Founders almost never publish the dollar number. The number is uncomfortable, the breakdown is more uncomfortable, and the opportunity cost is the most uncomfortable line of all. So this post does the uncomfortable thing and writes them down. Three startups, three shutdowns, four years, one founder funding the whole thing from his own bank account.

The full framework that came out of those shutdowns lives in three startups, three shutdowns and the three checks I missed. This post is the financial appendix to those: the line items, the totals, the things I would not pay for again, and the discount that bad decisions you cannot afford give you on the next bet.

The cash line

The direct cash number across all three startups, kept in a single spreadsheet I still update, is roughly forty-six thousand US dollars from late 2021 through the Vibe AI shutdown in early 2026. That includes infra, software, contractor hours, India incorporation and compliance, domain and brand, hardware refreshes, and the small ongoing costs of keeping companies legally alive while you decide whether to bury them.

Forty-six thousand is a small number for a Silicon Valley startup and a serious number for a bootstrapped founder. It is roughly two and a half years of senior engineer take-home pay in Bangalore, paid in full from personal savings with no investor diluting it and no salary covering it. There was no parachute. The decision to spend any of it had to clear a personal bar that VC dollars never clear.

Roughly half of the spend was on cloud infra, model APIs, and paid SaaS. Roughly a quarter went to contractors and freelancers, mostly during a single bad stretch on Super AI that I detail below. The remaining quarter was the slow drip of compliance, hosting, domains, accounting, and the assorted line items that nobody warns you about when you incorporate a Pvt Ltd in India.

The opportunity-cost line

The cash line is the small one. The honest cost is the opportunity cost: what I would have earned over the same four years at a senior software engineering role at a well-funded company. In Bangalore, that range is twenty-five to forty lakhs annually for a strong senior IC, which is roughly thirty to fifty thousand US dollars per year at current rates. Over four years, that compounds to between one hundred twenty and two hundred thousand US dollars of foregone income.

Add the cash burn and the all-in cost of three shutdowns is somewhere between one hundred sixty thousand and two hundred fifty thousand US dollars. That is the honest number. It is also the number that almost nobody publishes, because it sits in the uncomfortable place between "founder hardship" and "founder privilege". I had savings, family support, and Bangalore cost of living. Without those, the math does not work at all.

Cash vs opportunity cost across 3 shutdowns (USD) MindWave $3k cash + $36k forgone salary Super AI $22k cash + $48k forgone salary Vibe AI $21k cash + $60k forgone salary Direct cash burn Opportunity cost (foregone senior IC salary)
Direct cash burn vs opportunity cost. Bangalore senior-IC compensation used as the foregone-salary baseline.

Where the money actually went

The breakdown of the forty-six thousand dollars in direct cash burn looks roughly like this. Cloud infrastructure across AWS, Cloudflare, and self-hosted boxes accounted for around twelve thousand. Foundation model APIs, mostly OpenAI and Anthropic during Vibe AI, accounted for around six thousand. Paid SaaS subscriptions across the three companies, including design tools, analytics, email, productivity, observability, and the long tail of ten-dollar-a-month line items, totaled close to nine thousand.

Contractor and freelance work, almost all of it on Super AI, came in at roughly eleven thousand. India compliance, accounting, incorporation, GST, and the assorted Pvt Ltd costs added another five thousand. The rest, around three thousand, was hardware refreshes, domain renewals, brand assets, payment gateway fees, and the long tail of one-off costs.

SaaS sprawl is the line I am most ashamed of. By the third startup I had a stack of seventeen paid tools, half of which I used less than once a month. Cutting it down to five core tools by the end of Vibe AI saved roughly one hundred and twenty dollars a month, which compounds across a multi-year build to a number that would fund a real test of a real hypothesis. The economics post at economics of bootstrapped AI agents walks through the same SaaS-discipline math for the Gravity build.

