Why Some AI Companies Go from $0 to $10M+ ARR in Months

For the last few months, I've been seeing companies that are quickly going from $0 to $5M, $10M, even $15M in annualised revenue in a matter of months.

Broadly, they seem to fall into two buckets.

Bucket 1: Vertical AI Applications

The first are selling to SMBs, where it's a small ticket size. The product itself need not be that sophisticated. It's observing the problems of a large TAM (healthcare, legal, accounting, restaurants, HVAC, solo entrepreneurs, etc.) and building vertical AI products around it.

This is where all the voice AI application companies, AI employees, and "11x for X" companies sit.

Their fundraising blurb usually sounds something like:

We're building the AI operating system for [vertical]. Since launch, we've reached $700K ARR in four months and are now raising a $10M Seed round.

Bucket 2: AI Infrastructure

The second category is very different.

These companies are building infrastructure, memory, inference, security, reinforcement learning, developer tools, and other pieces of the AI stack.

Their fundraising memo often reads something like:

We've gone from zero to $12M ARR in six months since our pre-seed and are now raising a $30M Seed round.

The first time you see numbers like that, they almost don't register.

How does a company go from zero to eight figures in revenue in a matter of months?

This post is about that second category.

I've now met and worked with a few of these companies to notice what feels like the same pattern repeating itself.

They're Not Selling to Enterprise. They're Selling to the Companies Furthest Ahead.

The obvious answer is that they're selling to enterprise.

Maybe.

Maybe not.

The interesting part is which enterprises they're selling to.

Think about companies like Anthropic, OpenAI, Lovable, Harvey, Sierra, ElevenLabs, and a hundred other AI-native companies that have raised hundreds of millions of dollars over the last couple of years.

You know the companies I'm talking about.

These companies are themselves on a treadmill.

• Every 12 months (maybe even six?) they need to justify a much higher valuation.
• Every month they need to keep growing revenue.
• Every week they need to keep shipping new capabilities.

As a result, they're constantly running into technical bottlenecks that almost nobody (or a very small subset of companies) has yet experienced.

That's what makes them such interesting customers.

The Frontier Always Sees the Problems First

Imagine you're building something in memory.

Or inference.

Or reinforcement learning.

Or voice.

Or agent security.

And somehow you end up talking to engineers at one of these companies.

During that conversation, you realise they've just run into a problem.

Maybe they made a breakthrough.

Maybe usage exploded.

Maybe a new model architecture exposed a bottleneck they hadn't anticipated.

Whatever the reason, they've suddenly encountered a problem that only they have because they're operating at a scale or speed that very few others are.

They first faced the problem last month.

Since then, they've already tried solving it internally.

It didn't work.

And somehow you've figured out a better way.

You're now getting paid $100K a month.

After a quick trial, it converts to a $500K-a-month contract.

At first glance, those numbers sound insane.

Until you think about what they're comparing you against.

Sometimes They're Buying Time

Instead of spending six months building the solution internally, they can pay you today and keep their engineering team focused on the things that actually differentiate their business.

Other Times They're Buying Efficiency

Suppose they're on track to spend $20M a year on inference or token costs.

If your product reduces that to $12M, or even $15M, you've just saved them millions of dollars.

Paying you $2M or $3M a year isn't expensive at all.

It's one of the highest-ROI decisions they can make.

The interesting thing is that they're often not comparing your price to another software vendor.

They're comparing it to:

• Engineering time
• Compute costs
• Token costs
• Delayed product launches
• Lost growth

Against that BATNA, your software can actually be the cheaper option.

The Advantage Isn't Always Technology

One thing I hadn't appreciated before is that the moat isn't always technical.

Could somebody else build the same thing?

Probably.

But they won't even know it's a problem until they end up talking to the same customer you did.

These aren't problems you'll find on Twitter.

They aren't in Gartner reports.

They don't exist in the broader market yet.

They're problems that only a handful of companies have because they're the ones pushing the frontier.

The information advantage comes before the technology advantage.

As an investor, what you're really underwriting is:

Will many more companies have this problem six months from now?

These Companies Aren't Really Product-Led

The other thing I hadn't appreciated is that these companies almost don't operate like traditional SaaS businesses.

In a normal startup:

1. Identify a market.
2. Build a product.
3. Spend years finding customers.

Here it almost feels inverted.

You might have some basic version of a product.

But that's typically little more than an interesting feature.

You first find one frontier customer.

That customer exposes a problem that nobody else knows exists.

You build the solution (or change your product).

In the process, you discover the next problem they need solved.

And then the next one.

The roadmap isn't coming from some grand product strategy.

It's coming from staying incredibly close to the handful of companies that are already living six months ahead of everyone else.

On the product itself, I've seen multiple approaches.

Sometimes there's no product at all.

It's pure access.

That access leads to insights.

The insights lead to the product.

Other times, the team had spent years building something else, used that product to get in, but then stumbled onto the new direction—discarding the old business entirely.

Customer Concentration Is Almost a Feature

There's a very peculiar situation these companies often find themselves in.

Your revenue puts you in the top 0.1% of high-growth startups at your stage.

But 80–90% of that revenue comes from one, two or maybe three customers.

Traditional SaaS thinking says customer concentration is dangerous.

Eventually, it is.

But early on, I almost think it's a feature rather than a bug.

The companies paying you these amounts are the ones seeing these problems first.

Everyone else will get there eventually.

They're simply not there yet.

Which means your first few million dollars of ARR can come from just a handful of customers.

You're spending:

• Less time on broad GTM.
• Less time building an organisation.
• More time building alongside the customers that are dragging you towards the next problem they need solved.

Marquee Customers Create the Market

The other interesting thing is what happens once you land one of these customers.

They're usually companies everyone has heard of.

That logo immediately gets investors interested because the traction suddenly feels much more believable.

You raise a much larger round.

You publish a case study.

Competitors of that customer begin running into exactly the same problem and start reaching out.

Engineers move between companies and carry knowledge with them.

Word spreads.

What started as one customer slowly becomes an entire category.

It almost feels like category creation in reverse.

Instead of building a product and convincing the market it has a problem, you start with the handful of companies already experiencing the future, solve their problems first, and then watch the rest of the market slowly catch up.

So What Advice Would I Give?

If you're building AI infrastructure, developer tools or data-market businesses (there are probably other categories too), I wouldn't think about "enterprise" as one giant bucket.

I'd think about getting into the rooms where the companies furthest ahead are spending their time.

Because once you're in those rooms, you stop guessing what to build.

You start hearing about tomorrow's problems months before everyone else does.

Those problems are usually:

• Urgent
• Incredibly expensive
• Impossible to Google

Solve them well, and the next set of problems usually comes from the same customer.

The fastest-growing AI infrastructure companies I've met aren't necessarily the ones that predicted the future better than everyone else.

They're the ones that positioned themselves close enough to the frontier that they got to watch the future arrive first.

And yes, once the revenue starts flowing in...

Please convert them to annual contracts. 🙂