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.
The Company Is the Dataroom
The week's story is a 24-year-old. Situational Awareness, the $20 billion-plus AI hedge fund run by OpenAI alumnus Leopold Aschenbrenner, hit margin trouble in the sell-off and sold the bulk of its public holdings to Citadel in a deal hammered out overnight with Ken Griffin, a swoop the FT credits with helping stem a $3 trillion AI rout.
Elsewhere the capital keeps arranging itself. Amazon completed its $50 billion investment in OpenAI, banks are in talks to lend $15 billion against an Anthropic data centre, the FT reports Big Tech credit metrics at new highs, and Ares has held talks to buy Leonard Green.
Also inside: Goldman's new AI investing platform, an AI-native ETF factory chasing BlackRock, a $100 billion campus on a Cold War uranium site, and one-person companies clearing $1 million.
But first, my take on the future of deal making, because once buyers and funds can plug straight into a company's data, a surprising amount of the process starts to look like ceremony.
In today's Acquisition Intelligence:
From The Trenches:
The Company Is the Dataroom: the whisper chain, the 360 view, and going direct to stakeholder
What The Builders Are Saying:
The record earnings underneath the AI trade, and Guillermo Rauch on craft
News Digest:
The AI trade's golden child got a margin call
Amazon, Anthropic and the tightening AI capital web
Ares in talks to buy Leonard Green
Other Interesting Things I've Read or Seen This Week:
Goldman's AI platform, the Corgi ETF factory, Dwelly's rollup, a uranium plant turned AI campus, Lloyds' strategy reset, and one-person companies
From The Trenches
The Company Is the Dataroom
The most under-utilised source of value in private equity, I've come to believe, is the relationship between the fund and the companies it owns. A fund holds control positions, sits on the boards, approves the budgets, and it still experiences each portfolio company through a monthly pack that arrives weeks after the fact, four templates deep. All that distance is upside waiting to be collected.
Last week I argued that most knowledge work is transport, information moving between systems that cannot talk to each other, with a person serving as the pipe, and I used portfolio reporting as the example. The time cost is the visible part. The deeper cost is what the chain does to the truth.
The Whisper Chain
Watch the pack get made. The portco team exports from their systems and massages the numbers into their own template, the fund parses that into its template, the deck gets cut again for the board and once more for the LPs. Every hand in the chain has slightly different incentives and slightly less context than the last, so the message distils at every hop, a long game of telephone where the numbers survive and the meaning frays.
Nobody designed this and nobody is to blame for it. The systems could never talk to each other, so a chain of templates grew in the gap, and it has been in place so long it reads as normal. It's only in working inside these companies that the scale of it has become clear to me: the sheer energy the chain eats, and how little of what reaches the fund is the company as it actually is.
The 360 View
We spend a lot of our time now inside portfolio companies fixing exactly this. We connect the systems where the business actually runs, the ERP, the operating systems, the billing system, the HR system, and standardise everything they hold into a single source of truth. The operators get a 360 view of their own business, often for the first time.
Then the fund gets access to the same source, and the whisper chain has nothing left to carry. A partner can ask why margin moved in one segment and get the answer from the same data the company runs on, current and traceable, in minutes rather than at month-end. You see problems while they're still cheap to fix, you spot patterns across companies, and the board meeting stops being where surprises live.
My view is that this closeness becomes one of the biggest drivers of value in private equity from here. The gap between fund and portco was always a plumbing problem, and the plumbing is finally getting solved.
The Same Foundation, At Exit
For a while I treated that as a monitoring story, separate from what I think is happening to deals. The realisation crystallising for me over the past few weeks is that it's one story: a company whose data is structured and queryable is easier to run, and it's also prepared for a transaction, whether that comes next quarter or in five years.
Think about what a sale process is today. Bankers reshape management accounts into a databook and a CIM, buyers rebuild the CIM into their own model, and then diligence runs the whole relay in reverse, hundreds of portal questions answered in PDFs and re-keyed by associates on the other side. Translation stacked on translation, and every translation is a place for the truth to leak.
