# How I Made $20M with AI in One Year

[Watch on tuber](https://tuber.guzus.xyz/video/tjfaAsbaayc/summary) · [YouTube](https://youtu.be/tjfaAsbaayc)

**Score: 42/100 · skip**

Ben, a solo founder, says he took his AI "company OS" **Pulsia** from an idea in Paris (November) to a **$20M annual run rate** in San Francisco in under a year, and raised money at a **$250M valuation**. The video tells the story month by month: a publicity stunt brought in investors, then most of the work went into **fixing runaway AI costs, stabilizing an unreliable product, and cutting the price to grow**. Treat the numbers with some caution. The intro says he raised $30M, but at 4:40 he says $20M, and none of the figures are backed by any evidence in the video. The auto-transcript also mangles the product name ("Post AI," "Posia," "Polisur") and probably the infrastructure partner's name ("Sapien," "Sapient," "CPM," "CPO").

- **0:03–1:22** — **Cold open and pitch.** Ben describes Pulsia as "an AI that builds your company while you sleep" and states his headline numbers: a $20M run rate, $30M raised at a $250M valuation, all as a solo founder. A short teaser shows him saying he spends about $1M a month on AI and brushing off critics who point out bugs. That sets up the video's two tensions: very fast growth, and expensive, imperfect technology.

- **1:22–2:06** — **Origin and V1 build (November–December).** After a year of travel, he came back from Japan fixated on an AI that could run a whole business by itself. He spent a month alone in his Paris apartment with as few distractions as possible and a strict routine, building V1. By December the to-do list still had no end, so he **picked a launch date** rather than polish forever. That forcing function is one of the more useful lessons in the video.

- **2:17–2:42** — **First 100 paying customers.** After launch he did manual hustle: he emailed every user of every product he'd built that year, became a "reply guy" on Twitter and LinkedIn in AI-agent threads, and ran simple Meta ads. Nothing sophisticated, but together these got him his first 100 paying customers. The point is that he leaned on his existing audience and plain distribution, not on any growth trick.

- **2:42–3:19** — **Moving to San Francisco.** About a month after launch, with traction showing, he decided to move to SF to market the product harder and get closer to the ecosystem. He frames it as a bet on momentum: the product was working, and the next job was getting more users onto it.

- **3:19–4:52** — **The "AI raises its own round" stunt.** In SF he decided he needed to "make noise." He gave Pulsia access to his inbox so it could answer every email, spot investors, research them, and reply with product details and raise terms. The AI itself suggested aiming for a **$1B valuation**. He calls the stunt deliberately provocative. It went viral, investors started contacting him, and he used the momentum to take meetings. He says he raised **$20M at a $250M valuation**, which contradicts the $30M in the intro and in the later launch video.

- **4:52–7:50** — **Hardening the product and the $1.2M Anthropic bill (around April).** With the money in the bank, the product still wasn't ready to scale. Engineers from Sapien, a startup building infrastructure for AI agents, joined to fix bugs and harden the platform. The bigger problem was cost: about **$1.2M in one month, mostly paid to Anthropic**. That is unsustainable unless he charges customers $200–500 a month. His fix was to **rent GPUs and run open-source models**, and he visits the data center. A dashboard shows the migration away from Anthropic-only. He openly admits the tradeoff: open models are "not as good" as best-in-class ones but "can still be very good," and he hopes to bring the bill down to around **$50K**. Why it matters: this was an existential margin problem, and switching to cheaper models risks making the product worse.

- **7:55–9:56** — **Orchestrated launch video (May).** Revenue was nearing an **$8M run rate**, and the funding still hadn't been announced. Because the launch doubled as a funding announcement, investors wanted input on timing, so he took a few extra weeks. In the video, the AI narrates how it raised "$30M by itself." He coordinated friends to amplify it and cold-pitched press. He says it got millions of views on day one. Critics called it "AI slop," pointed out bugs and the cold-outreach tactics. His answer was that the goal was visibility, so the launch succeeded either way.

- **9:56–10:29** — **Offsite in Cabo.** He spent a week at an offsite organized by Sapien. He admits the product was "not yet super stable" and the timing was bad, but says he needed rest. It's a short look at what being a solo founder costs him personally.

- **10:29–12:13** — **Price cut from $50 to $20 (June).** At an **$11M run rate**, the GPU savings gave him a choice: keep the margin or pass it on to customers. He cut the price to $20 a month, and run rate dropped **from $11M to $8M within a month**. In exchange, acquisition became much easier. He describes a free-tier-style model that only spends compute when users reply to Pulsia's daily email, so people who try it and leave cost little. His reasons are reaching students who can't pay $50, getting more usage data, and improving "pivoting": when a user loses interest, Pulsia suggests a new project. The caveat is that this traded revenue now for growth later.

- **12:00–12:58** — **Rebuilding for scale (July).** The flood of new users broke the product, which "wasn't built for that many people." He rebuilt large parts of Pulsia with the partner's engineers to cut bugs and costs. He notes that an AI able to "build anything" means "any problem can happen." After the rebuild, revenue stopped falling and started climbing again.

- **12:58–14:42** — **Break in Europe (August).** After six months of nonstop sprinting, he went to Paris, Normandy, and Greece, still working but at a slower pace. On a windy Greek island he uses the nonstop wind as a metaphor for startup chaos. His point is that you can focus on the anxiety or on the "beautiful moments" around you. This section is reflective, with no business information.

- **14:42–16:34** — **$20M run rate and the next version (September–October).** He says margins were fixed, the growth engine was working, and the product was finally stable. He went from zero to about $10M run rate in roughly 5 months, with most of that ($1M→$10M) coming between March and June. He reached a **$20M run rate** by the end of September. He then went back to Tokyo, where the idea started after he left his previous company ("Gift Shop"), for a one-week creative trip to plan the next version. He calls it V2 at one point and "V3 / Pulsia AGI" at another, and describes it as a complete rethink of how the product works. The video ends on that note; the last minute is mostly scene-setting and outro.

**Practical takeaways**
- **Set a hard launch date.** Ben shipped V1 only after choosing a date, because the feature list "had no end."
- **Early distribution can be plain and manual.** Emailing past users, replying in relevant threads, and small Meta ads were enough for the first 100 paying customers.
- **For AI products, check inference costs before you scale.** A $1.2M monthly bill forced a move to open-source models on rented GPUs, which saved money but gave up some quality.
- **A price cut moves revenue around.** Dropping from $50 to $20 cut run rate by about 27% in a month. It only paid off because spending compute only on engaged users kept costs under control, and a rebuild followed to handle the new load.
- **Provocative stunts get attention and criticism together.** The AI-fundraising stunt and launch video drew investors and millions of views, along with "AI slop" backlash. Decide in advance whether you can accept that trade.

## Audience signal

- No high-signal audience themes available.