Apollo & Blackstone's $35B AI Power Play Redefines Infrastructure Financing

So, What Actually Happened?

So, Wednesday, and the headlines were all about who's going public while the real action was about who's quietly writing the checks. Apollo and Blackstone moved to finance $35 billion of raw AI computing capacity, the kind of number that used to come from a tech giant's bank account and now comes from private credit. We scanned 190,000 articles this week so you don't have to. Meanwhile Revolut lined up a sale at a $115 billion valuation, and an Australian lab quietly shipped the first AI to pass radiology's board exam.

The Bottom Line: The crowd watched the IPO stage. The money walked in through the back door, dressed as private credit, sovereign infrastructure, and a radiology report nobody applauded. Follow the financing, not the filing.

The Tracks That Matter

1. Private Credit Becomes The Quiet Banker Of AI

Here's the number that reframes the whole boom. Apollo and Blackstone moved to finance a $35 billion expansion of AI computing capacity, with the first tranche adding a full gigawatt, enough power to run about 750,000 homes, pointed entirely at training and serving models. That's not a venture round. That's a power utility being built on a private equity balance sheet.

What makes it a pattern, not a one-off, is the company it keeps. The same reporting notes Meta struck a $27 billion financing deal with Blue Owl Capital for its largest data center, and the broader arrangement aims to stand up more than 20 gigawatts of compute through 2028. When OpenAI and Anthropic both filed to go public this week, the headlines screamed equity. The structure underneath screamed debt.

This is the part that should change how you read AI spend. The build-out has gotten so capital-hungry that the people who normally finance toll roads and airports are now financing GPU farms. Private credit likes predictable, infrastructure-shaped cash flows, which tells you the market now treats compute like a utility, not a bet.

Here's what works: When you evaluate an AI vendor's staying power, stop reading the valuation and start reading the financing. A company funded by infrastructure-grade private credit can keep the lights on through a downturn. One running on equity hype and a burn rate cannot.

2. Revolut Eyes $115 Billion And Fintech's Reckoning

Here's a valuation that tells you where fintech actually landed. Revolut lined up a secondary share sale at a $115 billion valuation, letting employees and early backers cash out without the company going public. That's a private bank-sized number for a company that started letting you split a dinner bill in another currency.

The timing is the interesting part. While Revolut prices itself like a top-ten European bank, the rest of the sector is getting a cold shower. A sharp piece this week argued that after Europe's fintech fines, the Gulf's digital banks face their own reckoning, as regulators who let neobanks grow fast now want the compliance plumbing they skipped. The growth story and the governance bill are arriving at the same address.

The lesson generalizes past fintech. Every category that scaled on ”move fast, sort the rules later” is now meeting the regulator who took notes the whole time. The companies that priced in compliance early get the premium valuation. The ones that treated it as someone else's problem get the fine and the headline.

Here's what works: If you run anything regulated, audit your ”we'll fix it at scale” debts now, before a regulator does it for you. The cheapest compliance is the kind you build in before you're big enough to be worth fining.

Try It Yourself

3. S&P Global Hands Cohere The Keys To Its Data

Here's a partnership that says more about AI's real moat than any model launch. S&P Global expanded its AI ecosystem with Cohere to power trusted, enterprise-grade AI on top of its financial data. Read that carefully: the model didn't win because it was the smartest. It won because it was allowed near the data that matters.

This is the quiet shift under the loud IPO news. For two years the question was ”whose model is best.” Increasingly the question is ”whose data is the model standing on,” and the answer is rarely the model maker's. S&P owns decades of curated financial truth. Cohere brought the reasoning layer. Neither is interesting alone, and together they're a product a bank will actually pay for.

The strategic tell is who's doing the picking. A data owner choosing a model partner, rather than a model maker chasing a data deal, flips the power dynamic. When the company that owns the proprietary data is the one shopping for AI, the data is the scarce asset and the model is the commodity.

Here's what works: Inventory the proprietary data only your company holds, then treat it as the negotiating chip it is. In the next wave of AI deals, the side with the irreplaceable data sets the terms, not the side with the cleverest model.

