Two consortia of private equity firms are in advanced talks with Anthropic, with the stated intention of deploying its AI across their portfolio companies. Given that these groups own thousands of businesses between them, spanning mid-size manufacturers, software outfits, healthcare operators and just about everything in between, the reach of that rollout would be substantial.
The timing is not hard to explain. Private equity had a very good run of it five or six years ago, buying companies at keen valuations, loading them with debt, trimming the costs and selling at a handsome profit. That exit market has not recovered, and many funds are sitting on assets they cannot shift at anything like the multiples they were hoping for. Their executives, whose year-end bonuses tend to focus the mind in ways that strategy documents rarely do, need a new story. AI, with its promise of productivity gains and margin improvement, is the best one going.
What PE can do that individual companies cannot
There is something genuinely interesting in the model being proposed here. A mid-size distribution company or a regional care home operator does not have the resources or the incentive to build a centralised AI knowledge repository on its own. As part of a portfolio of thousands of companies sharing infrastructure, methodology and learnings, the economics change. It is the kind of thing that would not have been viable for these businesses otherwise.
There is also evidence that AI is working in specific areas. Customer service is the clearest example, where cost savings are real and implementation is contained. SaaS companies, which PE funds happen to own in large quantities, are adopting AI in defined workflow applications. The results, where they exist, are encouraging enough to make the investment case, even if they are not yet transformative.
The pricing question
A fair amount of current AI usage is subsidised. OpenAI and Anthropic are burning through tens of billions of dollars over the next five years to build market share, and many of the services companies are using are not priced at anything like a commercially sustainable level. The real test of enterprise demand comes when that changes, and businesses have to make a genuine return-on-investment calculation rather than an exploratory one. Some will find it stacks up. Others will discover they have been paying for something that does not justify the full cost.
Reliability is a separate issue, and one that gets glossed over in the enthusiasm. AI performs well in narrow, well-defined tasks. Outside them, the accuracy rate is less reassuring, and for companies implementing it at speed, under pressure from fund managers chasing their targets, the risk of deploying it in areas where it cannot yet deliver is real.
The cybersecurity angle
There is one category of business that benefits from all of this, whatever happens: cybersecurity. Companies installing AI tools at scale are expanding their attack surfaces, and the same technology that makes a customer service operation more efficient can make a breach more damaging. Security firms were well-positioned before PE started pushing AI across thousands of portfolio companies. They are better positioned now.
The SaaS problem
PE portfolios are full of SaaS businesses bought at peak valuations because they had recurring revenue, could absorb debt and had cost bases that could be trimmed. AI creates a new problem for some of them. Dashboard companies, those that aggregate data and present it in a usable format, are exposed to a technology that can do much the same thing at a fraction of the price. Software with deep workflow integration, the kind that takes years to embed in an organisation, is in a stronger position. Fund managers will be making this distinction over the next 18 months as they decide where to invest and where to cut their losses.
The exit question
If AI deployment works, margins improve, multiples recover, buyers return, and PE firms can move the assets they have been sitting on. That is a plausible outcome, and if Anthropic's own growth is anything to go by, the demand side of the equation is holding up. It is also two to three years away at minimum, which is a long time to keep limited partners patient.
What is not in doubt is that the experiment is coming. Thousands of companies that have never had to think about AI are about to have it imposed on them by their owners. Some will find it useful. Others will find it expensive and inconclusive, and the conversation about what comes next will be worth watching.