Summary

Marblegate’s Andrew Milgram put numbers to the K-shaped economy using anonymized Rapid Ratings data on ~1,200 US middle-market companies ($100–750M enterprise value): EBITDA down 20–25% since 2019, mid-single-digit margins vs. mid-teens for the Russell 3000, and net profits after tax “consistently negative over the past 2 years.” His verdict: “as much as the infomercial that is CNBC wants to convince you that everything’s great in the economy — it’s just clearly not.”
The default wave is already in the data. In 2023 almost 25% of the middle-market data set couldn’t cover debt service — and 2024 business bankruptcies hit a 14-year high. The 2024 data shows another 20% failing coverage, so Milgram expects 2025 filings to be persistent and possibly higher. Most likely resolution is the “meander solution” — a slow, S&L-crisis-style workout rather than a Resolution Trust-style fix, for which Milgram says he doesn’t know that there’s even the beginning of the political will.
Tariffs split the K further: Milgram can “make a pretty strong argument that public companies will actually benefit from the tariff regime,” but anything above a 5–6% tariff “will crush the US middle market” via margin and debt-service destruction — middle-market firms supply the big public companies, which hold the pricing power on both ends.
Private credit’s reported default rates are a mirage. Per Fitch, ~82% of private credit sits at single-B-minus or lower, and triple-C equivalents historically default at ~30% cumulative over 3 years — yet managers report north of 1.5%. “Defaults are the most easily manipulated statistic in the world… don’t ask the default rate. Ask the waiver rate. Ask the amendment rate.” PIK debt — “payment isn’t coming” — runs 17–18% in some BDC portfolios: those are equity-risk books wearing lender clothes.
The New York taxi medallion trade is the template for his “full-contact distressed investing”: two years of research (including getting his own taxi driver’s license), a $160k/medallion stalking-horse auction that created a year-end mark far below lenders’ prior ~$350k estimates, then over $600M deployed into 4,000+ assets bought from banks and the NCUA — an operation Marblegate took public a few weeks ago. His generalizable rule: “it’s almost like investment malpractice to invest in a distressed asset not taking an active operational role.”
The employee retention tax credit trade shows why returns hide in nooks and crannies: a CARES Act program estimated at $50B had paid out $200B a year in, and Marblegate bought conservatively underwritten claims at 85–86 cents on the dollar with a 6–7% statutory interest tailwind and put-back protection — roughly a 12% minimum-return estimate against a US government counterparty. Why was it available? “All profits emanate from the variant view. If you have the market view, you get the market return.”
Big-picture warning: large portions of the investing world are on autopilot — CLOs relying primarily on diversification and over-collateralization as risk control amid uneven credit work, and the “productization of investment decisions” that turns allocators into general contractors paying fee-motivated agents. “At Marblegate we say there is no outsourcing of critical thinking.”

Deep dive

1. The middle market is hollowing out — and now there’s data

Milgram’s frame for the “K-shaped economy”: everyone senses that parts of the economy are booming while others carry “a nagging slowness” they can’t locate, because CNBC, Bloomberg, and the Journal show “green arrows in the ticker-tape economy.” The middle market — roughly a third of the US economy and historically north of two-thirds of all restructurings — doesn’t file public financials, so evidence about it has been anecdotal and self-referential.
Marblegate fixed that by taking anonymized counterparty-risk data from Rapid Ratings and winnowing it to ~1,200 US companies with $100–750M of enterprise value. The findings: middle-market EBITDA “deteriorates every year,” down 20–25% since 2019; margins are mid-single digits versus mid-teens for Russell 3000 public filers; and net profits after tax are down “almost 200% over the measurement period” — consistently negative for two years.
The load-bearing metric is interest coverage, because — quoting his first Wall Street boss — “nothing so focuses the mind like a coupon payment.” In the 2023 data almost 25% of companies couldn’t cover debt service; 2024 business bankruptcies duly hit a 14-year high. The 2024 data shows another 20% failing coverage, so Milgram expects 2025 filings to stay elevated or rise, with some larger companies now starting to crack too.