4. Canada Wants A Divorce Plan For Its AI Systems

Here's the governance idea nobody else is saying out loud. A pointed policy piece argued that Canada needs an exit plan for its critical AI systems, the bureaucratic equivalent of a prenup: before you wire your hospital, your tax agency, or your grid to a vendor's AI, you write down exactly how you'd unplug it. Most organizations have no such plan, which means they're not adopting AI so much as marrying it.

The reasoning lands harder when you set it next to the harder edge of the same debate, where researchers are designing compute governance that actually works at the chip level. One end asks how a government keeps control of the compute. The other asks how it walks away from a system it already depends on. Both are really one question: who has the off switch.

For any leader, the framing is the gift here. We spend all our planning energy on adoption and almost none on reversibility. But a system you cannot exit is a system you cannot really govern, and ”the vendor changed the terms” is a worse position when there's no door out.

Here's what works: For every AI system you deploy in a critical workflow, write the exit plan before you sign. Document how you'd switch vendors, export your data, and run degraded-but-functional without it. If you can't describe the exit, you don't control the system.

5. Rio Bets A City On AI Data Centers

Here's where the sovereign-compute story stops being abstract. Rio de Janeiro deepened its partnership with Elea Data Centers for an AI city, after Elea pulled a $550 million investment from infrastructure manager I Squared Capital. A city government and an infrastructure fund are building physical AI capacity on home soil, the same move we watched Britain and Korea make last week, now landing in South America.

The connecting thread is who's financing it, and it rhymes with the top of this newsletter. Just like the $35 billion compute deal, this is infrastructure money, the funds that build ports and power plants, deciding AI capacity is the next asset class. Meanwhile in Europe, AWS named its first launch partner for a European Sovereign Cloud, a hyperscaler conceding that ”the cloud” now needs a passport.

For enterprise buyers, this quietly rewrites procurement. A year ago ”where does it run” was a footnote. Now there are real regional and sovereign options multiplying on three continents, which means data-residency and resilience requirements you'll be handed in a future contract suddenly have answers you can actually buy.

Here's what works: Add a ”whose soil, whose financing” line to your AI infrastructure review. The vendors standing up sovereign and regional capacity now are the ones who'll clear the residency rules headed your way, and the buyer who maps them early negotiates from leverage.

6. An AI Just Passed The Radiology Board Exam

Here's the medical story that should have led every health desk. Harrison.ai released Harrison.Rad 1.5, a radiology foundation model that drafts reports from images, prior scans, and clinical context, and it's the only model to pass radiology's new board exam standard. Not a demo that reads one X-ray. A system that does the actual job a junior radiologist does, against the actual bar the profession sets.

What makes it credible is that it's not alone in the data this week. A peer-reviewed study found that large language model consensus substantially improves the reliability of clinical answers, while another grappled with how clinical reasoning gets learned when machines can pass the exams. The breakthrough and the hard questions arrived together, which is exactly how you know it's real and not a press release.

The ”so what” is bigger than radiology. When a model passes the same board exam a human must, the conversation shifts from ”can it” to ”who's accountable when it does.” A draft report is still a draft, and somebody licensed still signs it, but the economics of who reads the first pass just changed.

Here's what works: If your industry has a certification exam, assume an AI is now studying for it. Map which expert tasks are ”draft, then a human signs” versus ”human from scratch.” The first category is where synthetic expertise lands first, and where you should be piloting, not panicking.

7. Don't Let AI Bulldoze Your Dashboards Yet

Here's the contrarian take the hype needs. A sharp analyst argued that the rush to replace dashboards with AI is a mistake, and the logic is hard to dodge. A dashboard is a shared, stable, agreed-upon view of the truth. A chatbot gives every person a slightly different answer to a slightly different question, and calls it progress.

I've watched this movie before, just with different technology. Every few years something promises to kill the boring reliable thing, and the boring reliable thing turns out to be load-bearing. A DJ can improvise all night, but the crowd still needs the steady kick drum to know where the beat is. Rip out the dashboard and replace it with a clever conversation, and you've traded one shared reality for forty private ones.

The deeper risk is plausibility. A dashboard that's wrong looks wrong, the number is off, someone catches it. An AI that's wrong looks right, phrased confidently, sourced from data nobody can see. Replacing a transparent view with a persuasive one isn't an upgrade. It's moving the failure point somewhere harder to catch.