Now sell the company that already runs on a structured foundation. The preparation has been happening continuously, just by operating, and the value story is evidenced in the data rather than asserted on a slide. Diligence becomes controlled, direct access: the buyer plugs in, asks the revenue recognition question, and the answer comes from the ledger with the source attached, in minutes.
The CIM and the dataroom were workarounds for a world where a company's data could not be accessed or trusted. Structure the data properly and the company becomes the dataroom.
You can tell where this is heading by watching the incumbents. One of the major dataroom providers deliberately blocks third-party API access to documents its own clients own, which is a rational thing to do when your product is the toll booth. It is also exactly what the fax machine people did.
What This Means for the Middleman
None of this makes the intermediary redundant. Judgment, narrative, negotiation, relationships, the discipline of running a competitive process: none of that was ever transport, and it's what clients pay for. What changes is the posture: today an advisor gates access because unstructured data is a liability in a process, and the advisor of the next decade wins the mandate by getting the company queryable earliest and making direct access safe.
So the takeaway is the same from every seat at the table. The structured company is the asset. It runs better day to day, it reports to its owners without the whispers, and when the time comes it transacts faster and at a better number, because buyers pay up for what they can verify. A fund that builds this across a portfolio collects all three, every day it owns the company and once more on the way out.
The economics of advisory fees in that world deserve their own issue, and I'll come back to them, because I don't think success fees at today's levels survive a diligence process that runs in weeks.
What The Builders Are Saying
Two posts this week, both from this morning, and they bracket the issue rather neatly: one on the earnings underneath the AI trade, one on where the individual edge goes. Worth following both accounts.
@KobeissiLetter (The Kobeissi Letter, markets commentary) on the earnings behind the boom
The post: S&P 500 companies are beating Q2 2026 estimates by an average of 27 per cent, on track for the strongest quarter in decades. Nasdaq 100 names are beating by 55 per cent, and the Bloomberg AI Value Chain index, chipmakers through cloud, memory, networking and power infrastructure, is beating by 71 per cent.
Why this matters: the same week a $24 billion AI fund got carried out on margin, the earnings underneath the trade came in at record beats. The blow-up above was a financing story, and the operating results climbed right through it.
@rauchg (Guillermo Rauch, CEO of Vercel)
The post: "AI alone is cool. But mastery + creativity + AI hits on a whole different level. Don't let anyone discourage you from pursuing excellence and craft. Keep studying the blade."
Why this matters: this is the middleman question from the essay above, answered in one line. The tools gear expertise rather than flattening it.
My take: read together, the two posts carry the issue's argument from opposite ends. The results say the boom is real however messy the financing gets, and Rauch says the returns on it go to the people who pair genuine craft with the tools. In deal work that means the banker whose judgment compounds once the transport is gone, and the fund that reads its portfolio straight from the source.
News Digest
The AI Trade's Golden Child Got A Margin Call
Leopold Aschenbrenner wrote the 165-page essay that named the AGI trade, raised a hedge fund on the back of it at 22, and ran it to $24 billion. Then July happened. The FT's account is the read of the week: debt-fuelled, concentrated AI positions, prime brokers calling for margin, and a Wednesday-night phone call with Ken Griffin that ended at dawn with Citadel owning the bulk of the fund's public holdings.
The details:
Situational Awareness was up more than 400 per cent in the first half of 2026, and 1,551 per cent since founding; it fell 67 per cent in July as positions like Bloom Energy and Sandisk dropped about 40 per cent
Citadel beat Millennium and Jane Street to a public portfolio once estimated at $16 billion, a swoop the FT credits with helping stem a $3 trillion AI rout
Aschenbrenner kept his private stakes, including Anthropic, which he had shopped at a premium to its $900 billion valuation
On July 24, mid-spiral, his investor letter called the unwinding "a particularly good time to add funds"
Reuters notes this is a familiar role for Griffin, who has stepped in as crisis buyer before; the WSJ's account runs under the headline "A Dire Situation"
Why it matters: leverage against concentrated AI positions is now large enough to move the whole market, and the FT reports collateral demands are spreading well beyond one fund.
My take: Aschenbrenner may still be right about AGI, and it didn't matter to the margin clerk, because the timeline of conviction and the timeline of financing are different things. A fund run by seven investment professionals moved the global market twice in one week, once on the way down and once by selling to Citadel.