2. A corporate class system — and tariffs will make it worse

Why is this happening? Market power. Middle-market companies typically don’t serve the end consumer; they serve large public companies, which hold pricing power over customers and suppliers — “pushing costs, pushing financing down onto those middle-market companies while taking that margin.” Patrick’s label — “a corporate class system” — Milgram accepts flatly: “the rich are getting richer and the poor are getting poorer,” and the frustration is expressing itself in the political sphere.
On tariffs, the K splits again. Milgram “can make a pretty strong argument that public companies will actually benefit from the tariff regime” — and assumes the West Wing and Treasury Secretary have reached the same conclusion. But for the middle market, “anything above like a 5 or 6% tariff will have a devastating impact on margin” — persistent tariffs “will crush the US middle market.” The uncertainty alone is corrosive: “Christmas is canceled,” as someone told him — order books must be committed now, with no way to know which way policy breaks.
How it resolves: probably “with a bang” that is nonetheless slow — either years of persistent restructuring like the early 2000s or a credit contraction like the late-’80s S&L era, which “sort of feels like the world we’re heading into.” Milgram says he doesn’t know that there is even the beginning of the political will for a wholesale Resolution Trust-style debt restructuring, so he lands on the “meander solution”: work through it problem by problem over time.

3. Distressed investing, properly defined: capital where there’s no supply of it

The term has been “abused,” Milgram says. When he started, it meant buying a company’s debt and exercising rights and remedies under the credit agreement to force operational improvement. Today’s wider covenants change the game: by the time a company trips, “the business is just worse off generally and needs a much bigger operational reworking.” And the “wait for the market to puke and buy” version is a tough strategy — Patrick notes that every year a handful of sectors run at 2–3x the system’s average default rate.
Marblegate was born in 2008–09 when his partner Paul called from Bear Stearns: “all great distressed investment firms are born out of crises, and this one’s ours.” They deliberately targeted the middle market as the Oaktrees and Apollos scaled up alongside the LBO complex — “we oftentimes refer to the private equity business as our manufacturing division, because they will produce a certain amount of problems pretty consistently” — leaving the middle market “underinvested, under-prosecuted, under-analyzed.”
The sourcing machine is the US banking system, where the middle market still gets most of its capital: “we are the number one buyer of steak dinners in the middle market and in middle America.” And the operating principle: banks make decisions “for three reasons — regulatory, regulatory, and regulatory. People think of banks as economic actors; they’re not. They’re regulatory actors.”
Everything stays in-house — sourcing, financial restructuring, operational restructuring: “At Marblegate we say there is no outsourcing of critical thinking.” FTI-style firms are too big and too costly for a middle-market situation to bear.

4. Taxi medallions: “the worst idea I’ve ever heard” — until two years of data said otherwise

In 2016, at peak Uber ascension, Paul twice brought him bank loans against NYC taxi medallions; twice Milgram called it “the worst idea I’ve ever heard.” Then the research began — two years of it, garage to garage in Queens, before a dollar moved. Milgram got his own taxi driver’s license (“I still go out and drive… you have to stay connected to the market”) — the only native-born American in his licensing class.
The first insight inverted the consensus: Uber wasn’t taking riders from taxis — it was taking drivers. By subsidizing every ride, Uber pulled riders from buses, subways, and private cars while pulling drivers out of yellows, which stacked idle. Uber was also exploiting an information asymmetry, pushing non-cash costs onto drivers who made cash-based decisions and structurally under-earned. Medallions had peaked at $1.2M each; the average unpaid principal balance of a taxi medallion loan was $550k, and there were 13,587 medallions — billions in, fleet owners “on yachts… while the drivers were struggling to make ends meet.”
His opening question to everyone in the space: “Tell me who your customer is.” One hundred percent answered “the rider.” Wrong — “the driver pays me. My customer is the driver.” That misidentification explained the extractive, combative industry structure. A 70-year-old garage owner supplied the thesis line: “nobody’s reinvented the economics of driving a car yet — and until that happens, taxis remain the most durable cash flow in the system.”
The TLC handed over ride-level data nobody had ever requested in full (“What are you looking for?” “All of it.”) — terabytes that broke Excel. Successful drivers looked “schematic… intentional”; under-earners looked like “a Rorschach test.” Marblegate later guaranteed drivers $200/day to follow data-driven routes like the “NASCAR loop” down Broadway (left turns only): dozens of experiments, zero guarantee payouts — the data-driven driver always out-earned.

