In the fall of 2000, a sheriff’s deputy put a padlock on the door of our office. He was acting for a creditor, and we couldn’t open the company until that creditor was paid.
The creditor was an advertising network. Eighteen months earlier, signing that contract had been the obvious thing to do.
Let me be careful about what this essay is and isn’t. I’m not predicting that the AI boom ends, and I’m certainly not predicting when. I’ve now operated through four downturns — the dot-com bust, the financial crisis, Covid, and the post-Covid inventory reckoning — and I failed to see every one of them coming. I have no special sight. What I do have is a reasonably clear picture of what the first two quarters afterward look like, because they look remarkably similar every time.
And the thing that determines who survives is almost never the thing people expect. It is not how good the product is. It is not how smart the team is. It’s what you owe that you can’t unwind, how fast you move once you know, and whether you can reach break-even without anyone’s permission.
Three companies, three cycles.
Creditland, 2000
I co-founded Creditland in 1998. (No relation to the site trading under a similar name today.) We built the first online marketplace to deliver real-time credit card approvals across multiple issuers — Bank of America, MBNA, First USA, and a set of auto lenders alongside them. Getting a genuine approval decision back from several independent issuers in seconds, over the internet, in 1998, was a hard problem and we solved it. We raised $35 million across three rounds: seed, Series A and Series B.
In the first quarter of 2000 we had a term sheet on the table for our Series C.
Then the market collapsed, and the investor pulled it.
Not renegotiated. Not repriced. Pulled. This is the part that people who have only operated in good markets don’t understand about bad ones: capital doesn’t become expensive in an orderly way that lets you adjust.
It goes from available to unavailable between signature and close, and the gap is measured in days.
We did some things right. We turned a ten-year office lease into a one-year extension, which in that market was close to a miracle. We cut costs hard and fast.
What we couldn’t get out of were the advertising commitments. Cars.com. Yahoo. Others. Multi-year, take-or-pay, signed in 1999 when distribution was the scarce input and everyone believed it would stay scarce forever. We had bought eyeballs the way you buy a factory, on the assumption that we’d need them for years.
When the market turned, the vendors we owed money to were facing their own collapse, and they pursued what they were owed with real aggression. Hence the deputy, and the padlock.
We reduced costs as far as we could and couldn’t reach profitability without another round. There was no round to be had. We folded the company in late 2000 and let around a hundred people go.
Here’s the part worth sitting with. Creditland didn’t die because the product was wrong. The product was right, and a version of it is a large business today. It died because in 1999 we signed obligations that assumed 1999 would continue, and in 2000 those obligations had first claim on cash we no longer had.
Boundless, 2008
I co-founded Boundless Network in 2004. By 2008 we were building a platform for managing promotional product spend — a business whose customers were corporate marketing departments, which is to say a business positioned directly in front of the first budget line anyone cuts in a recession.
When the crisis hit, three things were true that had not been true at Creditland.
We had money in the bank. We hadn’t been running a strategy that required the next round to exist.
We were already lean, because we’d never had enough capital to become otherwise. There was no accumulated slack to defend.
And we moved in weeks, not quarters. We took 20 to 25 percent out of the cost base. We eliminated positions. The executive team took pay cuts, starting with me.
My CTO and I made a deal that neither of us would shave until the company was at break-even. It took about six months. We both ended up with serious beards. I mention it because the shape of that period is the thing that gets lost in the retelling: it was not a dramatic decision followed by relief. It was half a year of unglamorous choices with no external validation and no announcement to make.
But we got to break-even. And once you’re at break-even you control your own destiny — you’re no longer negotiating with people who know you have no alternative. We grew through the rest of the recession while better-funded competitors spent 2009 in conversations about bridge rounds.
Gembah, 2022
Gembah is a product development platform. I co-founded it in 2018 and joined full-time as CEO in 2020.
Covid was very good to us. Consumer spending moved online, everyone wanted new product, and we grew five times in 2020 and three times in 2021. We raised a $10 million Series A in the fall of 2021.
The market turned in the first quarter of 2022.
I want to be precise about what happened, because it’s the most transferable part of this essay. Nobody stopped wanting products. Our customers were e-commerce companies, and for two years consumer purchasing online had grown at record speed. They had ordered inventory on the assumption that it would keep growing. When that growth stalled, they were cash-strapped overnight — holding stock they had already paid for — and they killed two things at once: new purchase orders, and new product development. Not just with us; Walmart and Target were doing the same thing to their own suppliers. Product development is the most deferrable line in a consumer business when the warehouse is full.
Our unit economics went from good to bad in about six weeks. Our customer acquisition cost went through the roof. And the investors who had said grow at any cost the previous quarter now needed a path to profitability.
We took around two-thirds out of the cost base and refocused onto our core product categories. I’d like to tell you we did that in one move. We didn’t. It took three rounds. The first took out the obvious fat. The second came once we’d streamlined operations and could see which resources we could actually manage without. The third came when we narrowed our focus and stopped the projects outside our core that were consuming people and money without paying for themselves. Measure twice and cut once is the right instruction, and it assumes you can see the whole shape of the problem at the start. In the first months of a turn, you can’t.
