$55 Billion Fortress Warns Private Credit Against AI Data Center FOMO
Capital piling into a hot category before anyone has loss data to price it is the oldest underwriting trap in alternative lending, and this week the warning about it came from inside private credit's own top tier. Speaking at the Milken Institute's Global Dialogues Toronto event on September 14, Fortress Investment Group co-CEO Jack Neumark told private credit investors to stop chasing AI data center lending out of fear of missing out. 1 2 His test is the one that transfers to any funding desk: is the return actually paying for the risk being taken on, or is it priced as if the growth never stops. 1 He is not calling the AI buildout a bad bet. He is calling the current pricing of that bet a bad trade for lenders specifically.
The warning: private credit is taking technology risk in AI data center debt without adequate compensation or a practical way out, and the person saying so runs one of the largest lenders in the category. 1
The setting: Toronto's Milken summit, where attendees described the mood as split between exuberance and unease about AI, not a uniformly bullish room. 5
Watch: whether AI-infrastructure spreads and covenants keep compressing faster than any default data could justify.
What Alternative Business Lenders Need to Know
What exactly did Neumark say, and why does it matter coming from Fortress?
Neumark's warning is narrow and specific, not a broad AI bear call. Credit investors who lend into AI infrastructure "get a largely fixed return if the technology takes off, while still bearing the risk that asset values can fall, potentially trapping them in an illiquid investment if things go wrong." 1 He is warning lenders "against taking technology risks without adequate compensation or a practical way to exit an investment." 1 The source matters here. Fortress is not an outside critic; it is one of the largest active players in private credit, with roughly $55 billion under management, and co-CEO status gives Neumark visibility into how his own peers are pricing AI-infrastructure deals right now. 1 4 3 A warning from inside the trade carries different weight than a warning from outside it.
What is the fixed-return, full-downside asymmetry, in plain terms?
Strip away the AI framing and the mechanics are the mechanics of any secured credit position against a fast-appreciating, thinly tested asset. The lender's upside is capped at the coupon. The downside is not capped at all: if the collateral loses value because the compute demand it was built for does not show up at the assumed pace, the lender is still owed the full facility and holds collateral worth less than the loan. Illiquidity compounds it, because private credit facilities are not marked and traded the way public equity is, so a lender who wants out before maturity may have nowhere to sell. 1 Equity in the same building takes the opposite shape: unlimited upside if the buildout works, and a position that can be repriced or sold if it does not. None of this is exotic. What Neumark is flagging is that the current AI-infrastructure lending rush is pricing facilities as if that shape did not exist, competing on speed and access rather than on spread and covenant strength.
Price the illiquidity and the technology risk explicitly. Sector growth is not a credit positive; it is the reason the loss data does not exist yet.
How does this apply to alt-lenders who have no direct AI data center exposure?
Almost none of Beyond Banks' core readers are writing checks into hyperscale data centers, but the same file shows up at a smaller scale. Equipment finance on AI-adjacent gear, the power systems, cooling units and networking hardware feeding a data center buildout, is a fixed rental stream against an asset whose resale value depends on the AI tenant still needing it. Revenue-based financing and MCA advances to AI-services borrowers are a fixed remittance against a revenue line that exists because of a demand wave, not because of the business that stood before it. In both cases the lender's return is set on day one and the collateral or the cash flow is the part that moves. The specific asset changes. The question Neumark put to his own industry does not.
Is the market actually listening, or is this one voice against the tide?
Mostly the tide. Milken's Toronto event, Global Dialogues Toronto, was explicitly framed by attendees as split between "exuberance and unease" about AI, meaning Neumark's caution is a real minority position inside a still-bullish room, not a consensus call private credit has already absorbed. 5 Other private-credit managers are continuing to expand into AI-infrastructure lending, and a separate line of coverage this month on how lenders are pricing risk premiums for AI data center debt against rising CDS spreads on lower-rated hyperscalers points to the same underlying tension from a different angle, one Beyond Banks could not independently verify this week after the source article sat behind a paywall on live access. Treat Neumark's remarks as an early, credible flag from inside the asset class, not as proof the market has already repriced.
What should lenders watch for or do differently?
Take the file most readers actually see: an advance or RBF deal to a business selling AI services whose deposits have doubled in six months. Neumark's test, applied at that desk, looks like this. Size the advance on the revenue that existed before the wave, not on the trailing 90 days, and treat the spike as upside the borrower keeps rather than capacity the lender funds. Check customer concentration: if the top accounts are themselves AI-funded startups, the borrower's demand is a second-order bet on the same capital flow Neumark is describing, and the term should shorten or the advance rate should drop to match. Name the exit before funding: if demand halves, does the pre-wave business still cover the remittance, and if not, what is the lender holding. Equipment finance on cooling, power or networking gear feeding a data center runs the same drill on residual value; who buys that asset if the tenant walks, and at what discount, is the number the residual assumption has to survive. For anyone with direct AI-infrastructure exposure, the market signal to watch is whether spreads and covenants keep compressing faster than any default data could justify, and the standard Neumark set for his own industry was adequate compensation for the technology risk plus a practical way out. 1
What this week's record does not yet show: whether Fortress itself holds AI-infrastructure debt at any specific scale, and whether the paywalled coverage of AI data center risk premiums corroborates Neumark's read from the pricing side. The sources carry the facts; the connections between them, where stated as ours, are ours.