5. The trades: manufacture a mark, then buy into the market

With no medallion transactions happening — “one of the all-time great indicators of when something is going to really sharply move” — banks still wanted ~$350k per medallion. The break came when Citibank seized “taxi king” Gene Friedman’s collateral for auction. Milgram offered a $150k stalking-horse bid; the bank “won’t do a penny less than 160” — done. Marblegate won 48 outright medallions and, crucially, hand-delivered a year-end mark far below every lender’s estimate, right at the end of November.
“Surprise, surprise, come January” lenders got serious. Marblegate bought the largest available portfolio from a federally chartered bank — chosen deliberately, because for a big federally chartered bank this was a small position it could dump at a sharp discount, whereas credit unions (the biggest lenders, whose medallion haircuts had been “next to nothing”) were rendered essentially insolvent and seized by the NCUA. The federal government became the largest lender in the space — and Marblegate’s next counterparty.
The NCUA negotiation took a long time for one reason: every loan carried personal guarantees, and the agency wanted assurance Marblegate “would not be rapacious” toward borrowers who had staked their homes on a medallion. With 4,500 line items to service and 97% of servicers refusing to touch the space, Marblegate stood up its own ~30-person servicer. End state: over $600M deployed, 4,000+ assets, by far the largest participant in the market. His rating of the investment: “Excellent” — on a risk-adjusted basis, since “we don’t have the benefit of… saying, ‘I think this Google thing’s got legs.’”

6. Endgame: a public company, and why the medallion survives autonomous vehicles

A few weeks before this conversation, Marblegate took its entire taxi operation public — chosen over a private equity sale so a “durable and persistent cash flow” could be valued by the market, with room to consolidate an ecosystem that “had been too disaggregated… too many people taking a profit margin out of it.”
On autonomy, he rejects “the simpleton’s answer — oh, you can’t fight technology.” The medallion system exists because LaGuardia’s 1930s Haas Act had to un-clog Depression-era streets, and that congestion imperative accelerates in an autonomous world: Greenwich residents could send empty cars to work in Manhattan, and people “with fewer scruples” might refuse rides north of 125th Street. The city’s regulatory instrument is the medallion — plus a moral one: the city has invested heavily protecting individual owner-drivers, and displacing their capital “would obviate all the work and investment the city has done.”
His hedged synthesis: autonomous “will probably someday displace the driver,” but the driver can swap earnings power for capital contribution — the medallion becomes a capital asset that its owner contributes to whatever autonomous operator runs the system.
Stakeholder management was half the trade. On Risa Heller’s advice, Marblegate pre-briefed every regulator and council member who touched the space. When the Taxi Workers Alliance picketed Marblegate’s Greenwich office, he sent out water and sandwiches — then put on a baseball cap and marched with the picketers incognito to hear their grievances, which he found “completely valid and real.” The relationship today is “constructive, productive, and partnership-like.”

7. The ERTC trade: returns in nooks and crannies

The employee retention tax credit came out of the CARES Act with famously loose language — a 20% revenue decline or impact from any government order, “not a federal government order… state, local, anything.” A program the government reportedly expected to cost $50B had paid out $200B about a year in — “and that was a year ago.” Overburdened IRS, paper filings, glacial processing.
Marblegate bought claims at 85–86 cents on the dollar, collecting the statutory 6–7% interest owed while refunds waited, with a put-back right returning capital plus a rate if a claim was disallowed. Underwriting was deliberately stricter than the government’s — “we want to be Caesar’s wife in our underwriting” — passing on huge numbers of claims. Patrick characterized the result as roughly a 12% minimum return against a US government counterparty.
Why did sellers take the discount? The K-shaped economy again: most, if not all sellers were capital-constrained middle-market companies monetizing an asset, and Marblegate turned documents in 2–3 weeks. Why didn’t Apollo or Baupost do it? The process was manual and paper-heavy and took months to build; the philosophical answer: “all profits emanate from the variant view. If you have the market view, you get the market return.” Same instinct behind their earlier work restructuring Native American gaming assets on sovereign territory.

