Sponsors put a number on AI as LPs and buyers start pricing the risk.
Hg turns portfolio AI into a budgeted EBITDA line.
Hg put a hard number on portfolio AI this week. On 21 September it said its portfolio runs more than 1,600 live AI projects carrying over $260m of budgeted EBITDA impact, a fivefold rise since its work with Anthropic began in early 2024 ([1]). It is extending that relationship into a new venture in which Hg Catalyst, its in-house AI product incubator, works directly with Anthropic to deploy agentic AI across portfolio companies ([2]).
The framing matters as much as the figure. Hg describes the number as budgeted EBITDA, which puts AI inside the value creation plan rather than in an innovation budget. Its argument is that its companies typically own the system of record and the workflows in their sectors, and that the value is created where those workflows meet the model layer ([3]). The stated aim is to move beyond experimentation to agents that carry out tasks inside business workflows ([4]).
Hg also tied AI to realisations. It cited Quantios, which launched its first agentic product in March with Hg Catalyst support before Hg sold it to Vista Equity Partners, alongside GTreasury and Intelerad, as evidence that businesses with credible, product-embedded AI are attracting stronger buyer interest ([5]). Treat that as the seller's account: Hg has not published the exit multiples.
For operating partners the lesson is method, not vendor. A portfolio AI programme that is counted project by project, rolled up into budgeted EBITDA and linked to named exits gives an investment committee and a buyer something to test, in a way that counts of pilots or licences do not. Hg has set a clearer standard for what counts as evidence, and other software sponsors are likely to be asked to match it in fundraising and at exit.
Hg's first greek deal shows the AI thesis at entry.
A day later Hg agreed a majority investment in EnterSoftone (ES1), a Greek ERP and business software group, in its first deal in Greece, valuing the business at around €1 billion ([1]). ES1 was formed in 2025 by merging Entersoft and Softone, serves more than 90,000 clients and is a category leader in Greek ERP and digital invoicing ([2]).
The investment case leans on regulation and AI together. Greek small and medium-sized businesses are adopting digital tools quickly because of mandatory B2B e-invoicing under the myDATA framework, and the EU's VAT in the Digital Age reforms are pushing other member states the same way ([3]). Hg plans to complete ES1's move to the cloud, expand into neighbouring markets and use Hg Catalyst to build AI products across the platform ([4]). Hg Capital Trust alone will invest about £33 million ([5]).
Read the two Hg announcements together and the playbook is plain. Buy a compliance-anchored system of record in a fragmented mid-market, where switching costs tend to be high and the data is proprietary, then add AI products through a central team. That is a different risk profile from mainstream business software, which sold off this year when plugins extending Anthropic's Claude Cowork to its tasks helped wipe roughly $285 billion off public SaaS valuations in 48 hours ([6]).
For deal teams in the UK and Europe this gives a practical diligence screen with three questions. Is the target's revenue protected by regulation or by ownership of the workflow? Would its data let it ship agentic features before a new entrant could? Does the sponsor have the in-house capability to deliver those features inside the hold period, rather than relying on management to find it? A target that fails the third question needs a plan for hiring or partnering priced into the model from day one.
Software exits reopen, but the split between assets widens.
Hellman & Friedman is exploring a sale of Applied Systems, which sells software to insurance agencies and brokerages, at a value of up to $10 billion, with JPMorgan and Goldman Sachs advising, according to Reuters reporting carried by Insurance Journal on 24 September ([1]). The business generates more than $550 million in annual EBITDA, and H&F bought it from Bain Capital for about $1.8 billion in early 2014 ([2]). On those numbers the price would be no more than about 18 times EBITDA. A sale would rank among the largest software buyout processes this year and test investor appetite for mature software assets, and sell-side activity in software has picked up in recent weeks ([3]).
