Not For Credit

THE NUMBER: $11B → $1.28B. That’s Airtable, from its 2021 peak valuation to the price Bending Spoons paid this week for the equity. Read it the way an investor reads it, because it’s not a blowup, and that distinction is the whole issue. Airtable still does roughly $480 million in revenue, still grows 20%, still counts 80% of the Fortune 100 as customers, and still had about a billion in cash — enough to run for years without asking anyone for a dime. Nobody put a gun to their head. The business didn’t break. What broke was the shape of the bet. An equity that used to trade like a call option — unlimited upside, the entire reason you hold the thing — got repriced like a bond, capped and cash-and-carry, the moment growth stopped accelerating. The downside was always zero. The ceiling is what got marked down. Hold that thought, because two very different men spent this week telling you the same thing about it.

📉 The Signal: Not For Credit

John Zito runs credit at Apollo. A quarter-trillion dollars of other people’s money, 25 years on the desk, one of the most respected fixed-income minds in the business. Somebody asked him at a conference for his top-of-house view — spreads, the wars, inflation, the usual menu a credit guy is supposed to have opinions about. He waved all of it off. “I spend very little time thinking about most of the things you just brought up. I’m not joking.” Only one thing matters, he said: whether what’s happening inside the AI labs is real. And if it is real, “it’s so hyper-deflationary to so many things over the long term that it’s really hard to take risk.” He’s been managing money for almost 25 years, and he called it as hard an environment to handicap the next 12 to 24 months “as it’s been in a really long time.” Then the four words that a credit lifer earns the right to say, and that ought to stop you cold: “That’s not for credit.”

Kieran Goodwin flagged it, and Kieran would know. Full disclosure — we worked the same shop years back, so I’m not a neutral party here. He’s one of the sharpest credit traders alive, a natural cynic the way every great distressed investor is, and he made his name in 2008 buying insurance on debt that everyone else swore was money-good, right up until it wasn’t. His translation of Zito is the whole newsletter, so I’ll give it to you verbatim: “Risk/reward in equity in fat-tailed scenarios can be very good. Risk/reward in credit in fat-tailed scenarios can be awful.” Somebody in his replies compressed it even further, and it’s the single best sentence I read all week: “Credit markets hate binary outcomes even if very +EV.” Even if the odds are in your favor. Sit with that, because it’s the hinge of everything below.

🎢 Floor Seven and a Half

There’s a floor in Being John Malkovich — 7½, wedged between seven and eight — where the ceilings run about four feet high and everyone conducts their business hunched over, shuffling around bent double, apologizing for their posture. That’s the image up top, and it’s the best picture of a bond I know. Here’s why a credit trader’s aside belongs on your desk and not just his. A bond and a stock share the exact same floor: zero. Both can go to nothing. What you actually pay for when you buy the stock instead of the bond is the ceiling — the headroom. Lend a company money, buy its bond at 80, and you’ve stepped onto Floor 7½: the very best thing that can happen is you stand up as straight as par will let you, which isn’t straight at all, and collect a coupon for your trouble. That’s it. You’ve capped your own upside at a modest, knowable number in exchange for standing in front of the equity in a wipeout. Buy the same company’s stock and your downside is identical — zero — but your upside is the moon. Same floor. Wildly different ceiling.

That asymmetry has a name — convexity — and it is the only reason to own equity in the first place. You’re not paying for the floor; you and the bondholder share the floor. You’re paying for the right tail, the fat, glorious, unbounded upside that the bond contractually cannot give you. Which is why Goodwin’s line is so sharp. In a genuinely fat-tailed world — one where the range of outcomes 24 months out is enormous and unknowable, exactly the world Zito is describing — that fat tail is a gift to the equity holder and a loaded gun to the lender. If AI is real and hyper-deflationary, the equity in the winners goes to the moon and the credit gets repaid at par, having eaten all the uncertainty for a coupon. Heads, the stockholder wins big and you win small. Tails, you both lose, but you’re the one who was supposed to be safe.

