A House of Cards

I round up the most relevant AI-in-finance news - the deals being done, who’s rolling out what, and what’s actually working on the front lines

…. Blue Owl halted redemptions on its tech-focused credit vehicle. Partners Group distanced itself from software. Thoma Bravo and Vista held emergency investor calls. And Bloomberg revealed billions in software loans hiding in plain sight, miscategorised across lender books.

Meanwhile, the job cuts are no longer theoretical. Munich Re announced 1,000 positions going to AI automation. AIG processed 370,000 insurance submissions last year using agentic workflows, without adding a single head.

But first, my take on the hidden exposure in private credit. And why the structural parallels to 2007 are getting harder to ignore.

In This Week’s Issue:

From The Trenches

A House of Cards

Apollo's David Sambur didn't mince words. "Only now have people focused on the multi-car pileup that's about to happen on the software investing highway," he told Bloomberg on February 20. "It was all there if you were willing to see it in 2022."

He's right. And his timing is no accident. Apollo has zero software exposure in its PE business. Less than 2% across the entire $938 billion platform. In a letter to clients, they called the decision to steer clear "an investing and risk management decision." They're doing a victory lap.

But the real story isn't Apollo's positioning. It's the scale of what everyone else is sitting on.

PE firms spent a record $348 billion on software acquisitions in 2021 alone, often at rich valuations, drawn to recurring revenue and loyal customer bases. I wrote two weeks ago that between 2015 and 2025, the industry acquired over 1,900 software companies for $440 billion. Those investments are now hitting the end of a typical holding cycle. And there may not be buyers at the prices GPs need.

The exposure is almost certainly larger than reported. Bloomberg reviewed thousands of holdings across seven major BDCs and found more than 250 investments worth over $9 billion that weren't labelled as software loans, even though the borrowing companies are clearly software businesses. A pricing-software firm categorised as "business services." A restaurant management platform filed under "food products."

If the lenders themselves can't accurately classify their own book, what does that tell you about the industry's ability to assess its real exposure?

Blue Owl found out the hard way. OBDC II halted quarterly redemptions on February 19 after withdrawal requests overwhelmed the fund. Its tech-focused vehicle, OTIC, saw requests hit 15% of NAV. The firm sold $1.4 billion in direct-lending investments to provide liquidity. Structured notes tied to Blue Owl fell below 50% of face value.

Partners Group, managing $185 billion, went on record the same day saying it had "intentionally limited and down-sized its direct exposure to technology companies." Thoma Bravo and Vista Equity Partners have been holding emergency meetings with investors. When the two biggest software-focused PE firms are in damage-control mode, the mood has shifted.

Does any of this feel a little 2007 to anyone else?

I don't say that lightly. But the structural parallels are hard to ignore. In 2007, the risk wasn't in any single mortgage. It was that everything was connected to everything else and nobody could trace the exposure. Mortgages packaged into CDOs, packaged into CDOs-squared, rated AAA by agencies that didn't understand what was inside them. The people closest to the risk couldn't see it because it was distributed across so many layers.

Now look at what's happening with AI. Private credit funds lend hundreds of billions to build data centres and AI infrastructure. Those same funds hold billions in software loans that AI is about to impair. The AI tools being deployed by AIG, Munich Re, and Goldman are eliminating jobs, which reduces consumer spending, which hits the portfolio companies that the same PE firms own. And the whole thing is being funded by Wall Street underwriting over $1 trillion in AI-related debt, concentrated across every asset class simultaneously.

It's recursive. AI needs infrastructure. Wall Street funds the infrastructure. AI makes the software those lenders hold worthless. The losses hit the same balance sheets funding the buildout. And at each layer, the exposure is miscategorised, under-reported, or hidden in SPVs designed to keep it off the books.

I'm not predicting a financial crisis. The scale is different, the instruments are different, and the regulators learned some lessons (some). But the pattern is familiar. Concentration risk that nobody's pricing. Exposure that nobody can accurately measure. And an industry-wide assumption that the music won't stop.

Sambur called it a "failure of risk management." He's being generous. The thesis that SaaS was safe, that recurring revenue was predictable, that customer lock-in would protect margins, all of it assumed the product couldn't be replicated by something better and cheaper. AI changed that assumption. Most of the industry didn't react until it was too late.

As Sambur put it: "You're going to see what happens when they sell these businesses. It's going to take a while for this to play out."

He's right about the timeline. But the direction is already clear.

News Digest

Munich Re Cuts 1,000 Jobs as AI Takes Over

Munich Re's primary insurance unit Ergo announced on February 17 that it will eliminate approximately 1,000 positions in Germany over the next five years. The cuts target call centres and claims handling, driven by AI automation. Ergo employs about 15,000 people in Germany.

The move is part of a broader programme to reduce costs by approximately €600 million annually by 2030. Ergo says there will be no forced redundancies. It also plans to retrain up to 500 employees for growth areas like retirement planning.

