BlackRock's AI Bankruptcy Playbook: Follow the Lender, Not the Builder

Larry Fink said something this week that most AI coverage treated as a warning. It wasn't.

At BlackRock's Infrastructure Summit, the CEO of the world's largest asset manager ($11.6 trillion under management) stated that the AI infrastructure race will produce "one or two" corporate bankruptcies among companies that are "third, fourth, and fifth" in the race. Companies that have "raised huge amounts of equity and debt that is now oozing through the financial system."

In the next breath, he argued that under-investment in AI is a bigger risk than over-investment. That if the US doesn't spend, China wins.

Most coverage framed this as contradictory. AI will produce bankruptcies, but we should spend more? How does that make sense?

It makes perfect sense once you understand who's talking.

The Position Reveals the Play

BlackRock doesn't build AI. BlackRock doesn't train models. BlackRock doesn't rent GPUs or hire ML engineers or ship products.

BlackRock lends money to the companies that do all of those things.

When Fink predicts bankruptcies, he's not warning the industry out of concern. He's telling his investors where the returns are. Distressed AI companies with real infrastructure assets (data centers, GPU clusters, fiber capacity) get acquired at steep discounts by creditors. BlackRock, as the lender, gets paid before equity holders in a bankruptcy. And if the company survives, BlackRock collects interest on the debt.

The "under-investment" argument serves the same function. More AI spending means more borrowing. More borrowing means more deal flow for BlackRock's infrastructure funds. Fink isn't contradicting himself. He's describing a business model where both outcomes (bankruptcy of non-leaders, continued investment by leaders) generate returns for the lender.

This is financial engineering 101. But most AI coverage doesn't speak finance, so the positioning goes unremarked.

The Capital Structure Nobody Talks About

The AI industry has a capital structure, and your position in it determines your risk.

At the top, you have lenders. BlackRock. Sovereign wealth funds. Infrastructure-focused private credit. These entities provide the debt that builds data centers and purchases GPU clusters. In a bankruptcy, they get paid first. They've already modeled the default probability and priced it into their loan terms. When Fink talks about bankruptcies, he's describing losses his team already underwrote.

In the middle, you have the mega-cap equity holders. Microsoft's investment in OpenAI. Google's internal AI spending. Anthropic's VC backers (Google, Amazon, Spark Capital). These companies are risking equity, which sits behind debt in the capital stack. If things go well, they capture upside. If things go badly, they absorb losses before the lenders lose a dollar.

At the bottom, you have the application-layer companies. Startups building on top of foundation models. Companies that have raised Series A or B rounds to build AI-powered products. In a downturn, they lose funding access, burn through runway, and either pivot or fold. Their equity holders (VCs) take the first and largest loss.

We noted the AI funding bifurcation on March 12, when February 2026 funding hit $189 billion with 90% going to AI. That analysis focused on the split between mega-cap infrastructure and application-layer startups. Fink's comments add the critical third dimension: where the debt is, and who holds it.

Which Companies Is Fink Talking About?

Fink specified "third, fourth, and fifth" in the race. He didn't name names. But the description, companies that have raised huge amounts of equity and debt for AI infrastructure and aren't in the top two, narrows the field.

The top two infrastructure spenders are clearly Microsoft (via OpenAI) and Google. Meta's $135B commitment is third by capex, but their Avocado delay (covered separately today) raises questions about whether that spending translates to frontier capability.

Below that, you have companies like Oracle (which just announced massive AI infrastructure expansion alongside 20-30K layoffs), CoreWeave (which went public to fund GPU clusters), and various data center REITs that have repositioned as "AI infrastructure plays."

These are the companies with "equity and debt oozing through the financial system." The ones Fink is both warning about and positioning to profit from.

The Pattern: Follow the Lender

This isn't unique to AI. The same capital structure dynamics played out in telecom (1999-2001), housing (2005-2008), and shale oil (2014-2016). In each case, the lenders predicted the bust, positioned for the bust, and profited from the bust, while simultaneously funding the boom.

The playbook is always the same:

1. Fund the boom through debt (infrastructure loans, credit facilities, project finance)
2. Predict the bust publicly (credibility positioning, regulatory cover)
3. Acquire distressed assets through the bankruptcy process (real infrastructure at discount prices)
4. Operate or sell the assets in the recovery

Fink's comments fit steps 1 and 2. BlackRock's infrastructure funds are actively deploying capital into AI data centers and compute facilities (step 1). Fink's public warnings create the record that BlackRock was prudent all along (step 2).

Steps 3 and 4 come later. But if you're building an AI company right now and wondering whether the capital markets will be friendly in 18 months, Fink just told you the answer. For the top two or three, yes. For everyone else, the lenders have already modeled your potential failure.

What This Means If You're Building

If you're a founder or technical leader at an AI company, Fink's comments should inform three things.

Your fundraising strategy. The window for raising AI infrastructure debt is closing for non-leaders. Lenders are getting more selective, not less, even as total capital flowing into AI increases. If you're not a clear top-3 player in your vertical, equity may be more available than debt, which means more dilution and less runway buffer.

Your vendor dependencies. If your critical AI infrastructure provider is in the "third, fourth, fifth" category, you have supply chain risk. Not just from the provider failing, but from the provider cutting corners as financial pressure mounts. Service quality degrades before companies go bankrupt.

Your exit timeline. The traditional VC model of "build, scale, exit in 7-10 years" faces a new constraint. If the AI infrastructure shakeout happens in the next 2-3 years (which Fink seems to think is plausible), application-layer companies need to either achieve profitability or find acquirers before the credit cycle turns. Being a growing-but-unprofitable AI startup when lenders are foreclosing on data centers is not a position you want to be in.

The Uncomfortable Question

Here's what nobody in AI wants to say out loud. If Fink is right that the race will produce bankruptcies among non-leaders, and Meta's Avocado delay shows that even a $135B budget doesn't guarantee frontier capability, then the amount of capital being destroyed in the current AI investment cycle is larger than the industry acknowledges.

Not all capital destruction is bad. Failed companies create available talent, discounted infrastructure, and proven negative results that help survivors. The internet bubble produced Amazon and Google from the wreckage of Pets.com and Webvan.

But the capital structure matters for who survives. Companies with manageable debt loads and diverse revenue survive downturns. Companies that levered up to build infrastructure for models that never reached competitive quality don't. The lenders always know this. That's why they charge interest.

Fink isn't warning you about the future of AI. He's telling you about the present, from the perspective of someone who has already placed his bets and is comfortable enough with the outcome to say it publicly.