The same week showed the other side. Clearlake and Charlesbank agreed to put around $175m of preferred equity into Symplr, a healthcare software provider, in a restructuring that pushes out debt maturities, after months in which its debt traded at distressed levels amid concerns about AI's impact on software ([4]). Symplr's first-lien loan of about $1.2bn was quoted at about 71.4 cents on the dollar, against 86.5 cents at the start of January, and the company is held through a continuation vehicle as its sole asset ([5]).
For portfolio CFOs, the gap between these two outcomes is the AI question in financial form. Buyers are still prepared to pay for vertical workflow software with sticky customers and cash flows they can underwrite. Lenders have cut their valuations of loans to Clearlake-backed Cornerstone OnDemand and Symplr by more than 30% ([6]), and the pressure falls hardest where the product looks easier to replace and the capital structure leaves little room to rebuild. Before launching a sale, sponsors should expect buyers to test AI exposure as hard as they test churn, and should have the evidence ready: product roadmap, revenue from AI features, retention by cohort and the cost of keeping the product current.
LPs are pricing AI exposure into commitments.
PitchBook data published on 23 September show tech private equity fundraising recovering but concentrating. Tech-focused funds have raised $31.3 billion in the third quarter so far, after only $24.7 billion in the first half, a drop of almost 70% year on year ([1]). Fund count stayed broadly flat at 17, and PSG and Francisco Partners accounted for over 80% of the total ([2]). PitchBook links the first-half lull to January's launch of Claude Cowork and the plugins that followed ([3]).
LPs are changing how they underwrite managers. Caisse des Dépôts' head of PE and infrastructure funds said his team now has to understand AI resilience and AI opportunity within the same company, and is looking at new criteria and KPIs to assess risk ([4]). PitchBook flags the 2018-2022 vintages as particularly hard to price because entry valuations were very high ([5]).
Caution also runs the other way. The $327 billion New York City Retirement Systems passed on a private equity fund because it was heavy on AI holdings, with its chief investment officer citing the need to diversify rather than fear of the technology ([6]). The wider backdrop is weak: managers had raised $211.9bn up to 11 September, on course for the weakest fundraising year since at least 2020, according to PitchBook data cited by the Wall Street Journal ([7]). PitchBook also estimates that private equity put an average of 14% of its capital into software over the past decade ([8]).
The consequence for GPs is that AI now sits in fundraising diligence from both directions: too exposed to AI disruption, or too concentrated in AI winners. Managers need a portfolio-level view of which companies AI threatens, which it helps and how they know, expressed in the KPIs LPs are starting to request. A generic AI slide in the pitch deck will not survive that conversation.
AI rewrites the thesis while hold periods stretch.
BDO's 2026 Private Equity Survey, released on 22 September, polled 400 US fund managers, deal partners and operating partners in May ([1]). 94% said AI is fundamentally reshaping their investment thesis, with firms applying it at every stage of the deal from sourcing to exit ([2]). At the same time, 80% are holding portfolio companies for five years or longer, and 82% expect deal prices to rise over the next 12 months ([3]).
Put those together and AI becomes a hold period problem as much as an entry one. Assets bought on a five-year plan are being held longer, so many will live through shifts in AI capability that the original value creation plan could not have priced. BDO says firms are coming to sellers with well-defined growth plans and partnering with management to win deals ([4]), which means the AI line in the plan increasingly decides who wins the auction and what the next buyer will believe.
The binding constraint is people. 48% of respondents named finding and retaining the right talent at fund and portfolio company level as their biggest limitation when trying to execute deals at speed ([5]). That is consistent with Hg's model, which runs AI through a dedicated central team working inside portfolio companies rather than leaving each management team to build its own ([6]).
Two cautions apply. The sample is US only, and the fieldwork took place four months before publication. Even so, the direction matches this week's deal and fundraising news: AI is now a valuation input at entry, a performance line during the hold and a diligence question at exit. Value creation leads should check that every portfolio company's plan states which of the three it is solving for, with a number and an owner attached.