Amazon, Anthropic And The Tightening Capital Web
Amazon has completed its investment in OpenAI, per the FT: roughly $50 billion of equity for a stake of about 5 per cent. The same week, Reuters reported banks are in talks to lend $15 billion against an Anthropic data centre backed by Google. Equity, debt and compute now interlock across the same handful of names.
The details:
Amazon takes roughly 5 per cent of OpenAI in an equity deal worth about $50 billion, a week after Microsoft's multibillion-dollar Mistral commitment
The $15 billion Anthropic data centre facility, first reported by the WSJ on July 30, would be one of the largest AI-linked debt packages yet
Nvidia is separately weighing arrangements to backstop financing for an OpenAI data centre project, per Axios, reviving "circular financing" concerns
The counter-current: OpenAI cut prices on its smaller models on July 30 as business customers scrutinise AI spend, and Microsoft's cloud beat on July 29 eased some spending concerns
Why it matters: the same names now appear as investor, lender, supplier and customer to each other, and every strand of that web gets priced somewhere, increasingly in the credit markets, where the FT reports Big Tech metrics at new highs.
My take: when the infrastructure owners hold equity in specific labs, the neutrality of the stack starts to matter. We keep DealSage model-agnostic partly for this reason: your vendor's cap table shouldn't decide which intelligence your firm runs on. Worth checking your own book too, because plenty of funds now hold the same AI names through three doors at once, equity, credit and infrastructure, and the Situational Awareness episode shows how fast the doors can slam together.
Ares Has Held Talks To Buy Leonard Green
The FT reported this week that Ares Management has held discussions about acquiring Leonard Green & Partners, the Los Angeles buyout firm. No deal is agreed, and talks of this kind often go nowhere, but the fact they happened at all says plenty about where the large alternative managers think this is going.
The details:
A deal would significantly expand Ares's private equity platform, adding one of the best-known LA buyout franchises
It would extend the consolidation wave among large alternative asset managers as fundraising concentrates with the biggest platforms
Why it matters: the intermediaries of capital are consolidating into platforms, and scale in distribution, data and technology increasingly beats boutique specialisation.
My take: the last round of GP consolidation was about fundraising reach. My bet is the next one is about operating infrastructure: platforms that can plug into their portfolios and run them on shared data will underwrite differently, and the boutiques will feel that gap in every auction. It's the fund-level version of the argument above, and Ares buying Leonard Green would be the clearest sign yet that the GPs see it coming.
Other Interesting Things I've Read or Seen This Week
Goldman Sachs' asset arm forms an AI investing platform (Reuters, July 30) - An internal memo lays out a dedicated platform for AI-related investing across the asset management arm. (The house view is now a product line.)
AI-fuelled Corgi chases BlackRock's ETF crown with a 500-fund push (Bloomberg, July 30) - The AI-native issuer is on pace for 500 ETF launches in year one, running about $3 million of revenue against an estimated $56 million break-even. (Venture math has entered the '40 Act.)
Dwelly raises $170 million for a real estate rollup (Bloomberg, July 28) - EQT Growth and General Catalyst back a UK startup buying letting agencies, 17 so far, and injecting AI into their operations, with KKR's Philipp Freise investing personally. (The rollup rediscovers its oldest trick: buy the mess, sell the tidy.)
A Kentucky uranium plant becomes a $100bn AI campus (AP, July 31) - The DOE picked Brookfield to build a data centre and gas power complex at the Paducah Gaseous Diffusion Plant. (From enriching uranium to enriching hyperscalers.)
Lloyds reports first-half profit up 23 per cent and outlines new strategy (Reuters, July 30) - The bank pairs the profit jump with a strategy reset. (Announcing a new strategy from a position of strength: a novel approach.)
How AI is helping one-person companies scale to $1 million and beyond (WSJ) - Solo founders are using AI to clear seven-figure revenue without hiring. (Key-person risk, in its purest form.)
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. The 360 view in this issue is the work we do: we connect a portfolio company's ERP, billing and HR systems into one source of truth, and give the operators and the fund the same window into the business. Reply here if you'd like to see what that looks like on one of yours, or have a look at dealsage.ai.