Here's what works: Don't retire your dashboards, layer AI on top of them. Keep the shared, auditable view as the source of truth, and let the chatbot explain and explore it. The win is a conversation grounded in one reality, not forty confident guesses.

Signal vs. Noise

🟢 Signal: The financing-and-compute layer. The money plumbing of AI, private credit deals, sovereign data centers, gigawatts of capacity, climbed hard in real influence on Wednesday while the model names cooled. The Apollo-Blackstone $35 billion and Rio's AI city are the same story: AI's physical and financial foundation is where the smart capital actually moved. Most coverage is still chasing model launches and missing where the money went.

🔴 Noise: The OpenAI IPO buzz. The IPO filing pulled some of the heaviest volume on the wires, but OpenAI's pull across the rest of the AI conversation actually slipped day over day. A filing is a liquidity event, not a capability shift. Anyone trading the headline as a product breakthrough is reading the press release, not the business.

From the 190K

We scanned 190,000 articles this week. Here's what no one's talking about:

Apollo and Blackstone moved to finance $35 billion of AI compute, Meta locked $27 billion with Blue Owl for a data center, and Rio pulled $550 million of infrastructure money into an AI city, all in the same window.

Each desk files these apart. The private-equity press covers the $35 billion. The Big Tech beat writes up Meta's data center. The Latin America desk notes Rio's deal. Read them on the same morning and one story appears: AI's build-out is no longer being funded out of tech giants' cash piles, it's being financed by the same infrastructure and private-credit funds that bankroll ports, toll roads, and power grids. The IPOs got the cameras. The off-balance-sheet financing got the steering wheel. The move on Wednesday is to ask whether your own AI vendors are funded like a utility or like a bet, because that's now the difference between a partner that survives a downturn and one that doesn't.

By The Numbers

• Apollo and Blackstone moved to finance $35 billion of AI computing capacity — The first tranche adds one gigawatt, enough to power roughly 750,000 homes, aimed entirely at AI. Private credit is now financing compute like it finances power plants.

• Meta struck a $27 billion financing deal with Blue Owl Capital — For its largest data center project, part of an arrangement targeting more than 20 gigawatts of compute through 2028. The AI build-out has gone off-balance-sheet.

• OpenAI is generating $2 billion in monthly revenue with 900 million weekly users — Growing roughly four times faster than the companies that defined the internet era, while targeting a valuation of up to $1 trillion and profitability only by 2030. Scale and burn at numbers no consumer business has carried.

• Revolut lined up a secondary sale at a $115 billion valuation — A top-ten-European-bank price tag for a company that started as a currency app. Fintech's growth story is now bank-sized, and so is its compliance bill.

• Rio's Elea Data Centers pulled a $550 million investment from I Squared Capital — Infrastructure money funding sovereign AI capacity in South America. The same fund class that builds airports now builds GPU farms.

• Microsoft's June Patch Tuesday addressed 198 CVEs — Nearly 200 security holes patched in a single month as the AI-era attack surface keeps widening. Every new capability is a new door, and someone has to keep counting the locks.

• See what's rising in our 190K-article corpus this week →

Deep Dive: Who's Actually Paying For The Party

Let me take you backstage, because that's where this week actually lived. The crowd buys a ticket to see the headliner. They never think about the promoter, the one who fronted the cash for the venue, the rig, the security, the insurance, weeks before a single light came on. No promoter, no festival. And this week, the AI industry's promoters finally stepped into frame.

The Stage Everyone Watched

The IPO filings got the photos. OpenAI and Anthropic both moving toward public markets, a $1 trillion number floating around, the kind of headline that sells the show. That's the headliner walking on stage to a roar. It's real, it matters, and it's also the part designed to be seen. The spotlight is pointed there on purpose.

The Check Behind The Booth

But the genuinely new money this week wasn't equity. It was credit. Apollo and Blackstone fronting $35 billion of compute. Meta and Blue Owl wiring $27 billion into a data center. An infrastructure fund putting $550 million into a Brazilian AI city. This is promoter money, patient, secured, structured like a toll road. When private credit starts treating GPUs the way it treats power plants, it's saying out loud that AI compute is now a utility you can lend against, not a gamble you bet on.

