> ## Content Index > Fetch the complete content index at: https://www.prodbistro.com/llms.txt > Use this file to discover other available public pages before exploring further. # The Off-Balance-Sheet Roadmap - URL: https://www.prodbistro.com/the-off-balance-sheet-roadmap/ - Published: 2026-08-24T14:00:47.000Z - Updated: 2026-08-24T14:00:46.000Z - Description: Every roadmap has an official version and an unofficial one. The unofficial one carries real risk and doesn't show up anywhere leadership looks, until it does. - Author: Geoff Anderson - Tags: Product You've priced a new AI feature into next year's roadmap based on your inference vendor's published per-token rate. That rate assumes the vendor's own underlying costs stay roughly where they are today. Do you actually know what's underneath that assumption, or did you take the rate card as a fact of nature and build twelve months of commitments on top of it? Most of us take it as a fact of nature. Worth changing that habit, and here's a real-world mechanism from outside product management that explains why. ## What's actually happening in AI infrastructure financing A lot of the compute capacity powering the current AI boom isn't sitting on the balance sheet of the company selling it to you. It's financed through separate legal entities, set up specifically so the debt backing the data center doesn't show up as debt on the parent company's own books. The parent keeps a small stake, a private lender puts up the real money, and a long-term contract, "we promise to buy this much capacity for this many years," is what convinces the lender it'll get paid back. That structure isn't new or shady on its own. It's the same tool that's financed toll roads and power plants for decades, and it's a completely reasonable way to raise money against a long-lived, cash-generating asset. What's newer is the scale, and how much of the "revenue" backing these deals is one company's promise to buy from a second company that's also, separately, promising to buy from a third. Estimates for total AI infrastructure commitments sitting off the major providers' balance sheets are now in the two-to-three-trillion-dollar range, and industry analysts have publicly compared the structure to the pre-2002 accounting rules that let a company called Enron hide its own debt the same way, before regulators tightened the consolidation rules specifically to stop it. I'm not telling you this to make you nervous about a specific vendor. I'm telling you because the underlying pattern, obligations engineered to sit somewhere they don't get scrutinized, is one you've almost certainly seen inside your own company, and recognizing it in the vendor's books should make you better at spotting it in your own org chart. Product Bistro doesn't run on ad spend or an algorithm surfacing it in anyone's feed. It runs on PMs passing it to other PMs. If this issue was useful, forward it to someone on your team, or subscribe so the next one lands directly in your inbox. [ Subscribe ](#/portal/signup) Share this post ## The same pattern shows up inside your company Here's the translation. Every roadmap has an official version, the one in the planning doc leadership reviews, with its budgeted headcount and its committed dependencies clearly listed. And every roadmap of any real size also has an unofficial version: the pilot nobody formally staffed, the integration one engineer is quietly maintaining as a favor, the dependency on another team that was agreed to in a hallway conversation and never made it into anyone's OKRs. That second roadmap carries real risk. It just doesn't show up anywhere leadership looks, the same way an SPV's debt doesn't show up on the parent's balance sheet, for exactly the same underlying reason: keeping the official numbers clean is easier than accounting for the whole picture, right up until the unofficial commitment comes due at the worst possible time, usually during a launch or a headcount review, when there's no slack left to absorb it. The fix isn't complicated, but it does take actual discipline: treat undocumented dependencies and unstaffed commitments as real liabilities, not as free capacity you get to ignore because nobody put them in a spreadsheet. If a launch depends on a favor, put the favor in the plan. If a vendor relationship depends on pricing staying stable, write down what you're actually betting on, not just the number on today's rate card. ## What to actually ask your vendors Three questions worth adding to any vendor conversation where you're building multi-year roadmap commitments on top of their pricing or capacity: - What happens to my pricing if your underlying costs change for reasons that have nothing to do with our contract? - Is the capacity you're selling me backed by your own balance sheet, or by a financing structure that depends on other customers' commitments staying intact? - If your biggest customer's demand assumptions turned out to be wrong, what happens to the terms I'm relying on? Most vendors won't have a fully satisfying answer to all three. That's fine, and it's not a reason to walk away from the deal. It is a reason not to build your roadmap as if today's number is a permanent fact, rather than a bet that's only as solid as the assumptions underneath it, some of which you can't see from where you're sitting.