The details:

Why it matters: This isn't a pilot programme or an aspirational target. It's a named company, with a specific number, on a specific timeline. The era of "AI will eventually impact jobs" is over. This is the era of "here are the job cuts."

My take: The numbers are modest. 200 per year across a 15,000-person organisation. But the signal is what matters. When a 144-year-old German insurer starts structured workforce reduction driven by AI, every HR department in financial services takes notice. And multiply this across every company running the same playbook. The efficiency gains are real. The question nobody's modelling is what happens to demand when enough companies make the same move simultaneously.

Read more

AIG Deploys Agentic AI Across Commercial Insurance

AIG's Lexington Insurance unit processed over 370,000 submissions in 2025, ahead of its 2030 target of 500,000. CEO Peter Zaffino told investors the company built "an orchestration layer in the technology stack to coordinate AI agents" across intake, risk assessment, and claims handling.

AIG Assist, the firm's internal AI tool, is now deployed across most commercial business lines. Zaffino described "a massive change in our ability to process submission flow" without requiring additional human resources.

The details:

Why it matters: This is what production-grade AI deployment looks like in financial services. Not a chatbot. Not a pilot. An orchestration layer coordinating multiple agents through the full workflow.

My take: The contrast with the private credit story is striking. AIG is one of the firms actually deploying AI at scale, processing more work with fewer people, and accelerating ahead of its own targets. The insurance industry has always been an early adopter of automation. But the orchestration layer is new. Coordinating multiple agents through a complete workflow is exactly the architecture that makes legacy software redundant. It's also the same architecture making those 1,000 Ergo jobs disappear. The efficiency gain and the job loss aren't separate stories. They're the same story.

Read more

Wall Street's $1 Trillion AI Bet

Wall Street is underwriting over $1 trillion in debt to fund AI expansion. Data centres, private credit facilities, hyperscaler infrastructure. AI is concentrating stocks, bonds, private credit, and the broader economy around a single bet. The eight largest companies, all tech firms with AI ambitions, now make up nearly half the S&P 500.

UBS estimates hyperscalers could spend up to $700 billion from their balance sheets on AI this year, while issuing unprecedented debt to access more capital. Morgan Stanley estimates private credit could supply half of the $1.5 trillion needed for data-centre buildouts.

The details:

Why it matters: Concentration risk across every asset class. AI exposure isn't just in software portfolios. It's in the equity indices, the bond markets, and the private credit books. A single thesis is being spread across the entire financial system.

My take: Read the Axios piece and then read the Blue Owl story again. Private credit is simultaneously funding AI's rise and being destroyed by it. The same asset class lending hundreds of billions for data-centre construction is also sitting on billions in miscategorised software loans that AI is about to impair. Lending into the boom with one hand. Holding the assets it destroys with the other. That's the kind of structural tension that doesn't resolve quietly.

Read more

Other Interesting Things I’ve Read of Seen This Week:

KPMG partner fined A$10,000 for using AI to cheat on AI training test (Feb 16) - A KPMG Australia partner uploaded training materials into an AI tool to pass the firm's mandatory AI course. 27 other staff were caught doing the same thing. You genuinely cannot make this up.

Mistral CEO: 50%+ of enterprise SaaS will shift to AI (Feb 18) - Arthur Mensch at the India AI Impact Summit put a number on it. Over 100 enterprise customers actively exploring replacement of older software stacks. But he drew a key distinction: workflow software is vulnerable, systems of record are not.

SaaS multiples collapsing from 20-30x to 5-10x (Feb 22) - Harry Stebbings shared Klarna CEO's take that software revenue multiples could compress further to utility-level pricing. If you own software in a levered equity fund, this is the chart that keeps you up at night.

Anthropic launches Claude Code Security, cybersecurity stocks tumble (Feb 20) - An autonomous vulnerability scanner that reasons through entire codebases like a human security researcher. CrowdStrike, Cloudflare, and Zscaler all dropped on the news. When the AI company starts eating the security companies too, nobody's safe.

Sequoia leads $1B seed for Ineffable Intelligence (Feb 18) - Ex-DeepMind founder David Silver raised Europe's largest-ever AI seed round. Backed by Nvidia, Google, Microsoft. A billion-dollar seed. The capital flowing into AI isn't slowing down. Even as the capital trapped in legacy software burns.

Acquisition Intelligence is a weekly newsletter on AI in M&A for finance professionals, private equity investors, investment bankers, corp dev teams, and deal-makers.

For questions, feedback, or to share what you're seeing in the market, reply to this email.

P.S. I'm Harry, co-founder of DealSage. We're building an AI-native deal intelligence platform to help professionals turn their institutional knowledge into better decisions. If you're curious what we're up to, check out dealsage.io or just reply here