Our Opinion
The useful thing about Neumark's warning is who said it, not what sector it was about. Fortress is not a hedge fund taking a contrarian short; it is one of the largest private-credit managers in the business, with every incentive to keep originating into a growing category, and Neumark chose a public conference stage to tell peers to slow down anyway. 1 Alternative lenders have watched the same speech get ignored in their own markets before, usually right up until the first wave of defaults made the pricing mistake visible to everyone at once.
For MCA, factoring and equipment-finance operators, the lesson is not to avoid AI-exposed borrowers. It is to underwrite them as what they are: a fixed return against a demand curve nobody has loss data on yet. Ask the question once per file, before pricing, and let the answer set the advance rate, the term and the exit. That question does not need an AI angle to matter. It needs to be asked every time a hot category shows up at the desk.
1-Minute Video: Catch Synthetic EIN Fraud Before You Fund It
A verified Secretary of State record only proves the entity exists, not that the applicant owns it
Synthetic fraud rings pair a real EIN scraped from a public filing with a shell business name, a mismatch invisible to a Secretary of State check alone.
Cobalt Intelligence connects directly to the IRS TIN matching system: submit a business name and EIN and get back a match code in seconds. Code zero clears the file, code one means the EIN was never issued, and code two means the EIN is real but the name on file does not match.
Free Tools for Lending Teams
Headlines You Don’t Want to Miss
On September 11, four banking agencies jointly proposed rescinding the 2023 interagency third-party risk management guidance and the July 2024 post-Synapse joint statement, replacing uniform check-the-box requirements with a risk-tiered approach; published in the Federal Register September 15 as document 2026-18859, docket numbers OCC-2026-0793, OP-1881, RIN 3064-ZA58 and NCUA-2026-1684, comments due November 16. 6 8 7 The proposal includes a standard-setting-organization provision that could let banks satisfy part of their diligence obligation through shared certification bodies rather than duplicating due diligence per fintech partner, a real cost and speed lever for bank-partnered MCA and factoring platforms once finalized. 6
Northwind Group closed a $208 million construction loan for 141 Willoughby Street in Brooklyn on September 11 and a $219 million construction loan for 100 Wall Street in Manhattan on September 14, its eighth NYC office-to-residential conversion financing this year. 10 9 11 Northwind is also the lender behind the earlier $135 million acquisition loan for 235 East 42nd Street, the former Pfizer headquarters where two steel columns buckled in July after reinforcing plates were left off during construction of an 11-story rooftop addition, a documented construction-quality failure that Parkview Financial's CEO says will push lending rates up 50 to 100 basis points industry-wide. 12 13 No outlet has connected the two directly; that reading is ours, and it is a useful precedent for any alt-lender pricing adaptive-reuse collateral.
An ALTA survey of 245 title professionals found 59 percent of firms hit at least one seller-impersonation attempt in the past year, up from 28 percent in 2024, and 58 percent now encounter deepfaked images or voices used to impersonate a legitimate seller, a tactic that barely existed two years ago. 14 15 The mortgage-specific survey has a direct read for MCA and equipment-finance underwriting: any lender relying on video or voice identity verification for counterparty or collateral checks needs deepfake-aware liveness detection, not just document-based fraud screening, and ALTA's red-flag list (reluctance to meet, mail-away signing demands, all-cash or lien-free deals) is a transferable checklist beyond mortgage closings. 14 One limit on the read: the survey counts attempts, not completed frauds, so the two-year rise measures pressure on the front door, not losses.
Sources
1 Bloomberg | Fortress Warns Private Credit Against AI FOMO at Milken's Canada Summit
2 MarketScreener / MT Newswires | Fortress CEO Warns Against FOMO in AI Data Center Lending
3 Milken Institute | Global Dialogues Toronto announcement, confirming Jack Neumark, Co-Chief Executive Officer, Fortress Investment Group
4 Fortress Investment Group | Jack Neumark, team page
5 The Logic | Milken Institute adds Australia's Future Fund, Microsoft and Cohere executives to Toronto conference
6 Fintech Business Weekly | Regulators Move To Rescind Post-Synapse Risk Guidance
7 GRC Report | Federal Regulators Propose Risk-Based Overhaul of Third-Party Management Guidance
8 Federal Register | Proposed Third-Party Risk Management Guidance (Doc. 2026-18859)
9 PR Newswire | Northwind Group Provides a $219 Million Construction Loan for 100 Wall Street
10 PR Newswire | Northwind Group Provides a $208 Million Construction Loan for 141 Willoughby Street
11 The Real Deal | Another office-to-resi conversion lands nine-figure construction loan
12 PR Newswire | Northwind Group Provides a $135 Million Acquisition Loan for 235 East 42nd Street
13 Bisnow | Conversions Likely Face Higher Costs, More Scrutiny After Former Pfizer HQ's Near Collapse
14 National Mortgage News | Rising Seller Impersonation Fraud Risk a Worry for Lenders
15 American Land Title Association | ALTA (survey sponsor)