Now watch what AI is doing to that ceiling, because it’s cruel. The narrative machine is pumping the imagined ceiling full of helium — unlimited-upside stories, every startup a rocket — while the actual ceiling gets sawed off in real time by price wars and free open-weight models. OpenAI cut its base model 80% per token this month. March’s full flagship intelligence now sells for about a thirteenth of what it cost four months ago. The thing you’re paying an equity multiple for is the ceiling, and the ceiling is the one thing nobody in this market can price anymore.

🪙 The Airtable Autopsy

Which brings us back to the number. Airtable is what it looks like when the ceiling comes down on a real company in public.

The arc is a period piece. Founded 2013, raised over $1.4 billion across its life, crowned at north of $11 billion in the 2021 mania. Earlier this year the secondary market was already marking it around $4 billion — the first tell. This week Bending Spoons, the Italian outfit that IPO’d in July at an $18 billion valuation and has made a business of buying Evernote, WeTransfer, Eventbrite and Vimeo on the cheap and running them for cash, agreed to pay $1.28 billion for the equity. With Airtable’s own cash the enterprise value lands around $2.25 billion. From eleven billion to a couple billion, on a company that is, by every operating measure, fine.

SaaStr’s Jason Lemkin wrote the eulogy, and his one-liner was “Grow or Die. That’s 2026.” Twenty percent growth at $500 million of revenue, not accelerating, gets you a 2.5-to-4x revenue multiple and a phone call from Bending Spoons. I’d put it one notch sharper, and I said so back to him: the day growth stops accelerating is the day the lever that pays equity holders quietly flips from growth to margin. It’s not “grow or die.” It’s “grow, or throw off cash — and God help you if you do neither.” That same $480 million of revenue at a 40% free-cash-flow margin is a $4 billion company. At 20% it’s a $2 billion company. The upside didn’t vanish when growth slowed. It moved, from the growth line to the margin line, and Airtable’s problem is that it was optimized for a growth story it could no longer tell.

The smartest people in Lemkin’s replies landed exactly there. Karim Ahmad: “This is not an arr trade, but an ebitda one… This should be a 30/40% margin business and a 20x ebitda.” That’s the whole game — Bending Spoons isn’t buying growth, it’s buying the fat it can cut into margin, and Lemkin himself spelled out the playbook: “Cut 80% of employees, double pricing, leverage the fact most can’t leave due to the data moat.” And Amplitude’s Spenser Skates said the quiet part: “A lot of VCs are still pretending their private marks on SaaS companies are still relevant. News flash, they aren’t.” The $11 billion was never real in the sense that mattered. It was a mark. The $1.28 billion is a price. The gap between a mark and a price is where a decade of venture returns quietly went to die this cycle — and note who got paid: the last money in gets its 1x back off the top of the preference stack, and the common holders, the employees, eat the difference. Somebody in the thread put it right: founders need to normalize raising common instead of preferred, because the preference stack is exactly what turns “we sold for $1.28 billion” into “the early team got a nice dinner.”

🏦 Why Apollo Lends Anyway

Here’s where it would be easy to score a cheap point, and I want to resist it, because the truth is sharper than the gotcha. The same Apollo whose credit boss says AI is “not for credit” is, right now, helping wire together one of the largest financing structures ever built — Google’s roughly $150 billion web of contracts to keep Anthropic supplied with chips. Apollo and Blackstone are supplying much of the private credit. The first tranche alone was a $35 billion special-purpose vehicle, the largest private-credit deal on record, priced around 5.75%. So does Zito not believe his own words?

He believes them completely. That deal is the exception that proves the rule, and understanding why is worth more than any headline. Apollo will happily syndicate $35 billion into AI — but look at what’s wrapped around that loan. It’s secured by the chips themselves, hardware worth most of the money. It sits under Anthropic’s roughly trillion-dollar equity valuation, an enormous cushion that eats losses long before the debt is touched. And critically, Broadcom stands behind a deficiency guarantee covering about $31 billion of the senior debt — meaning if Anthropic stops paying and the chips resell for less than the loan, Broadcom eats the shortfall. That single guarantee is what dragged unrated-startup risk up into investment-grade territory, the only reason a pension fund or an insurer can touch the paper at all. Enormous margin of safety, senior secured position, a fortress of collateral and third-party guarantees — and Apollo collects arrangement fees that, on a deal this size, plausibly exceed what it puts in as principal. That is not a bet on AI. That is credit doing the one thing credit can still do in a fat-tailed world: lend only where the downside is bricked up on all four sides, and get paid a fee to do it.