8. Negotiation rules and the human drama

His negotiation core: “it’s easy to know what you want — you look in the mirror and tell it to yourself every morning. The real effort has to be focused on understanding the other person” — their needs and hard constraints, which Marblegate commits to respecting once verified. No zero-sum deals, and when selling, “you have to leave something in for the next owner.” Boots on the ground are non-negotiable: managements tell him “you’re the first lender ever to show up and see the facility.”
The rest “you learned in kindergarten”: dignity, respect, honesty — “you don’t have to show all of your cards… but you want to deal with people on a heads-up basis.” And pace matters: “time kills deals.”
Paul’s line frames the psychology: “every single investment is both a complicated business problem and a human drama” — and the drama is the unknowable part walking in. Marblegate is “the avatar of people’s frustrations”; counterparties are living the hardest thing they’ll ever go through, often watching life savings vanish. His stance: “nobody finds distress — distress finds you… we’re the eat-your-vegetables guys. We’re not doing this because we have some personal animus to you.”

9. Credit markets on autopilot: CLO laziness and private credit’s hidden defaults

His most acute worry: “I sense a lot of laziness out there… large portions of the investing world are now on autopilot” — most acutely in CLOs, the primary corporate-credit creation vehicle of the past 10–15 years, which rely primarily on diversification and over-collateralization as risk control, amid uneven credit work. He’s careful not to disparage the whole industry — some managers are unbelievably good — but the “productization of investment decisions” turns investors into general contractors outsourcing thinking to agents “motivated by a stream of fees rather than an investment outcome.”
The private credit arithmetic doesn’t close. Post-2011 leveraged-lending guidance pushed levered credit out of banks (“he who wears the risk makes the rules”), and per Fitch, ~82% of private credit sits at single-B-minus and lower. Forty years of data says triple-C equivalents default at ~30% cumulative over three years — yet managers report north of a 1.5% default rate. Either someone “invented a new way to underwrite credit which avoids all losses,” or “defaults are the most easily manipulated statistic in the world — a default doesn’t exist unless I, the lender, call it.”
His diligence prescription: “don’t ask the default rate. Ask the waiver rate. Ask the amendment rate.” And watch PIK — “we oftentimes say PIK means payment isn’t coming.” Some BDC portfolios carry 17–18% PIK: “they’re no longer lenders at that point,” they’re taking equity risk in exactly the pressured middle-market companies his data set flags.
The tiering is structural: private credit “used to be a direct origination business” but is now “largely a brokered market, which is a kind of dirty little secret.” Houlihans and Lincolns show paper to Ares and Golub first, then down the cadres — so the best-known firms genuinely do hold the best portfolios, and the tail holds what everyone else passed on.

10. Manufacturing division, the future of asset management, and the boy from Beaumont

On private equity: “I love those guys. That’s my manufacturing division.” The firms he respects do “scratch and dent” value investing — unloved assets with real force applied. The rest — “great dealmakers who… don’t actually do anything other than buy the company and show up for board meetings” — “are troubled… and probably don’t have much of a future.” Same test he applies to himself: a purely financial investor is really just a traitor.
Asset management must eat its own cooking: “acquire assets that are troubled, reimagine what they could and should be, and then apply force” — applied to the industry itself. He buys the argument that retail is under-allocated to privates but warns “we’re going to go bump in the night trying to figure out where those lines exist”; interval funds have strengths and weaknesses; insurance and annuity-driven investing is “super interesting.” The golf-and-expensive-lunches model “is probably not going to be the successful model going forward.”
The origin story explains the style. His father — an immigrant born in Eretz Israel, raised in Latin America, who built a precast concrete business in Beaumont, Texas after visa restrictions on Jewish immigration first sent him back to Venezuela — died of a heart attack on a Boy Scout campout when Milgram was 11. They had read the Journal’s stock pages together daily. Patrick’s observation: the most common pattern among his mentors is losing a father young, followed by “this tremendous amount of agency.”
The close ties back to the K: “America is the greatest system that has ever existed. But countries, like companies, are delicate… subject to abuse.” A system calcified so “the haves will perpetually have and the have-nots will perpetually not have” breaks the magic that let his father’s son end up in Greenwich. The kindest thing anyone did for him: after his father died, a family took him in for breakfast every morning and drove him to school — “they leaned in.”

Arvind Jain on building Glean and the future of enterprise AI Market Overview August 5, 2025