The Promoter's Edge

Here's the part the hype missed. The operators who win the next phase aren't the ones with the loudest model or the splashiest IPO. They're the ones whose financing can outlast a bad quarter. Equity hype evaporates when the mood turns. Infrastructure-grade credit, secured against real assets and predictable cash flows, does not. The festival doesn't survive on ticket buzz. It survives because the promoter structured the money to take a punch.

What Actually Works

1. Read the financing, not the filing: A vendor's funding structure predicts its survival better than its valuation. Utility-style credit outlasts equity hype.

2. Map your single-supplier risk: Know which AI workloads depend on one company, and whether that company is funded to last.

3. Treat compute like a utility: Plan for AI capacity the way you plan for power and bandwidth, sourced, contracted, with a second supplier in mind.

4. Write the exit before the entry: For any critical AI system, document how you'd leave it. A system you can't exit is one you don't govern.

The festival is louder than ever, and the headliner is genuinely good. But the show doesn't run on applause. It runs on the promoter who quietly structured the money to survive the rain. This week, the promoters stepped into the light. The only question is whether you're still watching the stage, or whether you've checked who's actually funding the party you're betting your roadmap on.

What's Coming

Compute Becomes A Credit Market

With Apollo and Blackstone financing $35 billion of AI compute and Meta's Blue Owl deal in the same window, expect ”AI infrastructure debt” to become a named asset class by year-end. The vendors funded like utilities will quietly separate from the ones funded like bets, and procurement teams will start asking to see the cap table.

Sovereign AI Cities Multiply

Rio's AI city partnership with Elea is a preview, not an exception. Expect more national and municipal governments to pair with infrastructure funds to build domestic AI capacity, turning ”where does this run” from a policy footnote into a standard procurement clause across more continents.

Certified AI Forces The Accountability Question

After a model passed radiology's board exam, expect every credentialed profession, from accounting to law, to face the same question: when the AI passes the same exam a human must, who signs, and who's liable. The regulation will chase the credential.

For Your Team

Strategic purpose: Thursday is the day this week's shift lands on the leadership table. The headlines were about IPOs and billion-dollar valuations. The real story was that AI's foundation, the compute, the financing, the off switch, just got named. Your edge is refusing to treat ”how our AI vendors are funded” as the finance team's problem when it just became strategy.

Thursday's meeting prompt: ”If our top AI vendor hit a bad quarter tomorrow, do we know whether their financing survives it, and do we have a documented way to exit their system without breaking our operation?”

The Foundation-First Framework:

1. Read the financing — Judge AI partners by funding structure, not valuation. Utility-style credit signals staying power; equity hype signals fragility.

2. Own your data leverage — Inventory the proprietary data only you hold. In the next wave of AI deals, the data owner sets the terms.

3. Write the exit plan — For every critical AI system, document how you'd switch vendors, export data, and run without it before you sign.

4. Layer, don't replace — Keep your shared, auditable views as the source of truth and put AI on top of them, not in place of them.

Share-worthy stat: Apollo and Blackstone moved to finance $35 billion of AI computing capacity, with the first slice alone adding enough power to run 750,000 homes. When the funds that build power plants start building GPU farms, AI compute stopped being a bet and became a utility.

Go deeper: Track where AI money and infrastructure are heading in real-time →

The Track of the Day

”Private-equity firms have become a vital source of funding for AI companies that face a shortage of costly and supply-constrained AI infrastructure needed to meet rising demand.”
— From the Apollo-Blackstone $35 billion financing report

That line reads like a footnote, but it's the whole week in one sentence. The $1 trillion IPO targets, the $115 billion fintech valuations, the board-exam-passing models, all of it sits on top of one unglamorous truth: someone has to finance the compute, and this week it stopped being the tech giants and started being the bankers who fund infrastructure. The headliner gets the photos. The promoter who structured the money gets the encore.

We scanned 190,000 articles this week so you don't have to. Data Pains → Business Gains.

Published: June 10, 2026 | Curated by Yves Mulkers @ Ins7ghts