Which is exactly Zito’s point, not a contradiction of it. A reply under Goodwin’s post caught the apparent irony — “Then why is Apollo arranging $35bn tpu financings?” — and the answer is the lesson: because that structure has the margin of safety, and the rest of the market does not. Vendor financing at this scale isn’t new, either. Lucent and Nortel ran the same play for telecom gear in the late ’90s, lending customers the money to buy their own equipment, and it ended in tears when the customers folded. S&P has already downgraded Broadcom over the leverage this thing implies. The structure is elegant. It is not free. And it only works because someone bothered to build the walls.

🃏 The Cynic and the Optimist

Step back and you’ve got two temperaments staring at the same board, which is the real story of the week. The credit cynic — Zito, Goodwin, the distressed traders who got rich in ’08 being right about a tail nobody else would price — looks at AI and sees a fat tail that can kill him, so he demands a fortress or he walks. The venture optimist — Lemkin, the founders, the growth investors — looks at the same tail and sees the whole reason to play, because the fat tail is where the returns live. Neither is wrong. They’re just standing at opposite ends of the same instrument, one paying for the floor, the other paying for the ceiling.

And that hands you, whoever you are, a clean rule for the year. If you like the AI trade, express it in equity, where the convex upside actually pays you for the uncertainty you’re eating. If you hate the AI trade, express that in equity too — short the thing, buy the puts, where a fat left tail pays off. The one seat nobody should be sitting in is the lender’s, at least not without Apollo’s walls: you wear the entire downside and cap your best case at getting your own money back, which is a terrible trade when the range of outcomes is this wide. The bonds simply don’t pay you enough for a tail this fat.

🧭 We’ve Been Marking This to Market

This isn’t a new drum for us, and the through-line matters. On July 30, in On Tilt, we said the smart money was going to harvest the overhang, not chase it. On July 31, in Dumb Money, we watched Leopold Aschenbrenner be dead right about the AI buildout and blow up anyway, because being right and staying solvent are different skills — the leverage killed him, not the thesis. Today is the same lesson from the collateral side: the reason leverage kills you in this market is that the tail is too fat for debt to survive, which is precisely what Zito is telling his own desk.

There’s a second thread worth pulling, and it runs back through yesterday. When Goodwin posted, I replied that the deflation is already here for the tasks you can easily audit — math, code — and just beginning for the ones that ride on human judgment. That’s not a change of subject from our Terminator piece yesterday; it’s the same line drawn on a different map. The reason Zito can’t handicap the next 24 months is the reason we keep hammering: intelligence fell to nearly free, and judgment — knowing whether the confident thing the machine just did is actually right — didn’t fall at all. When judgment is the scarce input and the outcome range is enormous, you are, by definition, in a fat-tailed world. And a fat-tailed world, as the best credit mind on the tape just told you, is not for credit.

What This Means For You

Audit every dollar of CAPEX and every line of your strategy: offense or defense? Go down the list and ask, honestly, whether each spend funds a growth bet that can actually accelerate the top line, or a defensive play meant to protect revenue you already have. If it’s offense and it can compound, great — that’s you paying for your own convex upside, keep spending. If it’s defense, then stop pretending and cut costs aggressively to turn it into margin, because defense only pays equity holders through the profit line. And if you’re honest and find you’re doing neither — not accelerating growth, not expanding margin — then you already know what you are. You’re not a stock anymore. You’re a bond in decline, and the market will figure it out before you do.

Shorten the term on anything AI is making cheaper. Zito can’t handicap the next 12 to 24 months, and with respect, neither can you. So stop signing multi-year commitments at today’s prices for anything AI is actively deflating — models, tooling, seats, compute. Keep the contracts short and the optionality on your side of the table. The vendor wants your signature locked in for three years precisely because they can see the same ceiling coming down that you can. Match your contract duration to your actual visibility, and your visibility right now is short.