The single worst dollar

The single worst chunk of capital I deployed across three startups was twelve thousand dollars in contractor design and frontend work on Super AI in early 2024, before product-market fit. I had a vague product hypothesis, no validated user, and a contractor pitch deck telling me the screens needed to look "production-ready" before I could test anything. The deck was wrong. The screens looked good. Nobody looked at them.

The lesson is brutal and simple: pre-PMF, every dollar that does not directly test a hypothesis is wasted. The Super AI postmortem at super-ai-postmortem walks through the decision chain in detail. The short version is that I bought polish where I should have bought signal. A founder pre-PMF should build the testable thing themselves, ugly, until the test passes. Polish is a post-PMF cost.

The discount on bad decisions

Here is the part that almost nobody writes. Funding your own failures gives you a discount on every future decision that VC capital cannot give you. When the cash is yours, the unit-economics check is reflexive. You kill MindWave at seven thousand dollars of burn, not seven hundred thousand. You shut down Vibe AI before raising a Series A, not after. You walk into Gravity with a pre-budgeted monthly cap because the alternative is to repeat the Super AI freelance mistake at a larger scale.

CB Insights' analysis of post-mortems consistently puts "ran out of cash" at the top of the failure list, with around thirty-eight percent of failed startups citing it as a primary cause. The Wilbur Labs founder survey on shutdowns echoes the same conclusion. The bootstrap discount is that running out of cash is a much smaller event when the cash is yours and the burn is small. You shut down faster, learn cheaper, and bring the next bet to the table without crippling debt.

What Gravity does differently

Gravity has a fixed monthly cap I will not exceed before paying customers exist. The cap is published internally as a single line in the company doc; everything else has to fit underneath it. The 80-test methodology in how we test AI agents exists in part because reliability is the cheapest form of marketing for a bootstrapped founder. The unit economics post explains the per-task math.

The core principle is that the discount only works if you keep using it. Every line item in the Gravity stack has to clear a personal bar: it tests something, it directly serves a paying customer, or it is required for legal operation. Anything that does not clear the bar gets cut, and the cut is reviewed monthly. That is the financial discipline I am bringing into startup four, paid for in full by the four years and the forty-six thousand dollars and the salary I did not earn.

If you are building bootstrapped and want to compare line items, my email is at the top of /contact. The number I publish here is the floor; I will not pretend my mistakes were cheaper than they were.

Frequently asked questions

How much money did three failed startups cost?

Across MindWave, Super AI, and Vibe AI, my honest direct cash burn was roughly forty to fifty thousand US dollars over four years. Opportunity cost is bigger. The salary I would have earned at a senior engineering role over the same window pushes the all-in number past two hundred thousand. Cash is the small line item; time is the big one.

Where does most of that money actually go?

Cloud infra, model API calls, paid software subscriptions, and India company compliance costs eat most of it. SaaS sprawl is the silent killer. By the third startup I had cut subscriptions by seventy percent without losing meaningful productivity, which is the single biggest dollar lesson I carry into Gravity.

Is it worth funding your own failures?

Funding my own failures forced unit-economics discipline that VC capital would have hidden for years. Bootstrapping made me kill MindWave at seven thousand dollars of burn, not seven hundred thousand. The cost is real, but the discount on bad decisions you cannot afford to repeat is also real. I would still pick this path.

What was the single most expensive mistake?

Hiring contractors before product-market fit on Super AI. I burned roughly twelve thousand dollars on freelance design and frontend work for screens that nobody used. The lesson is brutal: pre-PMF, every dollar that does not directly test a hypothesis is wasted. Founders should build the testable thing themselves until the test passes.

Did the cost change how you build Gravity?

Yes. Gravity runs on a fixed monthly cap I will not exceed before paying customers exist. Every line item is a pre-budgeted decision, not an open-tab habit. The bootstrap discipline is encoded into the build process itself, which is why the economics post and the three-checks framework both exist as living documents.

Three takeaways before you close this tab

Sources