples, converts the workflow to software economics on a shared AI platform, and aims to exit at a software multiple. This is where the new capital is concentrating. Thrive Holdings raised $2B in August 2026 at a $12B valuation, with OpenAI holding equity that grows with portfolio performance, and runs 70-plus businesses across accounting and managed IT. General Catalyst's dedicated Creation strategy (~$1.5B) has co-created roughly ten platforms, among them Long Lake — thirty-plus acquisitions in under two years, and now the escalation trade: a $6.3B all-cash take-private of Amex GBT at $9.50 a share, announced in May and closed in the last days of September 2026 — the first time this playbook has swallowed a large public company. Closing settles the transaction, not the thesis: the equity came from Long Lake's investors and Koch Equity Development against $2.5B of committed bank debt, and no margin result exists yet for a business that has been private for days. It is also no longer a one-off: on 14 September Sequence AI Holdings and Dell’s family office agreed a $7.7B take-private of The Baldwin Group, an insurance broker, at $32.50 a share (about $4.6B of equity plus $3.1B of net debt, closing expected in the first quarter of 2027). Metropolis proved the operator variant by taking SP Plus private for $1.5B and becoming North America's largest parking operator. Capital formation and acquisition velocity are verified. What is not yet verified, anywhere in public, is the margin conversion at scale — and that conversion is most of the return. The entry-to-exit multiple gap, services EBITDA in and software revenue multiple out, is the bet itself. A buyer underwriting one of these platforms is underwriting the re-rating, not the current P&L.
The venture side has now named and codified the same conversion. Emergence Capital — among the earliest institutional backers of Salesforce, Veeva and Zoom — coined "AI-native services" (AINS) in 2024 for companies that use AI to deliver the complete outcome rather than sell a tool, closed a $1B seventh fund in March 2025 aimed at the category, and publishes its underwriting criteria in the open: the vendor owns the outcome; AI is built into delivery rather than bolted onto a labor process; the advantage is measurable in speed, cost or quality; AI leverage rises over time, read through gross margin and revenue per employee; the team has earned domain trust; and the workflows are high-volume enough that each engagement improves the next. The portfolio claims are the aggressive version of this model's thesis — a dozen engineers at Strala processing insurance claims more accurately than thousand-person incumbent TPAs, Mechanical Orchard running Fortune 500 mainframe migrations at a claimed 10× the pace of the Big Four — and Sequoia reached the same conclusion in 2026 under the heading "Services: The New Software." Two details in the Emergence playbook matter for anyone underwriting the fund layer above. First, the six criteria read as a diligence checklist for this model: a labor-conversion platform whose acquired businesses would fail them — no outcome ownership, no rising revenue per employee, no compounding data — is buying services revenue and calling it software. Second, the playbook's own M&A chapter warns founders against acquiring too early, while its companion essay argues the roll-up is the underused path to venture-scale AINS. Venture's sharpest thinking on the category disagrees with itself about timing, and that disagreement is the information: the build side and the buy side are now pricing the same margin conversion from opposite ends of the entry multiple.
The third model incubates the AI-native operator first and bolts M&A on afterward — creation at inception. It has the least evidence and the longest runway. The fourth is incumbent private equity's overlay: Vista's agentic tooling running in roughly a third of its ~100 companies, Hg's data platform, portfolio-wide model partnerships in negotiation at the largest firms. The deployment is real. The published, audited margin results attributable to it are scarce, and the February 2026 software selloff pushed the large sponsors into spending the year reassuring LPs that AI is not eroding their existing marks — which is a defensive posture, not a value-creation equation. The one quantified sponsor result found is self-reported and narrow: Partners Group attributes a $10M EBITDA uplift, about 120 basis points of margin, to two agentic tools built for Foundation Risk Partners, an insurance brokerage. Constellation Software's drawdown over the same period is the market repricing the decentralized-compounder design against the centralized-platform design, and its second-quarter commentary gave the market little to reverse it: AI is lifting development productivity, but management put the revenue benefit as still “a ways out”, with organic growth running below its historical 5–6% range. For anyone building a serial acquirer today, autonomy versus platform is no longer a culture preference. It is the load-bearing choice.