The business eventually approached break-even, raised additional capital on the strength of that, and is growing again at a healthy clip.
But notice where the shock came from. We weren’t an e-commerce company and no consumer ever stopped buying. It started one step away from us, on our customers’ balance sheets, and it arrived looking like a demand problem — our pipeline dried up. Underneath it was a cash problem, which is what Part 2 is about.
What the three have in common
Strip out the specifics and the same three variables decide it every time.
What you owe that you can’t unwind. Not your burn rate. Burn is a set of choices you can still make differently. Your unbreakable commitments are not. Leases, multi-year vendor contracts, take-or-pay minimums, anything with a termination clause you have never read. Creditland’s ad contracts were the whole story. Boundless had almost none. This is the single most predictive number about whether a company survives a turn, and almost nobody tracks it.
It’s also why the size of a cut tells you very little. Boundless took out 20 to 25 percent. Gembah took out around two-thirds, over three rounds. Boundless needed less because it owed less, and because the cost base had never been allowed to inflate in the first place. The number isn’t the lesson. What’s underneath it is.
How fast you move after you know. Every downturn I’ve seen, the companies that died took their first real action one to two quarters after the companies that lived. The instinct to stage cuts and protect people is a good instinct that kills companies, because you spend the same goodwill three times and never get ahead of the burn. One honest cut with real severance is kinder than three hopeful ones.
Gembah is a mark against me here, and I’d rather say so than tidy it away. We cut three times. What kept that from being fatal was that the first cut came within weeks, and that each round after it was triggered by something we had learned — how operations actually ran once streamlined, which projects outside the core were quietly consuming people — rather than by hoping the market would come back and finding that it hadn’t. That’s the real distinction. The failure mode isn’t cutting more than once. It’s cutting on the assumption of a recovery.
Whether you can get to break-even without permission. Creditland couldn’t. Boundless could, and did. That distinction is the difference between an outcome you control and an outcome someone else decides.
So what would this cycle’s version look like
I don’t know whether AI valuations correct, and neither does anyone telling you they do. But the useful question is not will it. The useful question is what are this cycle’s Cars.com contracts, and who is holding them.
The candidates aren’t hard to identify. Long-dated compute commitments signed when GPU access was the scarce input and scarcity felt permanent. Data center leases and power agreements with terms measured in years. Headcount at peak-cycle compensation — technically reversible, but in practice the slowest cost any company actually cuts, and the most expensive one to reverse politically and legally. In every case the structure is identical to 1999: a multi-year obligation for the scarce input of the moment, entered into on the assumption that the scarcity persists.
There’s a real argument that this cycle differs, and it deserves stating rather than waving away. In 2000 the losses sat with equity investors in companies that had no revenue at all. Much of today’s AI spending is enterprise budget, or hyperscaler capital expenditure funded from the cash flow of genuinely profitable businesses. In 2008 the transmission ran through household leverage and the banking system, which is what made it universal. If this cycle unwinds, it more likely runs through a concentrated set of corporate balance sheets and the private credit that has grown up alongside them.
That’s a narrower blast radius. Narrower is not the same as smaller for anyone inside it, and it doesn’t mean the rest of the economy is untouched.
What to do on Monday, regardless
None of this requires a view on timing.
List every obligation you can’t exit within ninety days, with its end date and total remaining value. Not your monthly burn. Your unbreakable commitments. Most founders can’t produce this list in under a week. In a downturn, it may well be the most important document in the company.
Know your break-even headcount. Not as a plan you intend to execute — as a number you already have, so that if you ever need it you’re not discovering it under pressure.
Know how many weeks of decision-making time your cash buys you, assuming no new money arrives. Runway assuming a raise is a forecast. Runway assuming no raise is a fact.
And on raising: if you can raise on reasonable terms now, raise now. The cost of raising twelve months early is dilution you can calculate today. The cost of raising twelve months late is a number nobody can calculate, because the closing of the window is not forecastable — as I learned in the first quarter of 2000, with a signed term sheet on the table.
I should say plainly that I have an interest in this advice. My firm advises Nordic companies on raising capital in the United States, so I’m not a neutral party on whether you should raise, and you should weigh what I say accordingly.
I’d add one honest complication. Plenty of venture funds are on the sidelines right now, sitting on large SaaS portfolios they can’t exit and facing limited partners who haven’t seen distributions in some time. The bar for a check today is already high. If a downturn arrives, it doesn’t get lower.
Which doesn’t change the advice. It sharpens it. If you can clear the bar now, clear it now. If you can’t, the answer is not to wait for the market to improve. It’s to change what you’re raising against — get to a revenue number, a retention number, or a break-even date that makes the decision easy for someone. Waiting is not a strategy. It’s a bet on timing, and I’ve now been wrong about timing four times.
Part 2 looks at how a correction reaches companies with no AI exposure whatsoever — and why it arrives through your balance sheet rather than your customers.