Express your AI view in equity, not credit. This is the portable version of Zito’s whole point, and it works whether you run a book or a business. If you love a bet, own the upside outright and let it run. If you hate it, find the way to be short that lets the downside pay you. The trap — the one seat that’s wrong in a fat-tailed market — is the lender’s chair, where you swallow the full loss in the bad case and get handed par in the good one. Any position where your best outcome is “I got my money back” is a bond, and this is not the market to be holding bonds you didn’t bulletproof first.

Three Questions We Think You Should Be Asking Yourself

Which of my bets are convex, and which can only ever get me back to par? Go position by position — vendors, builds, hires, your own equity — and sort them by shape, not by story. The convex ones are worth paying up for. The ones capped at “I broke even” are bonds, and you should be pricing them, and defending them, like bonds.

Where has growth quietly stopped accelerating, and have I switched the scoreboard? Airtable didn’t get repriced when growth fell. It got repriced when growth stopped rising. The day your second derivative goes negative is the day your job changes from chasing growth to manufacturing margin. If nobody on your team owns watching that exact inflection, you’ll manage for the wrong number a full year too long.

If I’m honest, am I offense, defense, or neither? Offense that compounds is a stock. Defense that throws off cash is at least an honest bond. Neither — spending to stand still, no growth and no margin — is a bond in decline, and it’s the one answer that should ruin your afternoon, because it’s the only one the market punishes without mercy.

The best credit investor on the tape looked at this whole beautiful, terrifying, unhandicappable thing and gave you two words of advice: not credit. If you like it, buy it. If you hate it, short it. Just, for God’s sake, don’t lend to it.

It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so.”
— The Big Short (2015), on the last time the smart money was certain about credit

— Harry and Anthony

Signal/Noise by CO/AI is published most weeknights from New Canaan, Connecticut. The point is to make you the smartest person in the room without taking more than fifteen minutes of your morning. If we did, forward it to one person. If we didn’t, hit reply and tell us why.

Sources

• Kieran Goodwin on X — Aug 4, 2026 (“Risk/reward in equity in fat-tailed scenarios can be very good. Risk/reward in credit in fat-tailed scenarios can be awful.”)
• Tropical Value on X — Aug 4, 2026 (John Zito, Apollo, at the Morgan Stanley US Financials Conference, June 10, 2026: “That’s not for credit”)
• Jason Lemkin on X — Aug 4, 2026 (“Grow or Die. That’s 2026.” 20% at $500M ARR = 2.5–4x ARR) + Harry DeMott reply (the growth-to-margin switch; $500M at 40% FCF = $4B) + thread: Karim Ahmad (“an ebitda trade… 30/40% margin, 20x”), Spenser Skates (“private marks… news flash, they aren’t”), the preferred-vs-common point
• TechCrunch — Bending Spoons to buy Airtable for $1.28B — Aug 4, 2026 ($11B peak, ~$4B secondary earlier this year, ~$480M ARR +20%, ~$2.25B EV with cash, 80% of Fortune 100)
• Capacity — How Google’s $150bn Anthropic chip deal reshapes AI finance — Aug 4, 2026 ($35B first SPV at ~5.75%; Broadcom ~$31B deficiency guarantee; Apollo/Blackstone credit; Alphabet owns ~14% of Anthropic; S&P downgrade; Lucent/Nortel precedent)
• Tom’s Hardware — AI companies are now racing to the bottom — Aug 3, 2026 (frontier flagship intelligence now ~1/13th its March token price)
• Ed Zitron — The AI Demand Bubble — Aug 4, 2026 (cloud growth propped by two unprofitable labs)
• Bloomberg / AdvisorHub — Family Offices Sidestep AI Fears in Deal Spree for Robotics Bets — Aug 4, 2026 (~$628B of patient capital buying convexity)
• CO/AI prior issues this builds on: On Tilt (Jul 30 — harvest the overhang), Dumb Money (Jul 31 — right thesis, wrong leverage), The Terminator (Aug 3 — judgment is the scarce input)
• The Big Short (2015) — the last time the smart money knew credit for sure